Procurement as a function is, at its core, a human discipline. Yes, it carries an analytical skeleton: data-driven vendor selection, historical pricing benchmarks, volume-based leverage, total cost of ownership models. These are the instruments. But the music is relational. The durable advantages in any supply chain come from trust built over years, from the supplier who calls you first when allocation tightens, from the partner who stretches terms during a cash crunch because they believe in the relationship’s long horizon.

I was presenting at a conference last week, and the panelists across industries kept returning to the same refrain. Their organizations had doubled down on proximity to suppliers and customers, particularly after the ruptures of Covid and now amid the cascading pressure of tariffs and geopolitical fragmentation. The lesson from the last five years of supply chain disruption is not that you need more systems. It is that you need better relationships. Depth, not diversification, is the hedge that actually holds when the world fractures.

And yet here is the paradox that nobody on those panels wanted to fully name: the time available to build those relationships is shrinking. AI has improved individual productivity, but it has also raised the floor on what is expected. Work expands to fill the capacity that technology creates. The procurement professional who used to run eight sourcing events per quarter is now expected to run twenty. The categories that went unmanaged because there was simply no bandwidth for them are now visible on a dashboard, waiting. The relationship-building that everyone agrees is strategically essential keeps getting deferred for the analytical work that is merely urgent.

This is exactly the tension that makes my conversation with Ben Leiken, Chief Technology Officer at Arkestro, so worth unpacking.

Chapters

Chapter 1 - Introductions (00:06 to 03:11)

Chapter 2 - From Bid-Ops to Arkestro: The Origin Story (03:12 to 07:28)

Chapter 3 - How a Procurement Bid Actually Works (07:30 to 12:16)

Chapter 4 - The Suggested Price: How It’s Built and What It Does (12:17 to 16:17)

Chapter 5 - The Model Stack: In-House vs. Foundational AI (16:18 to 18:54)

Chapter 6 - Training Without Sensitive Data (18:55 to 22:01)

Chapter 7 - Negotiation Science: What It Means to Predict a Negotiation (22:02 to 26:07)

Chapter 8 - The Car Buying Analogy: Anchoring, Suggested Price, and Supplier Coaching (26:08 to 30:09)

Chapter 9 - Game Theory Applied: Designing the Negotiation, Not Just Running It (30:10 to 33:38)

Chapter 10 - The Value Case: Time, Price, and Continuous Procurement (33:39 to 37:48)

Chapter 11 - Does Automation Erode Supplier Relationships? (37:49 to 44:09)

Chapter 12 - Customer Journeys: Chevron and Nissan (44:09 to 48:02)

Chapter 13 - The Numbers: Savings Benchmarks (48:02 to 49:22)

Chapter 14 - Business Model and Funding (49:22 to 53:07)

Chapter 15 - Building Technology at Startup Scale: Roadmap Discipline (53:07 to 55:30)

Chapter 16 - Closing (55:30 to 56:02)

Chapter 1 - Introductions

Nikhil Varshney (00:06) Hello everyone, welcome to another episode of Silk Road Nexus Conversations. I’m here today with Ben, he is the CTO at Arkestro. Arkestro is a native procurement AI solution and Ben will talk a little bit more about that in details. But the story is I was supposed to meet with Ben at Manifest which happened in Vegas in 2026. But unfortunately due to clashing schedules, we were not able to do that. And I’m thankful Ben is here. He gave me additional time to kind of record this and work with him. So thank you, Ben. If you could quickly introduce about yourself and Arkestro.

Ben Leiken (00:42) Absolutely. Yeah, no, it’s a pleasure to be here. Manifest was a blur. And I know we both had pretty busy schedules. So sorry we couldn’t connect there, but making up for lost time now. But yeah, just a little bit about me. So I’m Ben. I’m the chief technology officer over at Arkestro. We are the only predictive procurement platform working with enterprise procurement teams to improve their operations and ultimately drive them towards better outcomes unlocked by AI and behavioral science. So that’s a bit on what we do. But my background, I’m a technologist by trade and education degree in computer science. I think the through line on my career has always been a passion for bringing technology to the masses and particularly to underserved markets. As a high schooler, I had a business teaching people basic technology skills, how to use computers, how to get on the internet. Prior to Arkestro, I was at a company called Survey Monkey, which is kind of famously known for the direct to consumer, freemium business model, kind of bringing market research and research technology to the masses. That was a great kind of formative experience for me there too. But at Arkestro, which we began as a company called Bid-Ops, I like to joke that Arkestro and Bid-Ops were my first procurement jobs. And I was introduced to the space by our founder Edmund Zygren, who is a procurement practitioner and spent a number of years in the industry and really opened my eyes to this underserved area and business function that every business has that for a variety of reasons had been kind of left out in the cold. Back then, this was 2017, 2018, left out in the cold from the machine learning and big data boom. And I think seeing just the opportunity to really transform an industry and a business function got me really excited and brought me here and that’s if anything I’m much more excited now even than I was then because I think the opportunity has proven to be even more significant than I thought it was.

Chapter 2 - From Bid-Ops to Arkestro: The Origin Story

Nikhil Varshney (03:12) No, thank you. That was a great introduction. Before we jump in, I think one thing that you mentioned, Arkestro was Bid-Ops before it became Arkestro. Can you tell us a little bit about Bid-Ops and procurement? You also mentioned that there were some underserved needs in that procurement market. And then how did the pivot happen to Arkestro?

Ben Leiken (03:33) Yeah, great question. So the seed idea for Bid-Ops was born out of Edmund’s experience working in competitive bidding processes and running reverse auctions as a practitioner for many years and just seeing that there was a huge opportunity there to improve the outcomes of those processes and the efficiency. We really started Bid-Ops as a bid optimization software. That was our original focus. What became the foundation of predictive procurement is that in procurement, being able to predict the outcome of a process, which really talented, seasoned procurement veterans can kind of intuitively do. Folks who run a lot of reverse auctions, who run a lot of quoting processes, kind of have a sense before they even go in and run the event of how the event is going to end up. Things like supplier preference, knowing who the incumbents are, how many folks are participating. You get this kind of intrinsic knowledge, this horse sense, if you will, of how the event is going to run. Edmund saw this and asked, how do we build a product that enables procurement teams who want to run competitive bids to kind of ingrain all of that knowledge into the process and actually predict the outcome and then work backwards from that desired outcome to drive a shorter process and a better process. And so that was really Bid-Ops. Once we discovered that lesson in competitive bidding, the story from there is that we unearthed the fact that this way of thinking, which again became predictive procurement, applies to procurement writ large. The transition from Bid-Ops to Arkestro is saying, if we take this initial innovation and this model set that we built dramatically expanded, there’s a wider variety, a huge variety of use cases across the landscape of procurement that this way of looking at the world is relevant for. Again, this predicting the outcome and working backwards from that outcome to drive a better process. As the scope expanded and we had the chance to work with more customers and deliver a lot of wins to those customers, we started to see a kind of growing groundswell of momentum in the market. It made sense to rebrand the company to Arkestro and around that same time, we were fortunate enough to get connected with Rob DeSantis, who is the other co-founder of Arkestro, who is a very well-known figure in both the technology space, particularly procurement technology. He was one of the original founders of Ariba that became SAP Ariba. And Rob got a chance to see what we were doing at the time, got really excited to kind of jump back into the procurement tech game after a few years off, working for LinkedIn and a number of well-known companies as a board member and investor. And Rob coming on board and bringing his immense market knowledge and kind of pairing that with our technological vision, I think really threw a lot of fuel on the fire and has carried us forward to where we are today.

Chapter 3 - How a Procurement Bid Actually Works

Nikhil Varshney (07:29) That is a fascinating story and truly inspiring as well with some well-known figures on the board as well as on the founding team. One thing, Ben, I would like to understand if you provide this to me with an example of a bidding process in procurement. And that is probably the reason I’m asking this is how I see procurement is, for example, Apple is making an iPhone. They have around 300 parts that they procure for making that iPhone. There could be multiple suppliers that could support the procurement of those parts. So you are bidding with each of these suppliers on the price or are you bidding on different parts altogether? So can you give me like one example of how that breaks down in terms of bidding optimization?

Ben Leiken (08:20) Yeah, it’s a great question. So what we offer our customers is a full horizontal procurement solution. The most common use case is a competitive bid. Think an RFQ, a sourcing event, where in an NPI use case, a new product introduction use case where I’m building something new, I need to go out and secure contracts with a set of suppliers and I want to ensure that the prices that I’m paying are competitive, but actually more important than that, that I’m getting the right quality and business requirements associated with every part that I’m buying. The role that Arkestro plays in that example is kind of managing that process where a platform we sit in the middle, the procurement team initiates the demand request. We help them through a mix of our predictive models and platform capabilities kind of massage that demand request into an optimal event. We facilitate the supplier side interaction, recruiting the suppliers into the event, inviting them. The suppliers participate either live in the platform or over email, using a magic link flow where they jump into the platform and can drag and drop in a spreadsheet or use the UI to submit their quote. And that’s the overall flow. The supplier submits their quotes, they get feedback on those quotes and we support the procurement team in arriving at an eventual award decision. Where we are particularly unique is that at the outset of the demand request being received for procurement, we actually are able to go in using our technology and predict a suggested price for every single item that the procurement team might be trying to buy. So using the Apple example, these could be component parts. Using a services example, it could be a set of rate cards and suggested rates. What’s unique about Arkestro is that the approach that we take is identity agnostic. So we actually don’t need to know what you’re buying and you don’t need to have bought what you’re buying before to use our technology and to get a credible suggested price. So that alone is really cool, but we take it a step further in taking that AI generated suggested price and putting it in front of the supplier. So the supplier, when they log into the platform, as opposed to the way that procurement traditionally works, which is, I need a quote, let me go out and ask three suppliers what I should pay for this thing that I’m trying to buy. And I have the suppliers tell me the price. And then I say, this price is lower, this price is higher, this one meets my quality standards. That sort of process takes a lot of time. With Arkestro, by starting again short-circuiting that process, starting at the desired outcome and putting that desired outcome in front of the supplier, we’re able to drive a better result. Suppliers like it because it saves them a lot of time. They have a better sense for where they stand and what the buyer’s expectations are. And they can of course decide if they want to tweak the suggestion. But that use of the suggested price is an example of a smart default in a process that has a dramatic impact on the amount of time the process takes and the overall end result.

Chapter 4 - The Suggested Price: How It’s Built and What It Does

Nikhil Varshney (12:16) Makes sense and I think one part that I would like to double click on is the suggested price. If for example, there are let’s say 10 suppliers for one particular RFQ that has been released by the buyer, the suggested price is for the buyer as well as for the suppliers, right? It’s for both. That if you are bidding for something and someone says $100, the other supplier says $98, someone says $110, you are basically telling them that our suggested price is let’s say $101. So some people might be underbidding, some people might be overbidding on that contract. For the buyer, they have a genuine idea of like, okay, $101, but I have a quote of $98. So probably I can look at that buyer as well, which is below the suggested price. So then how does the equation actually fit over here? Because if the suggested price is for both the buyer and the seller, is the seller not going to adjust the price? And then basically you’re reaching the average rather than kind of having that competitive process of someone undercutting each other.

Ben Leiken (13:24) Yeah, so it’s a great question. I think that gets at how we arrive at the price suggestion to begin with. Before the price is shown to any supplier, we service it to the buyer first. And the input to the suggested price, one of the inputs, is a baseline. This could be a last price paid, this could be a value pulled from an existing catalog. We’ve also built functionality, we call this the intelligent counter offer, that can actually go out into market on the buyer’s behalf and assess a competitive baseline price, essentially what this thing should cost. And in some cases, if the buyer has a should cost model, it is used to set the baseline. From there, the machine learning model that we’ve built, the suggested price model, predicts a suggested discount off of that baseline. And that’s really key. First, the buyer has the opportunity to tweak the suggested price, but the price suggestion ultimately is arrived at by looking at the competitive context of the event. So the model, in addition to looking at the baseline, is looking at the suppliers that are participating or planning to participate, engaging in effective anchor based off of that competitive context. Some of this relates to the way that the models actually work, but some of this is also process science. The order and the way in which this feedback is shown to the supplier, in our case it’s shown at the outset of the event. We have a pretty high confidence that that price suggestion sits somewhere in what we call the zone of possible agreement, which is kind of the overlap of the supplier’s possible price range and the buyer’s desired price range. In some cases, the suggested price might be higher than one supplier is going to quote, lower than another, but we do a pretty effective job at creating it as a starting point. But it is just that, it’s just a starting point. And from there, the game theoretic negotiation engine kicks in. And this gets into the game theory, which we’ll talk a little bit more about later on, I’m sure. We can dig into that. But that’s the other really key piece of the platform, is the way that the negotiation experience actually interacts with suppliers and provides them line item level game theoretic feedback to coach them to a desired outcome. Sometimes that outcome changes based off of how suppliers react to that initial offer.

Chapter 5 - The Model Stack: In-House vs. Foundational AI

Nikhil Varshney (16:17) Makes sense. And I know you mentioned a little about the machine learning models that you have. Are these models in-house and you’re training them on your own data or is that like what is happening right now in the market is that models are primarily owned by hyperscalers and then you build applications on top of that. So what kind of model are you working with? Do you own these models or are you building agents on top of these models that can help you do multiple stuff?

Ben Leiken (16:45) Yeah, it’s a great question. And it’s of course one that we get a lot. I think there is the kind of stereotypical, where’s the AI question. And AI is kind of implicitly framed as this one model, it’s one entity. That’s not our reality. And that’s also just not how we, the version of thinking about AI that we found is right for our market space. We, to answer your question directly, we use a variety of models. We have some trained in-house, some more classical analytical machine learning models, and as well as we’re using the same set of generative AI foundational models that others are, hosted off of Bedrock or OpenAI or Anthropic. So we do it all, but what we’re really focused on is how do we apply the right form of artificial intelligence to the right business process and particularly at the right point in the procurement team’s workflow for it to drive maximum value. And I think that’s really key because there’s a number of examples in procurement and outside of it where these generative models, these Claude or GPT wrapper type products, folks are kind of applying AI as a cure-all on often fairly broken processes and procurement is full of those. Looking back to the early 2000s, most of the incumbent players in procurement, your Aribas, your Oracles, were really focused on digitizing physical processes. And the reality is many of those processes have not changed for at least 20 years and in many cases longer. This is where, with predictive procurement, we’re focused both on applying AI, but fundamentally creating and driving towards a transformed set of processes that drive better outcomes. You apply AI to a broken process, you’re just making that broken process run faster.

Chapter 6 - Training Without Sensitive Data

Nikhil Varshney (18:54) Exactly. Now I think that’s very interesting because I think one part that I would again like to double click on is that you said that you have your own models that you’re training and what has been your experience around the quality of the output. The reason I ask that is, are you getting better outputs during pre-training or during inference? How are you matching pre-training and inference within Arkestro? The reason I ask that is the negotiated data that you’re collecting or the suggested price that you’re building is probably going to be highly sensitive to the customers you work with. And that kind of goes back into your pre-training models. But when you use your inference models, you probably can get better results in real time. So that’s the reason I’m asking this question.

Ben Leiken (19:48) I think the answer is it varies based off of the use case. And I think we’re focused a lot on negotiation science, what we call negotiation science, the suggested price, where the approach that we’ve taken to create the suggested pricing recommendation, we don’t see super wide variance across customers because of the way that we’ve designed that model. It’s a behavioral model. We’re really focused on training that model in such a way that it’s looking at non-identifiable attributes of the event and of the suppliers participating to make their recommendations and it’s predicting a discount. And that’s one of our core aha moments, is we’re saying the set of training features that you can train a model on to drive consistent discounting behavior across customers and across a variety of use cases is a different set than most people think. Many other price prediction models are centered around an item identity resolution sort of approach. We get around that by focusing more on supplier behavior and competitive context. And looking at the identity information, a lot of that is encapsulated in the baseline process. That works really well for us from an efficacy standpoint, but it also allows us to say we haven’t designed a system where we need to train a machine learning system on sensitive customer data. Enterprise customers do not want their pricing data and their supplier identity data used to train a large language model or really any sort of predictive model. And so the way that we’ve architected this thing is done in such a way that we can drive really powerful results without needing to leverage that confidential data, which is really important to our enterprise customers.

Chapter 7 - Negotiation Science: What It Means to Predict a Negotiation

Nikhil Varshney (22:01) And I think that is so true because I think the data is so sensitive that if you use it to train your models and you provide a suggested price based on that to their competitor, it becomes kind of challenging around losing the pricing advantage that they might have just because of the relationships with the vendors, et cetera. So I think it makes sense. I think pivoting from here to learn a little bit more about the negotiation sciences, that is the one part that has fascinated me a lot about Arkestro as I started reading about it. Before I kind of ask my question, I’ll just lay the premise of my understanding of negotiation. Negotiations are highly behavioral, right? I mean, it’s based on relationships. It’s based on your understanding of what the market is. It’s also highly analytical in the sense that you have previous data that can lead you towards better price negotiations. Having said all of this, which is a behavioral context, the market context and the analytic context, how does predictive negotiation or predictive procurement actually work? What is predictive negotiation? Like how can you predict the negotiation?

Ben Leiken (23:17) Well, I think to build on a lot of things you just said there, the approach that we take at Arkestro is really rooted in behavioral science. And we believe there is a ton of relevant information that suppliers kind of metaphorically throw off, they and buyers as well, as they go through a negotiation process. Everything from, and we all kind of intrinsically know these things, like the proximity to the end of a sales quarter, the emotional aspect of needing to get a deal closed, of understanding your leverage in the market. All of these things matter a great deal when it comes to negotiation. And our platform is designed to capture these factors and leverage them into making better recommendations and decisions for procurement teams as well as for their suppliers. So we really see negotiation as a, yes, an analytical art, but a human driven construct and process. That’s really what we mean when we talk about negotiation science. We’re really focused on modeling the strategic dynamics of the sourcing decision or any procurement decision and looking at some of the factors I named in addition to just digging into who is participating, how many suppliers are participating, understanding the suppliers’ market position, strategic motivations and their current and past behavioral data, understanding that and then applying it and designing a process that maximizes, shortens the distance to get to the desired outcome and focuses on the optimal outcome from the very beginning. That’s fundamentally what negotiation science is all about. And I think there’s parallels to draw. I think most of the major tech platforms that we all use every day are rooted in this way of looking at the world. You look at Meta, Google, all of these major online advertising platforms. This is the premise that they’re built on, that we can, by tracking clickstream data, tracking end user behavior, we can infer a lot about what someone is going to buy, what someone is going to do. That is essentially the core seed of what negotiation science is all about and our perspective on negotiation.

Chapter 8 - The Car Buying Analogy: Anchoring, Suggested Price, and Supplier Coaching

Nikhil Varshney (26:07) Okay, I’m going to take one example and it’s going to be a very generic example and I think everyone probably has gone through this scenario. Everyone goes to buy a car, right? We all try to negotiate from the sticker price or the window price of the car. And I know it’s going to be very simplistic in terms of the complex world that you deal with, but I think the principles could be similar. So if I’m going to buy a car, what should be my behavior? How would Arkestro guide me towards a negotiation strategy? What has worked for me is how I remove the behavioral part of it is I don’t deal with the dealership directly face to face. I deal with them either on email or over the phone during the negotiations. Because then it keeps the emotions out of the conversation and you’re not dealing with a person, you are replying to a message. So that has been my philosophy, it has worked. But how would Arkestro then guide me to a better deal working with a dealership?

Ben Leiken (27:07) Yeah, so I love that example because I think for one part, you know, the fact that part of what we do is drive a uniform process and optimize that process towards a more impartial medium. Negotiating over the internet removes some of the, in the case of buying a used car, maybe the intimidation factor or the feeling that you’re standing in this crowded or intimidating dealership and getting pitched. So we remove any competitive leverage related to the location or the context of the negotiation. But the main area where Arkestro adds value in that example is arming you, the car buyer, with a suggested price. You’re coming into that negotiation knowing roughly what you should be paying. And I think we all know if you’ve ever bought a used car or paid for some kind of service on Thumbtack, right, had paid somebody to paint your house or mount your TV. Having an understanding of what the thing should cost is really powerful. And it’s also super rare. We’re all put in positions, procurement is put in these positions every day, where they’re asked to buy things that they’re not experts on. Their job is to buy, but there’s no possible way that any procurement person could be a detailed subject matter expert on every single SKU. So that’s really powerful. Again, being able to put a stake in the ground and know roughly what you should be paying. But on the other side of it, we would benefit the supplier, the used car salesman, with a feedback experience, a gentle, friendly kind of coaching experience to say, here are some levers you can pull for this person who’s trying to buy your car. You know, this specific feature of the car is really important to them, or hey, they actually need being able to drive this car off the lot tomorrow is much more important. They’re going to be more game to negotiate if they can get access to the specific model sooner. So again, just an example of how these behavioral factors and an understanding of the participants and their desires and what the optimal outcome is. What a used car negotiation would look like to a procurement user and to a car seller would be, here’s a suggested offer, if you meet this and you can meet these requirements, you’re going to have a sale today. And to the car buyer, it’s confidence that you’re getting what you want and that you’re paying a fair price.

Nikhil Varshney (30:08) Right, and I’m just going to ask a counter argument related to that because when we talk about suggested price, how I look at suggested price is that it creates an anchor bias, meaning that I could have gotten a lower price, I could have gotten a higher price, irrespective, now I’m anchored on that suggested price. From a behavioral psychology standpoint, my mind is now seeing dollar X as the right price, which could be wrong. At the end, it’s the suggested price from the model. Is that a concern that you have seen or have you faced from any of your customers that they have stuck onto that price, which is now anchored on? But there could be, it could be wrong because it’s kind of creating that bias in your head now.

Ben Leiken (30:58) Yes, that is a common question that we get from customers and from prospects when they’re first introduced to the concept. The inclination to say, what if the price is wrong, logically follows. I think what that largely boils down to in Arkestro is confidence in the baseline, right? Confidence that either the last price I paid or the baseline that Arkestro has assisted me in setting, that baseline might be too high and I’d be getting ripped off. So super common pain point and question that we get. The way that we address that in Arkestro, and again bringing it back to this used car example, is for these use cases, for these things I’m buying that I haven’t bought before, making it really easy to get market data. To get a representative sample. In Arkestro, we call this the intelligent counter offer, which goes out to market, identifies a set of relevant suppliers, and asks them for quotes before any negotiation begins. So in the used car example, it would be equivalent to going to CarMax or your Kelley Blue Book and pulling in an approximation of that price. It’s actually a little bit better than that because the Arkestro intelligent counter offer understands your business, understands the competitive context in which you operate. The companies that we work with are large enterprises, think Fortune 500, Chevron, Nissan, these companies procure on the scale of small countries and they have almost entire macroeconomic effects of their own. That’s just the scale at which they buy. And so building in an understanding of that into the intelligent counter offer process allows those companies to drive a contextually relevant baseline for the thing that they buy. Nissan buying paint looks very different than Rivian buying paint. Yes, they’re both car companies, but comparing those two baselines is not actually going to tell you very much. So that’s where our product is really focused, is how do we make it easy to get to a context-relevant recommendation that you can trust and trust quickly.

Chapter 9 - Game Theory Applied: Designing the Negotiation, Not Just Running It

Nikhil Varshney (33:38) All right, now I think I’m going to switch and flip that example of car buying from car selling. And the reason I want to do that is I want to now talk about game theory. And I’ve actually done that. So maybe it will fit the discussion that we’ll have around game theory. But the example is now if I have to go sell my car, I go to 10 different dealers and get a quote from each one of them. What I have seen is CarMax quote is usually the highest in terms of the money that they offer and the other dealerships offer lower money for that same car. But then if I have a CarMax quote I can play the game theory thing where I can go to the other dealership and say, CarMax is offering this much. Are you willing to give me a lower price than this? Because I’m not going to sell it at the same price. If it’s the same price, I can go to CarMax. So what’s the benefit of working with you? 80% of the time, the dealership would give you $500 or $1,000 better deal compared to CarMax while buying your car. How does that game theory aspect be built through technology and presented to your customers? That is one thing that I want to understand. How does that game theory model really fit into predictive negotiations, predictive procurement and how have you scaled it because every situation in game theory is different. How can you scale it as a technology?

Ben Leiken (35:05) Yeah, it’s a great question. I think a lot of what you’re describing there, in Arkestro land, relates to the way that we actually design the game, that we design the process and negotiation. And the way in which game theory is applied to that largely relates to event design. So not just which event to run, like RFQ, RFP, RFI, spot buy, but how to actually tailor the feedback that suppliers receive at each step, each turn in the game. So I think in the car buying example, the ability for the supplier to take the offer and shop it around, the way that that relates to Arkestro is we’re very thoughtful about the types of feedback and competitive context that we offer suppliers and how we do that. One example of this is what we found is that revealing a given supplier’s standing in the game, where they stack up on an individual line item or in the event at large, there are big benefits and big potential drawbacks associated with how that is done. A supplier who knows that they’re in first place often takes their foot off the gas. They stop competing. They know that they’ve got the best offer. They know that they’re going to win. And we’ve actually found telling suppliers they’re in first place often causes them to lose. If there’s another round of negotiation, and there’s a real chance that they get their position usurped, it’s not in the supplier’s best interest to tell them that they’re winning. So that I think is one concrete example that relates to the example you gave of like the way in which game theory is applied allows for suppliers to exert their leverage. And we’re thoughtful in applying it so that suppliers can understand where they have levers, where they don’t, and the same for buyers. And again, it’s all in the interest of arriving at that optimal outcome within the zone of possible agreement for both parties.

Chapter 10 - The Value Case: Time, Price, and Continuous Procurement

Nikhil Varshney (37:48) Got it. Makes sense. And I think what you just mentioned is around the negotiation is not over until it’s actually over. So even if you are the front runner in the first round of negotiations, you could lose in the second round, could be on multiple factors. From there, I think what you have just described, I want to switch gears a little bit and talk a little bit more about how do customers get value? And I think you have already talked and touched upon two great points. One is the price, which is the suggested price is usually lower than what they would have paid. So there is an absolute value gain and ROI emphasis that you can apply on day one once you have the suggested price in front of you and you can benchmark against it and compare your savings. The second you mentioned is the time that is saved in the negotiation process, which is usually months is now down to like days or even hours in some cases. What other benefits do you think the customers get or value customers generate from using Arkestro solutions? And if you want to double click on any of these examples of price and time, that is also great.

Ben Leiken (39:07) Yeah, it’s a good summary you just gave. What’s interesting and I think the common misconception of Arkestro is that it’s all about saving money, and the competitive leverage that Arkestro provides. What the predominating way in which Arkestro provides value to our customers, perhaps surprisingly, is the time savings. And the ability, yes, to run shorter, more effective processes that drive consistent results. In other words, the way that this shows up for our customers is, think of your most talented procurement folks. What if you could clone them? What if you could ensure that the minimum bar for all of your procurement activities is consistent with the output of your best people? That’s ultimately what we provide. And so ultimately what Arkestro is all about is allowing procurement teams to touch more spend, to actually get through their to-do list, to turn through their backlog and drive better outcomes. And that relates to time savings, but it’s really this is what process transformation unlocks for our customers, is this ability to supercharge your procurement people. And we talk at Arkestro, our vision we describe as continuous procurement. How do you put your procurement processes into a kind of continuous cycle, moving from what are traditionally very episodic processes in procurement. I run my roughly annual sourcing events. Sometimes I don’t run it. My plate is always overly full. There isn’t enough time or enough people to do all the things that procurement teams would like to do and for procurement teams to really operate from a true strategic position. This mindset shift that Arkestro allows our customers to get to in this transformation that we facilitate is one in which you have these continuous cycles. Everything that you buy is consistently managed and monitored and surfaced to the procurement team when an intervention is necessary. Ultimately, humans are at the center of this process. We believe the product is built in such a way that it’s centered around excellent particular way users orchestrating.

Chapter 11 - Does Automation Erode Supplier Relationships?

Nikhil Varshney (41:50) But isn’t that like if the process is centered around humans, then technology is basically replacing that human, right? Because now at the end of the day, negotiation has been the art about building relationships which build forward leverage. So if you have been working with one company and they have been giving you a good price, now you lose that relationship. Is that something that customers have felt or do you think that is going to be a problem where forward leverage that used to get generated through human to human relationships will now not be available and everything will be so mechanical in terms of negotiation and the beauty of negotiation was human relationship.

Ben Leiken (42:34) That’s a great question. So we’re big believers that great supply chains are built on great supplier relationships. And so we are not here to replace your supplier relationships and that fundamentally human relationship that matters so much to great supply chains and building great products and companies. And this is what I like, ultimately this cycle time compression time savings value add that we offer. The reality is that today, procurement teams are not spending as much time as they’d like to be spending with their suppliers and building great supplier relationships. That’s ultimately where Arkestro helps. Let’s get the cruft, the busy work, the amount of time it takes to prepare for negotiation, to decide what process to run, to figure out the target price, to figure out the baseline. The reality is that, not to mention just cutting POs and making sure that stuff shows up on the loading dock when it’s supposed to. It’s all of that stuff that occupies an inordinate amount of procurement time today. And we believe, with great software, that stuff becomes an implementation detail and you bake in the optimal outcomes into all of that stuff so that procurement teams can ultimately be more focused on the human element even more. And that’s a radically different perspective than some of our competitors for sure, but we think it’s the right one. I think we’ve seen significant evidence so far that it’s the right approach.

Chapter 12 - Customer Journeys: Chevron and Nissan

Nikhil Varshney (44:09) Got it. And Ben, if you could walk us through like with a couple of customer journeys around how this process actually starts, what is the part of behavioral science and the negotiation that is getting automated, and how does that result in more time for the customer to spend with their suppliers to build that relationship. If you can walk us through like one or two quick examples that might help viewers understand a little bit more about the nuances of where the technology is actually applied.

Ben Leiken (44:46) Sure, yeah, it’s a great question. So I mentioned a couple customers already, Chevron and Nissan are a couple of our notable lighthouse accounts, in the oil and gas energy vertical and automotive vertical. The story that we’ve seen with both is they came to us with a problem set very similar to what I described earlier. We’ve got too much spend, not enough time to address it all. And in Chevron’s case, an awareness that there were a number of specific use cases and verticals where there was just not enough people hours to stay on top of all of the procurement activity and to really, in some cases, negotiate at all. So there were POs going out the door that were just not negotiated whatsoever. And that’s a very common story. With Arkestro, in addition to our procurement approach, the thing that sets us apart from the pack is customers can get started really quickly with us. We have this implementation methodology called Live in Five. It’s exactly what it sounds like. It means we can actually get a brand new customer onboarded and off to the races running events in just five days. And that’s, looking at Chevron specifically, they were able to run hundreds of millions through the platform in a very short period of time. And that has allowed them to get their arms around a much larger swath of spend and spend a lot less time managing each individual process with each supplier. So they’re seeing a significant cycle time reduction that’s allowed them to do a whole number of different things, address other categories, focus, spend more time with their most strategic suppliers. That’s really in broad strokes the story with all of our customers. We approach these implementations focused on an area or set of areas where there’s acute pain. And then as we progress, we kind of climb the ladder of use cases with the customer. And that’s where the product really shines as well in terms of its flexibility and its horizontal application, climbing the ladder with Arkestro. We’re not just focused on a specific spend segment like MRO or tail spend. With Arkestro, you’ve got this broadly applicable tool set that goes back to the idea that it’s not just one form of AI. We’re really focused on applying the right form of AI to the right business process. And so that ability to climb that ladder and add more spend and really kind of build the flywheel of value for that customer, of accruing value, is really strong and something we’re super proud of.

Chapter 13 - The Numbers: Savings Benchmarks

Nikhil Varshney (48:02) Makes sense. And I think what I would also like to double click on in terms of value, and I think maybe I’m going back to this again in terms of suggested price. Like if I were, and I know and I also know that you said that it could not be the most important part of the value that Arkestro provides. But I think this is one of the more tangible and measurable outcomes that can be quickly measured in terms of the value gain from any platform, which is dollars. Can you give us directional numbers, probably not the absolute numbers, but directional numbers in terms of how much savings, for example, a Chevron or a Nissan or any other customers have received because the suggested price that the model came up with was better than what they would have negotiated otherwise.

Ben Leiken (48:54) Sure. On average, we’re seeing it’s typically somewhere in the range of 12 to 15% with those customers. So it’s a significant amount. Wide range, depending on the category as well. We often see outcomes significantly higher, but that’s where we hang our hat as far as savings are concerned. And then in addition, cycle time reduction of typically on the order of like 3x what they’ve been spending.

Chapter 14 - Business Model and Funding

Nikhil Varshney (49:22) And I think this kind of pivots me to one of the last sections that I want to talk to you about. Funding story, etc. But I think before I go into funding, a lot goes into the cap table before any funding is decided is how much revenue can you generate and how will you pay back to the investors, right? And I think for you, the revenue model, I believe is going to be somewhere around the service based as well as if you’re taking a cut of the savings. I don’t know if that is the revenue model, but can you give us a directional understanding of what your revenue and the business model is? And then take that to your funding story as well.

Ben Leiken (50:04) Yeah, it’s a great question. So we try to keep our revenue model very simple and very aligned with a win-win for when the customer wins, we win. And so like a lot of other AI native companies, a consumption-based model is what we use. So the more spend you manage with Arkestro, the more you pay us.

Nikhil Varshney (50:30) So it’s basically like a fee for service model, which is like the usage of the API. Is it like the token based model that AI companies use these days?

Ben Leiken (50:38) Very similar. We don’t call it tokens exactly, it’s a spend under management type of model.

Nikhil Varshney (50:46) Got it. And please take us to the funding story as well and the valuation story as well.

Ben Leiken (50:53) Yeah, for sure. So as you can imagine, I can’t share a ton of details on the specifics there, but really excited by the momentum that we’re seeing in the market and particularly customers. I think that the investments, the investors, there’s a lot of investor attention on Arkestro right now which we’re really proud of. We’re super proud and humbled by the collection of investors that we currently have. We’ve got an amazing board, an amazing set of pretty recognizable names there. I think a headline by NEA and Jeff, former CEO of GE, sits on our board and is a tremendous strategic asset to the company. We’ve most recently announced a $36 million strategic investment round led by Aramco Ventures, that was last summer. And there’s a huge amount of momentum behind the company and a huge amount of demand for predictive procurement. So we’re striving to meet the moment and excited to have a set of investors and just a very receptive market. I think there’s, we’re at this big moment that AI is having paired with, I think really dating since the pandemic, this focus on supply chain and procurement, and only made more relevant by all of the macro uncertainty around tariffs. It’s a tumultuous time, but predictive procurement is incredibly relevant in the market right now. And that’s something that we’re excited about.

Chapter 15 - Building Technology at Startup Scale: Roadmap Discipline

Nikhil Varshney (53:07) Sure. And one last question that I would like to end with is, I know you have worked with big technology companies as well in the past, working at Arkestro, what is the difference in terms of building technology for a startup that is getting started? And you have a variety of clients, so you’re not focused on a specific industry and scaling in that industry. You have a breadth of clients and you’re working across the industry. How messy it gets in terms of building and prioritizing technology, what is your mantra and how are you making sure that your roadmap stays sane and does not ship features that you might not want to ship at certain point in time.

Ben Leiken (53:50) Yeah, it’s such a good question. And I’m sure you can relate at your role at Wayfair that there’s always, no matter what size you’re at, there’s always a challenge balancing the various demands on the business. I think the way that we look at it on the product team at Arkestro is we are fundamentally a transformational technology company and our ability to continue to chart a course for predictive procurement and to deliver game changing new functionality, that’s really core to what we do. But we pair that with the other equally relevant reality that we are serving enterprise customers, we take that job very seriously. Investment in everything from security to compliance to quality. That’s certainly something where we’re always looking to raise the bar. And I think maybe getting to the root of your question, looking at the company seven years ago when we were Bid-Ops versus now, I think it’s an increasing focus and need to focus on those sorts of capabilities that’s risen on our roadmap to the point where it’s at now where we’re serving again, some of the largest companies in the world. And we’re really proud of that. But they’re ultimately signing up with us and counting on us as a transformation partner and really a thought partner for how to thoughtfully apply AI and behavioral science in procurement. So we always need to be pushing the envelope there. And we’re really proud of that.

Chapter 16 - Closing

Nikhil Varshney (55:30) Now, thank you, Ben. It was a fantastic conversation. I learned a lot. And thank you for taking up with my examples, which are not complicated around car buying and explaining the thought process behind Arkestro’s science of negotiation and procurement. So really appreciate your time. Thank you so much for coming.

Ben Leiken (55:49) Thank you so much, Nikhil. It’s a pleasure and yeah, pleasure meeting you and just so impressed by what you’ve built with Silk Road Nexus and excited to follow the progress. Thank you.

Nikhil Varshney (56:02) Thank you so much.

Thanks for reading THE SILK ROAD NEXUS! Subscribe for free to receive new posts and support my work.

Thanks for reading THE SILK ROAD NEXUS! This post is public so feel free to share it.

This is the problem Andy Kohm, CEO of SCIP (Supply Chain Intelligence Platform), set out to solve. A biomedical engineer by training, Kohm first encountered the pain of fragmented data while sourcing vendors for R&D projects. The information existed somewhere in the organization, but it was inaccessible, outdated, or trapped in systems he couldn’t reach. That experience became SCIP’s founding insight: the gap in supply chain technology isn’t another application. It’s the intelligence layer that sits across all of them.

SCIP ingests data from ERPs, PLMs, warehouses, and procurement systems, then aligns it against roughly 200 universal supply chain attributes. The platform doesn’t replace existing tools. It activates the data already living inside them. Where AI enters the picture, Kohm is deliberately pragmatic. Agents handle data cleansing, manufacturer matching, and part-number resolution, but they operate behind the curtain. The customer never builds or manages an agent. AI creates the rules; deterministic logic applies them at scale.

The business model is equally deliberate. No seat licensing. Four modules (visibility, data health, optimization, and risk scoring), priced by company size. The logic is simple: more users means more interaction, more feedback, and a stronger proprietary data layer that compounds across the customer base. That data, anonymized and enriched over time, is the moat. Software can be rebuilt. The accumulated context of millions of cross-customer data points cannot.

Thanks for reading THE SILK ROAD NEXUS! Subscribe for free to receive new posts and support my work.

Thanks for reading THE SILK ROAD NEXUS! This post is public so feel free to share it.

(Transcript)

Nikhil Varshney: Welcome to Silk Road Nexus Conversations. My name is Nikhil. I am privileged to have Ita with me today, who is the CEO of Veho which is the national carrier disrupting the entire logistics industry at this point in time. Ita, thank you for doing this with me.

Ita Zur: Thanks for having me.

Nikhil Varshney: And if you could please introduce yourself and Veho to the customers.

Ita Zur: So my name is Ita, and I’m the co-founder and CEO of Veho. Veho is a delivery platform that helps e-commerce brands provide an incredible customer experience with every delivery. So we take care of the entire delivery experience all the way from the distribution center of the brands that we work with to the customer door. And our focus has always been around the customer experience. Customers who receive the box from Veho have full control over their delivery. They can communicate with customer support in real time, can leave instructions for the drivers, they can reroute, reschedule their deliveries. And then if there’s any issue, they can troubleshoot in real time.

Really what we’re going after is proving the delivery should not be an afterthought. It should be part of the customer journey and the better delivery experience results in a higher customer lifetime value. So we’ve been doing this for companies almost 10 years old now today in 65 US cities working with some of the top household brands in the United States and growing very fast.

Nikhil Varshney: And I know you’ve said a lot of big brands which are Lululemon, Luxotica, Saks, Adidas I believe as well. And I think when you work with these brands and any other brand for that matter of fact, trust is very important. Ita what is your mantra for trust? How do you define it? And what is one KPI that you use to measure trust?

Ita Zur: Well, the first thing we do with brands is educate them about how we think differently about logistics. And I want to explain that, but also, obviously, the education is not good enough to build trust. actually have to live up to the high standards you’re setting for yourself every single day. That’s where trust really matters. But education matters too, right?

When I started Veho, it was important for me to reframe the conversation from logistics as a cost center to logistics as a value center. Think about it. You’re spending a certain amount of money and in delivery, usually it’s the second largest client line item on the e-commerce company’s So you’re spending a lot of money. You know, what are you getting in return for it? You know, when you compare two different delivery providers, does one reduce your...

Incoming customer calls, customer complaints, refunds, credits. Does a certain provider allow you to increase your customer retention, customer lifetime value? That is what I mean by return on shipping. So what we do with the brands that we work with is we basically frame the conversation about how we can help you improve your own business goals. Not just cut shipping costs, and not maybe one of them, but improve your own strategic business goals.

What metrics are you optimizing for?

Is it Net Promoter Score, customer lifetime value, et cetera, and we hold ourselves accountable to those metrics. Then there is execution. We are extremely focused on providing a high on-time delivery with high customer satisfaction. We look at these metrics every single day. We hold ourselves accountable. We held weekly calls with our clients. We exchanged deeper data QBRs. We go very deep. get feedback. Ultimately, trust can only be built.

if you live up to the expectations you set for ourselves. We set a very high bar for ourselves. When we’re not there, we’re not always perfect. We take accountability, we will compensate when we’re worse. We want to make sure that the brands know that we see ourselves as their extension to the end customer. And if you don’t do that, that’s not good enough for us. And that’s how we are able to provide that every single day with that level of seriousness around this.

Nikhil Varshney: No, that makes absolute sense because when you look at Cost Center, which has been the mindset in the retail industry for a very long time. For logistics, it’s like a backend service and you basically just hand it off to a legacy carrier and hope that everything goes well and that your customer doesn’t call you back. How does that conversation actually go? Because the mindset is still like logistics is a cost center for a lot of these companies. What is that convincing power that goes in to say like, okay, it’s not a cost center. You talked about value through ROIs, through multiple levers that you can pull, but it’s still the mindset. Like how do you change that mindset or is it changing right now?

Ita Zur: First of all, it’s changing on its own. I think more and more companies are starting to realize they should think more strategically about their logistics as a value creation center. And by the way, this doesn’t come from Veho per se. I think Amazon has really shaped the way e-commerce companies think about their delivery as a way to increase conversion on the checkout page, as a way to delight the customer, to create customer loyalty. We just tell the story and we bring our own data to do that.

But let’s say somebody comes to the conversation and they’re not convinced. Well, we show them case studies, but we also do another exercise that is really interesting. It’s called the return on shipping calculator. And basically what this means is we ask them to say, how much do you spend per month with a certain provider, a carrier of their choice out there.

How many customer calls do you get, complaints? How many of those packages have been lost or stolen? And how many of those packages arrive, but they arrive late? What does that cost - to your retention when they arrive late. And we build basically a spreadsheet that calculates all these adjacent costs, or I would say even costs, that shipping managers don’t always think about. And we compare it one by one versus Veho’s metrics. And why would it cost you to work with Veho? On time delivery, the number of packages get stolen or lost or whatever that is, customer service calls. Also apply the factor of retention. If you’re able to get it back on time, deliver it on time, roll the customer in.

Then keep on buying from you. And then you bring all those numbers down to a number per package. And again, that is nice in theory. Not everybody buys into it. They have to see it in their own eyes. So usually they launch a pilot with us. We’ll ask them to hold us accountable to certain metrics and compare us with other providers so they can see. And then usually what happens is that we launch in three, four markets. They see that this is far better than what they expected. That’s how we’ve been very successful so far.

Nikhil Varshney: And I think one part of trust that I would also like to kind of ask you about, and this has been one of the questions that I have taken a survey on, on LinkedIn and other platforms as well, and this survey had close to 5,000 responses. My question was, would you trust a gig driver versus a professional driver making a delivery for a luxury product? So my question was specific to luxury. And the answers were heavily skewed in the direction that they would want a professional driver making that delivery.

Ita Zur: Yeah, that’s understandable.

Nikhil Varshney: How do you then change the mindset? It’s not only the brands, it’s also the customer buying from the brand that has that mindset that if I buy from it, I want a professional delivery service. So how do you encounter, how do you counter that shift in the customer’s head as well as the brand who’s selling to the customer?

Ita Zur: Well, one thing to remember, and I’ll answer your question directly, but one thing to remember is that for years and still today, Amazon uses gig economy to deliver packages.

I think we can all agree, [Nikhil Varshney…yes], that Amazon is a premium delivery service in the United States. Consumers, typically, 99 % of time, don’t even know that it arrives from a crowdsourced driver. And by the way, if we are, what is the most expensive or the most valuable thing in our life? It’s ourselves, it’s our families. Would you take your family on an Uber? The answer is yes.

So the problem is not really with the Gig economy driver. The challenge is with the technology, the visibility, and the accountability. When you don’t have technology to hold the driver accountable to a high level of standard, or to track and trace, then we all defer back to the old system that we learned to trust over time. But when technology is there, technology can not only make the crowd source driver, the Gig economy driver, just as good as… somebody’s been doing this for 25 years, they can actually make it better. And the reason is that now you have a carrot and stick approach, if you will, right? Customers get to rate the service.

Drivers want to drive on the platform. Drivers have to be able to be compliant with certain metrics. And if don’t hit the metrics, you don’t get the opportunity to drive. Through technology, we have visibility across the board. Anything goes wrong, we’re able to catch it in real time. So again, all that is a nice story. It comes down to the data.

We work with some of the top brands in the country. And what they’re saying is that Veho has the highest on-time delivery rate in the country. Customers are happy. I think ultimately that’s what matters. It’s not really what the driver is, but what is the product and the result of the delivery that we’re so obsessed with?

Nikhil Varshney: Makes sense. And just because we are in the economy and we are talking about drivers a little bit over here, I know you have close to 120,000 drivers in the gig model working for you. There is a lot of criticism that happens about the gig industry in general.

How do you think this particular industry, the gig driving industry can be made better for drivers? What are your thoughts on that? So it’s more a general question, not specific to Veho but I just want to capture your thoughts around this.

Ita Zur: I think about delivery or ride sharing or any work you can do on a flexible basis as a product. So when you go to a store,

I used to be a brand manager before I started Veho, so I understand this pretty well. You can buy Pampers or you can buy Huggies, which is a different type of, I’m saying Pampers because I was brand manager of Pampers. And these are different products. For an individual looking for a work, they’re thinking about all these platforms as a product. And what we’re trying to do at Veho is build a better product for the driver. So let’s talk about the specifics. When you’re an Uber, you know, it’s on demand.

You don’t necessarily know how much money you’re gonna make during that day. You’re also driving passengers around. There’s always, at least from a perception standpoint, a little bit of a risk with another person sitting behind you, you don’t know them, driving at night, et cetera. When you’re doing food deliveries, for example, that’s tough. You gotta wait in line in a restaurant, they may not be ready on time, you gotta hustle away, you know. And again, you don’t know how much money you’re gonna be making. If it’s a rainy day, a lot of people order a lot of food online, you’re making good money.

If it’s sunny outside, people go out, you’re not necessarily making a lot of money. We change the way drivers look at the gig economy. With Veho, routes are posted the night before. You know what route you’re signing up for. You decide where you want to work. You decide whether the price matches your needs or not. You can sit and wait. Prices may go up, and then you make more money. You’re not driving a passenger in the car. Specifically, day hours. So if you’re a stay-home parent or you need to, you know… pick up your kids from school, you’ll probably still be able to go and pick up your kids from school because you signed up for a route earlier in the day.

There are many reasons why this is a better product for the driver, right? And because it’s a better product for the driver, one of things that we see is that many of those drivers come back and want to drive with us again and again and again. We cannot guarantee that they get routes every day, but there’s a tendency to see themselves as affiliated with our brand. Some of them will actually go and put the Veho. shirt or Veho logo that’s completely up to them.

They are independent contractors but you can see themselves part of the mission. And that gives you a sense that they see us differently than you see other Platforms.

Nikhil Varshney: And basically what you’re saying is that the technology has enabled you to make life flexible for them in multiple ways and that flexibility is what is required in this economy as well for them to feel like they’re part of the industry and they can work full time over here as well.

Ita Zur: Flexibility is important, but flexibility without consistency does not work. If I want to have a gig and drive every now and then but I don’t know how much money I’ll be making and can’t make ends meet, then this is not going to work for me. I got to feed my family. So what Veho does is provide consistency. You know exactly how much you’ll be making. You have a lot more control over how you make the money, where you drive than I think on any other platform. And that sort of seals the deal for them.

Nikhil Varshney: I would like to shift gears a little bit and talk about a couple of products that Veho offers. FlexSave. which was recently launched, is one of the prime products. I think no one is offering that kind of service right now, which is you have a flexible window that allows brands or customers to kind of make sure that they can purchase on the cheapest available day. That saves money. I think I can compare it to Google Flights, where it kind of makes your flight flexible. Are you flexible or are you fixed on the dates? So flexible, it gives you cheaper options on plus, minus one, two days.

How are you seeing the reaction to FlexSave? What are some of the other benefits of products like this? And how are brands using this and perceiving this?

Ita Zur: I love the analogy to Google Fights. I saw that, I think you posted it on LinkedIn as well. I saw that as well. It’s a great analogy, right? Essentially what FlexSave does is it says, well, some brands or some products, they really want a day-shortening delivery.

So if you ship on Saturday, you and you promise two day shipping on the check-out page, it is really important to live up to the brand promise to bring it by monday In other cases, for other brands, actually saving money on shipping is far more important. So they are willing to flex and open the delivery window a little bit. And when they do that, that gives us more time to make optimization in the warehouses. We have a technology platform, it’s AI, it’s called Maestro AI, it’s trademarked orchestration platform that allows us to make optimization in real time, drive up utilization, drive down the cost, pass down the savings to some of shippers.

So, you just launched this last week. We’re seeing a lot of demand, a lot of people, a lot of interest from a lot of brands.

What’s generally happening in the market, a very strong tailwind. Every year, January 1st, the legacy national players will post annual GRI increases, typically 5.9 % a year. Postal service last year also increased by more than 10 % or so.

So if you’re an e-commerce company, you are [under the hammer]. You have to pay more and more to your shipping providers while your customer is more and more expecting free shipping. So there are categories where this is so vital, right, to be able to free shipping and the margin may not be very high, that those companies really need a better solution.

FlexSave gives them that solution. It’s a great customer experience, it’s fast delivery for the most part, but that flexibility allows them to save money. By the way, from the customer standpoint, it’s still a great customer experience. Customers can interact with a customer supported real-time, leave instructions, upload a photo of where you want the package to be delivered, talk to live customers to see if there are any issues and troubleshooting, et cetera. It’s a great customer experience. It just may be a little bit slower in a way that allows the brand to offer free shipping, so it’s a win-win for everybody.

Nikhil Varshney: And you touched a little bit about this where customers are able to customize their delivery options. How does it really work? Because Veho is a carrier partner and brand owns how they communicate with the customer. So at the end of the day, let’s say I’m buying something from Lululemon. Lululemon is showing me the delivery date. I have already provided my delivery address. Where is the customer going and updating? Is it the brand’s own page or brand… transfer to your website where they can go and update the information.

Ita Zur: When you buy it online, Some Brands will offer you to leave instructions on the checkout page. Amazon does it, Some Brands as well. Those instructions then get handed over to us through an API. So our driver partners, when they show up at the address, they see on the app all these instructions. But let’s say the Brand doesn’t give you that option. And you can have the conversation directly with Veho. In case we send text messages, you can reply to those messages.

And so a message would say, from Veho, we’re delivering your certain brand’s package. It will arrive tomorrow between this time and that time. You can actually reply and say, I’m not going to be home. Or I need you to leave it in the back door, because otherwise somebody will steal it. And you’re going to have a live conversation. A lot of it is automated these days with AI. But there are edge cases where you have a human being jumping on a call to have that conversation.

If you can’t get it delivered because you’re out of town, you’re concerned that somebody may take it from your doorstep, know, Porch Piracy, you can tell us, we’re gonna hold this another day, get it out before anything. Now the idea over time is not only to slow down the delivery if you want us to hold it, but also speed up. Let’s say the brand that we work with buys FlexSave And let’s say that you get a notification and it says, today it’s Saturday, we’ll deliver it by Wednesday. You can then say, well, I know I got free shipping from the brand,

But I actually want it faster. I’m going to a wedding, I want to wear that dress, I ran out of socks whatever that is. The idea is that you’ll be able to then upgrade yourself and get it delivered faster over time. So that is the kind of control we’re talking about. It’s not just communicating, it’s also deciding when and how it will be delivered.

Nikhil Varshney: And I think that is a fantastic lever for the business as well because now you have a direct relationship with the customer, not just the brand. And at some point in time, that relationship allows you to understand the behavior of the customer as well. Like, Nikhil always prefers deliveries on Wednesdays, so if Nikhil is buying on any of the brands, that gives you the power to kind of make suggestions. Is that something that you think would happen or Veho would like to pursue that opportunity as It is true that the data can be used in many ways.

Ita Zur: One thing to consider is that this is not our customer. It is our clients’ Customer, Yes. And so the first thing and the most important thing in our mind is how do we protect the data for them?

And how do we make sure that we don’t use the data in a way that allows us to benefit other brands unless we have an agreement about this, right? And so this is probably the most important thing. But you are correct. When a customer gives us instructions, right, as an example, they don’t want to give us the instructions seven different times for seven different brands.

So if we get the approval of the brand no way are we going to associate those instructions with the address, never with the person. Within the address and then basically drivers can follow those instructions for the next brand. And it makes the system better and better. For everybody, right? It’s like a network effect because you know when you join Veho as a brand, you’re able to enjoy some of the benefits of having had those interactions. But again, I want to caveat everything. All that is subject to agreement and privacy laws that we hold ourselves very, very accountable to.

Nikhil Varshney: Absolutely, absolutely.

Coming to the second product, I think you already touched on it, Maestro AI, which is the great platform that has been built, which is optimization, as you said, is an optimization engine. Can you tell us a little bit more about how that works? How are you using prioritization? How are you deciding what goes on Monday, what goes on Wednesday? Is it volume based? Is it price based decisions? How are you making these decisions

Ita Zur: Well, Maestro AI essentially allows us to, as we scan millions of packages into our network, make split-second decisions about how to optimize those packages. Think about it. We have multiple data points about how we get packages. We get the data from manifest, from the API of our clients, but we also have the physical raw data where packages flow through our network when we scan those packages. What Maestro does essentially compares what we expect to get in the manifest versus the real-time data.

There’s usually never 100 % absolutely. There’s never 100 % overlap. There’s always packages we expected to receive and we did not receive. There are packages that we did not expect but we did receive. And so Maestro takes all that into account. And in real time, based on the priority of the package, it allows us to orchestrate the platform. So let’s say we are selling to a certain brand, Ground Plus, two day delivery from, say, New Jersey to Miami.

Ground Plus is a fast ground product. That means that those packages are always going to be prioritized in the warehouse. Let’s say another brand is FlexSave. So Maestro takes into account the fact that it’s FlexSave, and then allows the warehouse employees or workers to say, okay, this package should get on the truck to maximize utilization, or should we set it aside because we technically have two more days to deliver it, we don’t have to do it right now. And it does it constantly in real time. So you can see how when you have more and more data.

You’re able to maximize utilization without degrading the quality of on-time delivery because it’s not, the most important package is always pushed out first, if you will.

Nikhil Varshney: Great, I think that makes sense and I think we definitely need tools like Maestro AI which can in real time decide how to optimize, how to make sure that you have economies of scale and you’re able to make cost savings from that. Just to wrap it up, what is one advice that you would like to give to a younger person or to budding founders who are looking to break into supply chain specifically, not in any other space, but supply chain? What advice would you give to those founders?

Ita Zur: Well, I can give a lot of advice, but look, I mean, when I joined this industry, it was 10 years ago, I did not know anything about supply chain. And I was willing to challenge myself to go into a space that I did not know about and learn everything from scratch. I had my own convictions about what needs to be built. I had a conviction, which is still today,

The customer experience is the most important thing, but there are a lot of things that I didn’t know along the way. And the advice I would give myself, and I would give anybody else, is you have to be open to ask a lot of questions and challenge yourself all the time. The things that we think we know, A, they change, B, sometimes we learn new things. If you have a very clear picture and you just follow this picture and you’re not willing to be challenged about that picture, you may be missing a lot of big things along the way.

So I’ve been fortunate to have a lot of amazing advisors and people who contributed their knowledge and experience to make Veho a better company all the time. I think it makes us unique because we constantly learn and adopt and change as opposed to being very rigid in our own ways. I think that flexibility when you go into a supply chain that is changing so fast and it’s so complex, it’s probably one of the most important things you can do.

I will say one more thing, stay the course. Stay the course.

You know, it’s not… A lot of logistics companies aren’t built overnight. It takes time to build a company in this space. And if you have conditions, if you’re really passionate about this, keep on doing the thing. It’s not going to happen immediately. There’s going to be ups and downs. As long as you keep on doing this, ultimately, if you have a really good product, the value will be adopted, and then value will be created from that.

Nikhil Varshney: Thank you so much, Ita. It was a pleasure having you.

Ita Zur: Thanks for having me today.

Nikhil Varshney: Hopefully you had a great manifest. Thank you

Thanks for reading THE SILK ROAD NEXUS! Subscribe for free to receive new posts and support my work.

Thanks for reading THE SILK ROAD NEXUS! This post is public so feel free to share it.

The gaps in supply chain

I’ve been covering supply chain technology for a while now, and one thing I keep coming back to is a gap I see in how AI is being discussed versus how it’s actually being deployed. The press releases and fund raising rounds talk about agents, autonomy, and transformation. However the operations floor tell a different story — contextual manual processes, indefinite co-ordination and working around exceptions.

That gap is exactly what drew me to this conversation.

I sit down with a lot of founders and executives. The conversations are deep, problem driven and cover the breadth of the supply chain industry. But the ones I find genuinely valuable to talk to share a specific quality: they make bold predictions. Great founders and leaders back these predictions with operational logic, not just market narrative.

Yair Weinberger is one of those people.

Four Predictions Worth Taking Seriously

Within the first few minutes of our conversation, Yair had laid out a set of views that are both counter-consensus and, I think, correct:

→ The SaaS moat is shrinking. Margins that ran at 70–80% will compress toward commodity levels as the cost to build software collapses. Pricing power will follow.

→ The new moat is process context. Not the model. Not the data pipeline. The accumulated knowledge of how work actually gets done — edge cases, exceptions, institutional memory — extracted from real-life handling of hundreds of thousands of cases.

→ AI forgiveness is asymmetric. Users forgive human mistakes far more readily than AI mistakes. This changes everything about how you have to build trust, score performance, and deploy systems.

→ AI software must mirror human onboarding. For it to work in operations, it has to learn the way a new employee does — through real cases, real feedback, real exceptions. Not from documentation. Not from SOPs. From doing.

His views are testament to real operational know-how and expertise. Yair has actually lived inside the problem. Before founding Reindeer, he built Luma, a data pipeline company acquired by Google, and spent nearly a decade running large-scale ERP integrations at Google scale. He knows what it looks like when enterprise software fails to embed into operations. He’s built the integrations that made it work. That operational credibility runs through everything he says.

The Edge Case Problem

What struck me most was his framing of the edge case problem. Something I’ve seen play out firsthand in my own work on optimization systems. Here’s how he put it:

“There are a lot of edge cases. When you try to write them down, you never think of all of them. But once you’ve spent enough time with a company, you know what to do in every situation. At some point you become trained in all those edge cases. Trying to map them out in advance is very, very difficult — you need to understand, in real time: is this an edge case? Do I need to ask a question? And if yes, who do I ask?”

This is precisely the problem that breaks most enterprise AI deployments before they start. The process isn’t in the SOP. The process is in the people. And Reindeer’s core thesis is that the only way to solve this is to build AI software that learns the way employees learn — through experience, feedback, and accumulated exposure to exceptions.

His view on the shifting SaaS moat is equally clear-eyed:

“I do think there is still a moat around who owns the context and process knowledge — the business context, the product knowledge — because these are things extracted over time and only from real-life handling of cases.”

I’ve written before about how AI hype in logistics often misses the operational physics of what’s actually happening in the yard — read that piece here. This conversation is a useful companion: what does it actually look like when AI does embed into the operational layer? Yair has a concrete answer, and a live CPG client processing 7 million invoices a year to back it up.

This is the kind of conversation I come to Manifest for. I hope you find it as useful as I did.

What I’m Still Thinking About

A few things struck me about this conversation that I keep turning over. Yair’s distinction between system of record and system of work is one of the cleaner frameworks I’ve heard for explaining where AI agents actually fit in a tech stack that isn’t going to be rebuilt from scratch. Most enterprise operations aren’t replacing their ERP. The question is what sits between the ERP and the human — and that’s where Reindeer is betting.

The 7 million invoice case study is also worth sitting with. That’s not a speculative use case. It’s in production. And the workflow Yair described — Outlook inbox to vendor portal to SharePoint to approval, fully automated with an audit trail — is the kind of thing that sounds trivial until you’ve tried to build it. The variance alone across vendor formats and contract terms would have defeated most pre-AI automation approaches.

The question I’m still chewing on: how durable is the process-context moat, really? Yair argues that 500,000 handled cases is hard to replicate. I think that’s largely right — today. But the pace at which foundation models are improving changes that calculation over time. The companies that will win aren’t just the ones with the most cases; they’re the ones with the best feedback loops and the deepest operational trust with their customers. That last mile is still very much a human problem.

I want to be clear about something before you close this tab. I find Yair’s thinking sharp, his operational credibility real, and the problem Reindeer is solving genuinely important. But I’d be doing you a disservice if I left this conversation without a word of caution, because I’ve been here before. We all have.

During the dot-com bubble, somewhere between 1995 and 2001, thousands of internet companies flooded the market with compelling pitches, credible founders, and real problems worth solving. Roughly 90% of them didn’t survive. Not because the internet wasn’t real. It was. Not because the problems weren’t real. They were. They failed because demand didn’t materialize at the pace that capital and enthusiasm assumed it would.

The market was front-loaded with supply — websites, platforms, storefronts — long before customers were ready to actually use them at scale.

We are in a structurally identical moment with AI applications.

The tax on creating software is now effectively zero. A founding team with a sharp thesis and access to foundation models can ship a working product in weeks. That is genuinely remarkable — and genuinely dangerous, for exactly the same reason the low cost of spinning up a website was dangerous in 1999. When the barrier to supply collapses, markets flood. And flooded markets don’t sort themselves out by quality alone. They sort by who has the runway, the distribution, and the customer relationships to survive until real enterprise demand catches up.

Enterprise AI adoption in supply chain is real. The pain points Yair describes — invoice processing, quote automation, load optimization — are real. But enterprise procurement cycles are slow. Change management in operations is slow. Trust, as Yair himself noted, is hard-won and easily lost. The question isn’t whether the problem is worth solving. It’s whether enough companies will move fast enough, and spend real money, to sustain the wave of vendors now competing to solve it.

A Word of Caution — Because History Deserves One

we will see significant consolidation in this space within the next 24 to 36 months. The companies that survive won’t necessarily be the ones with the best technology. They’ll be the ones with the deepest customer relationships, the most auditable outcomes, and enough capital to wait out the gap between market enthusiasm and actual enterprise budget cycles. Process context — as Yair argues — may be a moat. But only if you’re still standing when the market figures out what it actually wants to buy.

Be curious. Be engaged. But be skeptical. The best ideas in any bubble are almost always real. The question is timing, and timing has killed more good companies than bad technology ever has.

If you found this useful, explore more analysis at The Silk Road Nexus — the only newsletter analyzing the unit economics of the physical world.

Thanks for reading THE SILK ROAD NEXUS! Subscribe for free to receive new posts and support my work.

Thanks for reading THE SILK ROAD NEXUS! This post is public so feel free to share it.

Visibility without action is just an expensive dashboard. After decades of enterprise software investment, that gap between information and decision is still wide open.

Before I dive too deep, I want to talk introduce both Nauta and Valentina.

Nauta

Nauta is an AI-native operating system for global supply chain, built for wholesalers, distributors, manufacturers and traders. It sits on top of existing ERP, TMS and WMS systems, extracting and contextualizing data to create a single source of truth. Its core thesis is connecting inventory with logistics at an SKU level, giving shippers control over what matters: not where the container is, but what's inside it. Nauta runs 18 agents that flag risks, identify patterns and convert insights into automated actions. The company prices by order volume and targets companies between $200 million and $2 billion in revenue.

Valentina Jordan

Valentina Jordan is the co-founder and CEO of Nauta. Her background is in product and technology, with twelve years of experience building for last mile and B2D across supermarkets, retailers and restaurants. She came to global supply chain through Nauta, where her co-founder brings deep operational expertise spanning an entire career in the industry. Valentina's product philosophy is deliberately unglamorous: build the foundations right before chasing the flashy stuff. In under a year of operation, she has grown Nauta to the point where clients are proactively opening their financial and procurement data, the clearest signal of trust an enterprise customer can give.

Coming back to the conversation with Valentina.

The Data Is Not Bad. It’s Chaotic.

The persistent misconception is that the data problem is a quality problem. Clean the data, standardize it, and the rest follows.

It won’t.

A single product moving from Guangzhou to a shelf in Dallas touches twelve or more independent companies, each with their own systems, formats, and definitions. You are never going to normalize that at the source. The fragmentation is load-bearing. Most enterprise software assumes clean data as a prerequisite for intelligent decisions. In supply chain, that assumption has been quietly killing products for thirty years.

The Context That Lives Nowhere

Even if you solve the data chaos, you still haven’t captured the most valuable information in the operation. Because it isn’t in any system. It’s in the people.

It’s the coordinator who knows a carrier always runs late in Q4. The procurement manager with a handshake understanding that appears in no contract. The operator who knows exactly which exceptions resolve themselves by Thursday.

When you reduce the team that carries this knowledge, you don’t just lose headcount. You lose the institutional memory that makes the operation function. Strip out human context too fast and you’re not running a leaner operation. You’re running a more fragile one.

Agents Are the Easy Part

Agents are easy to build and easy to demo. The problem is they’re the least differentiated layer of the stack. An agent is only as good as the data and context beneath it. The real work is in the foundations: the data architecture, the context engine that knows not just what happened but what it means for this company, with these suppliers, in this market.

Most companies are building the agent and hoping the foundation is good enough. In supply chain, where edge cases are endless, it usually isn’t. Unreliable in production is a trust problem. In an industry where trust determines adoption, trust problems are fatal.

An Honest Word of Caution

The cost of building software has collapsed. A small team can ship a working product in weeks. That’s remarkable, and it means the market is flooding. Flooded markets don’t sort by quality. They sort by distribution, customer relationships, and runway.

The companies that survive the coming consolidation won’t necessarily have the best AI. They’ll have the deepest customer relationships and enough capital to wait out the gap between market enthusiasm and actual enterprise spend.

The founders who understand what the technology can do today versus what it will eventually do, and build accordingly, are the ones worth paying attention to.

Valentina Jordan is one of them. Enjoy the podcast.

Thanks for reading THE SILK ROAD NEXUS! Subscribe for free to receive new posts and support my work.

Thanks for reading THE SILK ROAD NEXUS! This post is public so feel free to share it.

Last week I posted about Terminal Industries, while I was still working on finalizing the video. This is the full interview with Darin Brannan, CEO of Terminal Industries.

Most supply chain tech focuses on the warehouse or the customer, but one-third of the territory remains largely unmodernized: The Yard.

Darin is a serial entrepreneur and Terminal Industries is his fifth startup. Learning and understanding the value chain before investing are his key traits that I believe has led to a great value proposition for Terminal which is ready to disrupt the traditional and legacy platforms that manage the yard.

Full interview here.

Subscribe for free to receive new posts and support my work.

Nikhil Varshney: Hello and welcome to Silk Road Nexus Conversations. I am at Manifest 2026 in Vegas over here and I’m privileged to have Darin with me who is the CEO and co-founder of Terminal Industries based out of Austin. Usually when we talk about supply chain, we talk about a lot of tech that goes into warehouses, that goes into making sure customers are buying the right products, but there is a hidden aspect of supply chain which is very crucial and it’s the yard.

Nikhil Varshney: Yard is a place where all the freight come in, they stop, they unload, and they finally leave. But that portion of managing that yard is a very big pain point and a problem in the supply chain that leads to lot of disruptions, that lead to lot of delays, and a lot of wrong shipments as well. Terminal has been founded to kind of solve that particular problem. But before I jump into that, Darin, why don’t you quickly introduce yourself.

Darin Brannan: So first off, I’m delighted to be here. Thank you for having us on this podcast. My background is I’m a transplant. I live in Austin, Texas, but I’ve been transplanted from the Bay Area where I spent a third of my career as a VC and then lept across the table to try my hand at entrepreneurship. So I became a tech entrepreneur and now I guess a tech business builder and a terminal is my fifth startup slash business to build. Staggering.

Nikhil Varshney: Yes, so I love the business of starting businesses that lead to market leadership that drives real outcomes for customers. Awesome. And like fifth one, what happened to the previous four? How are they doing?

Darin Brannan: Yeah, so the prior four, three of the four had really nice unicorn IPOs. OK, so fantastic returns for all employees and investors. And the last one, we were close to that trajectory, but COVID was not our friend and hurt some of our growth, but the company is still doing well and maintained a market leadership in its category. And then finally, terminal.

Nikhil Varshney: So I’m assuming when you’re talking about yard, you probably have spent some time studying what the real problems were in yard. And I know I’ve worked in supply chain for last 10 years. I’ve worked with Wayfair and we have a big problem with yard management as well. But at the end of the day, I want to hear from you. What have you seen are the biggest problem in yardage?

Darin Brannan: Yeah, one of the reasons I was attracted to build another business with some of the brightest investors and strategic investors here is one, they did fantastic due diligence, competitive analysis and landscape on the market. Then I did my own prior VC and entrepreneur landscaping and market analysis. And it confirmed that the yard is one of the last and it’s large. It’s one third of the supply chain territory is one of the most unmodernized nodes in all of supply chain.

And there’s reasons for that. It’s unstructured. It’s highly variable. SAS tech has been hard to ROI in that category relative to the warehouse and TMS space. And so it’s been largely overlooked as a node of high, can just almost congestion in a sense that it’s just a cost of doing business. They’ve just dealt with it with processes and a patchwork of, of technology. And it’s now finally reached the point where you have adoption of WMS and TMS in the 70 to 80 % of tech that ROIs and only 25 % of the art. And it’s now causing, the investments in TMS and WMS to be bit diluted because it’s chipping away at the efficiency and throughput when you have congestion in an unmodernized environment.

Total TAM and Available Value - Outside of Interview (Added by SRN)

And so to me, that was a large growing and underserved market that had a patchwork of tech companies that were sort of 1.0, 2.0 tech that outpriced themselves. There’s a number of reasons why there wasn’t adoption. And then when I teamed up with ABC and their strategic investors, Rider, NFI, Lineage, some of the biggest players in the industry, they all affirmed that this is a cluster of pain points that they’d love to have a single full stack modern AI agentic computer vision, if possible, build that company so that they have a really sustainable platform that’s modular and configurable for the art that works. And that’s a long answer to you.

Nikhil Varshney: I mean, it’s fantastic because I mean, what it leads to is like, if you can help us understand the size of the problem, because I mean, when we look at logistics, it’s a very big industry and it has a lot of touch points. And the value chain of logistics include yard in it because at the end of the day, trucks have to come, trucks have to go. So there is a value chain associated with it. But what is the size of the problem if we were to kind of like ballpark it in certain figures?

Darin Brannan: Yeah. And so I’ll bring it down to the street and then I’ll bubble it up to the macro view. But the thing that struck me is the biggest problem in the industry relative to the yard is that because it’s under modernized, the tech doesn’t work there the way it should in terms of real orchestration, real movement, real efficiency, real velocity, visibility. It’s created nearly a 30 to 40 % under utilization of truck time. Normally you have 11 hours of truck time. It only gets on average six hours. So 40 % and the majority of that is buffered in the art because the scheduling, the... The stage is the minute you’re off schedule by a few minutes, it just creates congestion because everything’s manual written. So that’s the main problem. And then that in terms of the market spend, it’s in a four billion range is the market size for the problem. Yeah, that’s the problem. And then we add, we’re adding security fraud because we have cameras there that can detect all sorts of things. And then we do damage detection. And now you’ve just quadrupled that time.

Nikhil Varshney: Wow. So basically we are looking at 12 to 15 billion dollar of a market, which is underserved at this point in time and is a very crucial aspect of the entire supply chain.

Darin Brannan: Right. Within just the art notes, just the art. And that’s inland. And then we have opportunities else international and for tomorrow. Obviously.

Nikhil Varshney: But just kind of double clicking on that one number that you mentioned that the truck utilization is 11 hours and that kind of gets cut down by 40 to 35 percent. What does that 11 hour mean? Like, is it the drivable duration of a truck. And then basically we are only driving it for six hours. So we are losing four to five hours of drive time of the truck.

Darin Brannan: Just because it’s in the yard. that manifests in trucks waiting outside of the gates for four or five hours, three hours, waiting inside the gate, waiting to be slotted. And that’s just the way the business has been done for decades. And again, because there’s fewer people in the yard, So the warehouse may have 20, 50, 80 people, more structured workflows data. That was easier for tech companies to essentially attack that industry. But the strong ROI case, same for over the road. But the yard, highly unstructured. So many different moving assets, moving actors, so many different movements that traditional software was hard to build with kind of clumsy IoT devices, RFID, GPS, when you add it all up, it just becomes a cost prohibitive tech stack that’s 100, 150,000 a year. And that doesn’t work for that node. So they’ve just left it as largely, 80 % of it is not digitized, it’s not optimized, it’s certainly not automated. And that’s our business is to come in and modernize it with those three things in mind with our tech stack.

Nikhil Varshney: Makes sense. And then, when we look at the problematic part, is it also the paperwork that happens in the yard? Is that a problematic part or are you saying that would be outside of the scope of like what yard management actually includes?

Darin Brannan: So our platform, again, we’re out to, we’re a very mission driven company. Our mission is to make goods flow better, faster, cheaper, cleaner by modernizing the yard, first digitizing it, which was that paper element. Then that allows us to optimize the yard. optimize the gate, yard and dock, and then that leads to the autonomous yard of the future. And we’re doing that with the only agentic AI platform in the industry in the world today, combined with best in class computer vision that we’ve built that has the highest detection, highest accuracy rates.

So when you combine those two, and you’re able to get the full benefits, almost irrefutable benefits of AI and agentic, which is one third, one half the price. 10 times the capability, one third the deployment time, three times the ease of use. That’s a game changer. That model, if it’s deterministic, will work in the yard to help reduce the paper flow. And at the gate, that’s where we see the biggest problems. You have manual check-in with paper, and we can eliminate that within two seconds. We capture all the ID information, pull it up on their web app, and it just confirms it. Or it can be a fully autonomous gate.

And so now you have there was anywhere we were finding anywhere from 20 to 40 % accuracy issues at the gate that they just funged through the art. That’s just how they did things. Like it created chaos and then they’d go try and solve for the chaos. Now we bring that accuracy rate to nearly 100%. So that’s one main efficiency and velocity. And the other is just the throughput. We, on average, when you set up our cameras, if it’s hybrid, we’ll do a check in in 34 seconds versus. two, five, we didn’t see 15 to 20 million check-ins.

Nikhil Varshney: Got it. And then just to double click on that number where you say that there is a gap in efficiency at the check-in itself, like 20 to 30%. What is the inefficiency over here? Is it the gatekeeper is not able to identify the trucks properly? It’s taking a lot of time or the inaccuracies are in the form of wrong data?

Darin Brannan: It’s so it’s it’s it’s both. It depends on the type of yard and the type of clients they have in that yard or the manufacturer. But oftentimes the universal workflows where it’s not well digitized and sometimes there’s a little bit of digitization there, but it’s just a record keeping digital record keeping. It’s not doing the door. The database correlation lookups and confirmation that that data is right, that this is the right truck with the right load. at the right time going to the right spot in the yard.

typically happens, most of that happens at the gate where captioning information, running intelligence on it, and then reaffirming that the accuracy is complete and that the asset can now be tracked with additional cameras in the yard for full live real-time visibility of that asset. So no longer are they running around with papers are stickies trying to figure out, especially in moderate to sophisticated yards. And there’s a lot of paper, have yard checkers that are constantly tracking what’s been moved where that’s no longer needed with our sort of simple deploy easy to run tech that was the most advanced.

Nikhil Varshney: How do you see the adoption going for yard management today? Like our company is open to that. it like? still a back burner issue and people are not talking about it. What’s the market?

Darin Brannan: It’s an excellent, it’s an excellent question. That is the question I always ask before I get started in building the business is why, why now and why us? And if I don’t have positive affirmation of a short checklist on each of those, the timing could be off. And my view is the success of startups or business building is one third market timing. especially when you’re selling into mid to enterprise. So that’s a backdrop.

What I found here when I joined up with this, the investors and strategic investors is, A, we had the signals from the strategic investors, the riders, et cetera, that their view was, post-COVID, we’ve made our investments in WMS and we’ve cycled through the things that don’t work and work and now we’re feeling comfortable with that. We’d love to revisit the yard. And we’ve heard there’s new smart yard tech and they did that over three or four years ago, but the tech just wasn’t scalable or effective. So they felt like that got burnt.

OK. So AI is now a catalyst for them to revisit to see if there’s new tech vendors and sure enough, terminals, the dominant space is. So I’d say, sorry, I just want to finish that thread. Cause this gets to your point is I’d say a couple of years ago, maybe six, six to 10 % of the market was in market looking for yard tech. OK. My sense, and having done this multiple times, it’s somewhere in the 20 to 50 % or 60%. It’s a big range. But it’s quadrupled in terms of people in discovery to look at smart yard tech. Because they don’t want it to be, a patchwork of point solutions that’s bringing down their investments here, and then congestion, accuracy, all those things, efficiency, velocity. They know with the right data and the right tech, they can optimize that part of the node.

Nikhil Varshney: Got it. who is the customer in yard management? Because yard management could have third party carriers, would have their own fleet carriers. And then there obviously is the company that owns the yard. So who is the customer and where the tech goes at you? So if you can just break it down at multiple customer points and then explain like how the technology is working across different players.

Darin Brannan: Great question. So we break up our target market customer into two segments, three PLs, mid and large and shippers. Mid, large and large mean small, mid, large enterprise and mid market. So those are two massive segments. The shippers and three PLs. And on the three PL side, it’s a partnership where they’re looking to provide as much value in their tech and service stack. And they already You know, have a WMS or a TMS investment or a couple of vendors selected. And the YMS has been again less than 25 % adoption. It’s been a real challenge for them. So they’re also looking for new solutions.

And so we partner up with them. That’s the rider. That’s NFI to help them down select and standardize on one platform that they can that they know will bring 20 to 30 % efficiency, 20 to 30 % throughput when fully deployed. Those are meaningful numbers that drop to the their bottom line when they sell for that. And it makes their customers more loyal and sticky. If they have that full stack pulled together with YARF, and again, only 20 % of the market has it, and even half of those are dissatisfied. And then it helps them get new business as well. So when they’re in an RFP, we’ve been brought in by 3PLs to RFPs, it helps complete the story. we have the full stack digitization, including the yard with the most modeling yard tech stack, which is where a lot of that congestion and pain can happen today.

So that’s 3PL. Then shippers, for them, it’s all about getting their right product to the consumer at the right time and us helping them thrive by getting that faster, cheaper, cleaner by solving that yard four hour buffer that they have. And you could even look at the real estate in their environments is the ability to reduce them. Because you no longer need staging if we’re bringing the truck. If there’s a tight synchronization between the driver, even if there’s delays, usually that’s where the cognitive load falls apart. When you have delay upon the delay, then they’re stacking up.

With our system, it determines delays even two days in advance where there’s going to be condensed share. And so it’ll do the right slotting and optimization for the right deck. Doc, it’ll notify the unload load and then a fast pass for the truck driver on the way out. 10 second checkout, 30 second check in, 10 second checkout, full synchronization. That’s profound for shippers if done right.

Nikhil Varshney: But one of the problems that I see, and I just pick up on your last point over here, is that the quick check ins and quick check outs happening within seconds. But the bigger problem that I see in the yard is that there is no space in the yard. So even if you can do a quick check-in, the technology is not solving for that real issue, which is there is no space. I can’t park another truck in the yard. So is that going to still be a drawback in terms of adoption or the gains from the technology? Or do you see from your optimization software that the two can be combined together, where now you have an optimal slot creation? where you know that the truck would come in at 10 AM and there is one slot waiting for that truck, versus the truck came in and there are no slots.

Darin Brannan: Yes. So, you know, it’s case dependent, but overall we see, again, deployed properly where we configure workflows for that yard. And again, agentic is different than SAS. So you’re building ontologies around recalls. I don’t want to geek out too hard on this, but actors, assets, moves, and dynamic training. And so an actor could be the gate guard, the yard checker, the dispatcher. There’s things they do and things that happen to them and there’s movements with them.

So we create workflows, all their workflows plus all their potential attributes and we do that for assets as well. That’s trailer, chassis, et cetera. And same for dispatcher. And so what that allows for the on the ground movement is we can make sure that the scheduling happens, driver experience scheduling happens properly before they arrive. Then when they arrive, can simply, again, the velocity of the gate check in through a web app on their phone, it’ll notify them. just in this conversation will say, who are you? You can literally say I’m Chuck, which company? Rider, okay, we’ve correlated. Take a picture of your e-ball or your license. And it then says, doc four was your doc, but it’s now congested. We’ve slogged you here and we notified the pickup. Stay in your unit.

Yeah, and they’re prepared for you because they knew exactly when you were arriving and then exactly what slot and and so we unload load immediately and they’re out. That’ll reduce the that’ll optimize the archimedes to where, you our view is if done well, you’ll need very little staging. They are right because you’re you’re you’re just fully optimizing the warehouse. up the dynamic at the point. It’s it’s auto dynamic. It does auto decision making. That’s that’s the. That’s a promise of AI and agentic is once you’ve built that and it’s learned quickly, it is you’re leaning, you’re leaning towards the autonomous yard of the future kind of lights out of the future because of autonomous decision making and the way we set it up.

There’s a lot of AI that doesn’t work in enterprise. So we’ve built it from the ground up to be deterministic, which is a lot different than probabilistic product. So many weird things that can happen. Terministic is within our rules. And it ensures that that happens.

Nikhil Varshney: And then these rules, how scalable are these rules? Because every yard is different. Every company has a different structure. So when you go from, if you’re building deterministic rules, then you are kind of saying, I will have 100 kinds of different rules based on the infrastructure requirements. is that a kind of a...

Darin Brannan: That gets into a little bit of our secret sauce, but I’ll give you a hint on it. You start to build a library of these rules. And even though every yard is different and there’s different carrier shippers, cetera, you can more that leads to the instead of months of pilots and POCs, it leads to days of setup with value. You’ve got a robust library of assets, actors, dynamic moves and optimizations. that’all for it.

And that’s why it’s our belief that SAS, traditional SAS, has had a really tough, like there’s been a massive adoption problem of tech in the yard. It’s because they end up with config tables upon config tables and then brittle API stacks and you need event-based integration. it’s an entirely kind of revolutionary platform. And we’re now in a state where it’s proven out and we’re forward.

Nikhil Varshney: So when you say that you have a library that’s basically a set set of structures that probably are unchanging across multiple dynamics reusable, reusable. And then you have definitely tentacles coming out for different industries in different segments that you can now on top of that library build, which kind of reduces your deployment.

Darin Brannan: Yes. And, and, and by the way, that, that is rocket science. This is not easy to build. You build, it all together. You need AI, agentic engineers. There’s a lot of, of a heavy lift to get in here. Yeah. On the AI computer vision alone, that, takes PhDs, machine learning data scientists. to make the computer vision, which is the sort of powered by elements. That’s the most effective cost efficient way to extract the data and then track the data and then use that data and enrich the data in that deterministic model. And so we feel like we have five or six moats with that because of how hard it’s been, but we’re now set up, I think in the most sustainable tech platform the world has ever seen. Like I can’t see another transformative tech beyond this state. For anything, if you can reach an agentic AI platform state, I just don’t know what would be next beyond that.

Nikhil Varshney: True, that’s true. Because it’s autonomous, it just creates autonomous decision making, which is the ultimate goal. One thing that I want to double click on and learn more about is this computer vision. Can you help me understand what problems did you see that required the computer vision solution? And how is that working? I know we talked a little bit about check-ins, check-outs, slotting, which is optimization, but how does computer vision fit into it? Like, what is it doing?

Darin Brannan: Yeah. So the industry had kind of waves 1.0 OCR computer vision several years ago. And that claimed to do some really interesting work, but it wasn’t going to replace these GPS and RFID trackers. And so it didn’t get much traction. And the advent of real AI combined with that is a game changer. You go beyond OCR, where you’re tracking container IDs, you’re solving for occlusions, which is inclement weather, it’s covered over IDs. Like you can solve for those through AI extrapolation and annotation. it just leveled up the possibility of computer vision.

And the reason why that is becoming a pretty exciting wave for the supply chain is because they’re kind of reaching a point of, I’d say, a cluster of pain points around GPS RFID. that’s part of how the yard breaks is if they don’t scan that right, if it doesn’t have the right RFID. And then pretty soon you’ve got tags everywhere. And you’re just involved like three more manual workflows associated with that versus a camera properly. And again, lighting position, inclement weather, annotation.

There’s just a lot of complexity to build multiple model pipelines to then be able to instantaneously convert that data into meaningful insights and tracking. And so within typically less than two seconds, we’ve scanned the type of power unit. trailer chassis ran all the IDs done some database correlation seamlessly integrated with TMS WMS. We have a really good sense of who this is, what they have, where they should go, how quickly we can turn this around.

And we’ve taken just to give you a sense of that. We’ve taken environment that had GPS RFID and we’ve done pilots on this where they had five to 15 minute check-ins that then created an hour 47 minute line on three days in particular. And we were able to get that from gate to dock to unload in eight minutes through our full system. That’s 34 second check in two seconds to capture information. The guard, they kept the gate guard and they can process the correlation, make sure it’s all extra accurate. And then it slots in at the right place and notifies the right people and manages all that carrier appointment scheduling.

Nikhil Varshney: Makes sense. And now I think this is a good segue for me to kind of talk a little bit about the results, right? I mean, we talked about the problem, the size of the problem that we are looking at dollar terms, was like close to 12 to 15 billion dollars. And you’re saying 40 % of the time is unutilized for the trucks. When you have worked with these customers, how much value have they regained from that? And where has been the biggest value capture? And what has been a surprise value capture for you that you realize, oh, we didn’t think that this particular process would result in such high value capture.

Value Capture chart - Outside of Interview (Added by SRN)

Darin Brannan: Excellent question. I’ve been selling enterprise platforms for nearly my whole career and one of the key tenants to any strategy of selling a platform in that industry is tech is great. Even agentic AI, boy, that’s great to check that box so you have sustainable. But really all that they care about is cost benefit. And that cost benefit is where are the ROIs and how quickly can you prove those and are they real?

And so we start with that as our DNA. And so I call it kind of the three V’s at each category, the gate, the yard, and the dock. Are we providing velocity, visibility, which is efficiency, and then value at each of those areas? And can we ROI each of those? Because they can be modularized in enterprise. They may have made some investments, but they want to start. So that has to ROI. It has to cost benefit. So we lead with cost benefit to value. And we’re typically, because of the the benefit of all the hard work investment we made in AI, we’re seeing ROIs inside of 12 months and that’s three to six times our ROI because again, we’re able to come in lot less expensive, 10 times the capabilities, less time to deliver and really easy to use. So a great question. It’s our most important question.

Nikhil Varshney: Okay. And then one part of that question was like, what has been the most surprising element that you thought that would not provide this value, but has created a surprise value element for you?

Darin Brannan: Yeah, so we went in with yard operations in mind, and then we were quickly asked for fraud detection. Can your cameras do face recognition? Can they do bad actives? Can they start to correlate all the information and determine double brokerage bookings? Can you get to the massive fraud that’s building up? And our answer was, let us go experiment and figure it out. And now we’re coming back with that full module suite, which our customers and prospects are excited to see and deploy. And the other was, oh, we also have not only security, but perimeter security.

And then there’s a safety element that happens. People get hurt in the yard. Can you do predictive alerting? Can you do perimeter security all on that camera in addition to yard operations? And it turns out we can do that as well. That was a great surprise. So they’re more delighted. And then last is, as trucks come in, leave, and user owned it, can you do high correlation, high accuracy on data on the The truck damage detection.

Nikhil Varshney: Truck damage detection.

Darin Brannan: Yeah, so there’s damage that happens in the yard and there’s different types of customers that that matters more. OK. Where they might have rental areas and maintenance and so on. And we now have run that through our models and built models around that so we can do damage detection. That again, that’s what kind of ballooned our TAM. We knew that those were adjacencies, now they’re real. on the yard, and then now you have like multiple alternative. Yeah, so security and damage detection, those are massive market problems that nobody has been able to solve. And I think we’ve got the secret sauce to do.

Nikhil Varshney: Awesome. And I think kind of moving from this value more towards yard management to yard operations. And I know Terminal has a larger suite of solutions and the time is not just associated with trucking yards. There are ports, there are train yards. How do you see the overall correlation and how big that problem becomes when you add multiple other segments of the industry to this?

Darin Brannan: Yeah, that’s a great call out. We looked at ports and rails because our technology can address all yards. There’s a million and a half yards in the US, about 55 that are applicable in the 25,000 in the inland. We hadn’t focused heavily on the ports, on the rails, but we do view that as our next step opportunity because they made massive investments in tech. They have their own terminal operating system and they did some initial 1.0 computer vision. so now that we’ve been out in the market for a while, we’re now being pulled into those conversations and solving for some point solutions.

And then we’ll build out more of that yard operation tailored for their environment, which is different than in line yards to a degree. So that’ll be a separate segment because a separate adjacency segment or super excited and then international. Yeah, because we have, again, the best computer vision that we can train quickly. We think we have an advantage to. It will satisfy multinational companies that have international operations.

Nikhil Varshney: Makes sense. Moving from here, I would like to talk a little bit more about Terminal as a company. How big is the company now? How many customers, if you want to share that, how many customers do you have? And what are some of the greater learnings that you had with Terminal, which you didn’t have with other startups?

Darin Brannan: Great question. I don’t typically provide vitals about the company because competitors, but I appreciate the question. But I will say that I think we’re the largest and we have the best in class team for what we’re doing. I haven’t seen any competitor that comes close to having this kind of dominant product with the best in class team. Of course, I’m wildly biased on those two things. And then in terms of. I forgot the last question.

Nikhil Varshney: The learning that you had, which you didn’t have.

Darin Brannan: Yeah, there’s it’s it. There are a lot of universal truths to building a business. I break it into four stages. And so as soon as I got together with this team, I figure out what stage we’re at. And there’s a list of must do things. And then there’s a list of. Idiotic things not to do how you stay is how I call it. And so I think if if you use that playbook, you have a better probability of getting to each stage towards market leadership.

So all that to say. I try to take out the chaos and provide some predictability and performance and the way we would go about the market. But there’s always surprises. There’s always market hold, stretch, some ideas around pivot. And so far, the only surprise, I guess, is how quickly. I knew the market was going to start to go into discovery for YARC, but I’m finding that they’re moving quickly. I think they’re doing more pilots with AI native companies and they’re finding lower cost, greater benefit. And so we’re starting to see that spike where the industry wants to kick the tires with us.

They want to see, you know, what do we have and is this my long-term two to five year solution? So I don’t have to rip and replace or deal with, you know, point solutions. That that’s happening faster than I thought in this kind of heavy industry. So that’s a pleasant surprise.

Nikhil Varshney: I mean, no negative surprises yet. Just taking from that point, right? I mean, I look at I have been through two eras only. So internet era and the AI era, if I can call them. The internet era basically gave us unlimited distribution and the tax or the cost was writing software. in AI, distribution was already there with internet. Writing code is almost free if you can do good wipe coding. What becomes the most challenging part in your opinion? Because if someone can write wipe code, a terminal like application, let’s say, what sets up our terminal in modes that no one else can copy?

Darin Brannan: Excellent framing. Excellent question again. You know, I look at it on a couple of fronts. You’re right. AI, even since I’ve been building this business, it’s becoming less expensive to code. And our engineers are all over over that. And that’s why we’re coming in as AI natives, you know, half the industry cost or less. So we’re riding that wave well. You know, I think if you focus on mid to enterprise customers, they have a lot of legacy and inertia and they’re okay to set course for a 10 year modernization.

And so the way in which you design your business, enterprise, modern tech stack, the data layers are correct, agentic, the security, the scalability, the intercalable modularity and configurability. that’s not easy to vibe code straight out of the box. Like that in itself, that modernized enterprise tech stack is a moat.

The other is I brought together operators and technologists. So the fusion of domain experts and technologists make sure that we’re speaking a language, we’re on the yard with them, we’re building the workflows that are tightly integrated to the way in which they’re going to modernize and scale. That’s a relationship business. That’s a DNA verticalization business. that’s hard for any Vibe coders to pop up and say, look, we have an AI. And that all comes with having worked with enterprises for years. they demand high quality, they demand intergable, modular, configurable, and they demand a cost benefit in ROI. And so the technology is kind of a means to get to all that. if you can, like computer vision, I know that’s a moat. That was millions of dollars to get right. There is no hacking through that today. And the rest is is the way in which you sell, service, deliver and support is its own IP.

Yeah. And then we’ve got hardware, services and AI. That’s kind of the future of sustainable startups in my opinion, is you can bring those together. That gives you moats.

Nikhil Varshney: Yeah. And I think that’s a very valid point. And I know I’m going to make a generic statement not specific to supply chain over here. Over the last couple of weeks, there have been a lot of noise about SaaS being dead. And I think my mind always goes to the direction that, you know, maybe the cost of building the software has gone down or probably will become zero over a period of time. But what you just said, like, you know, services, infrastructure, horizontal and vertical integrations that big tech companies and enterprises have provided, that is just hard to replicate. Building software is just one part of the job, which is a smaller part of the job, because it’s a process you want to automate. But then how you want to use that automated process across multiple other applications is where the real differentiation lies. And that’s where like big companies have survived for so long.

Darin Brannan: Yeah. And that’s well said, but even those big enterprise folks, they’re in for a dogfight for companies like us that get all of that, but then have way better tech transformation platform. They’re going to have a really tough time competing with that platform unless they throw out their tech and start over. Then we’ve got a headset. That’s my bias opinion.

The other piece that I almost forgot to mention, which is also part of IP. The way in which you deliver matters a lot. And part of that delivery is you can’t rely on their change management. Otherwise you’re stuck for 10 years. So we built a change management DNA and team. So we have our business model is we’re selling you this and we’re selling you change management. Guess what? Change management is free. We include it. We’re going to train you. going to send it. We will transition you. We do all that where we bring the donuts and pizza for your team to get them excited. And then we do hyper care after that to make sure that they have now transitioned into eliminating workflows and making their day easier and that they see like this is the good value prop and it make their life much easier.

Nikhil Varshney: Makes sense. And I think that was one part that I actually forgot to ask you about change management because you’re working with employees that sometimes could be hard to adopt new technologies and that’s where if the adoption is low, irrespective of how great the technology is, you won’t get the real benefits coming out of it. So what you’re saying is Terminal has boots on the ground, working hand in hand with those operators to help them understand how to use the technology in the best possible fashion.

Darin Brannan: Absolutely. And just to amplify that point, part of the massive adoption problem in tech is that many of these YMSs their first or second site, but they couldn’t get beyond that because they couldn’t get the adoption because they couldn’t get past the 50 year process as in trans processes because they didn’t hyper focus on change management as well as the tech was old and it was just observability tech versus moving and orchestration tech, is it has to provide 10 to 20 % increase on velocity throughput and efficiency. We’re in the throughput efficiency game. If it doesn’t do that, plus if we don’t manage the change management, And make it super easy to operate use and low, low lift on tech. You’re just not going to get an option and that’s what’s happened. So we’re trying to solve for all those prior adoption challenges.

Nikhil Varshney: This is one of the last questions that I’ll ask you. Thank you so much. But one question that I would end with is. At what point when you segment your customers. Do you realize that. below this threshold, it’s very difficult to convince the customer to buy a product because they’re too focused on building their brand or if they’re a shipper, they are too focused on kind of increasing their revenue and sales and we should not focus on that. So do you have a threshold above which you try to approach the customers?

Darin Brannan: Yeah, so initially as a startup, early stage, you refine your ideal customer profile so that you’re not doing too much customer work and slows down your initial momentum. So initially we had an ICP filter where below 20 moves a day in a yard, below 250,000 square feet in a warehouse. But we’ve since matured our tech to where we have low cost solution with or without cameras.

And part of that was by design, it was an easy lift once we built everything. But then we had our big three PLC. What are we going to do with all these really many yards that we have? They’re all on paper. I’d love to just, is there a low cost digitization way? We have a very low cost, very easy to deploy way to digitize those small yards. So now we have the full spectrum. I couldn’t be more happy about it.

Nikhil Varshney: So now we can come in with like no ICP restrictions. And that is fantastic because now you are also capturing the low hanging fruits, which are the cause of big disruptions as the industry and the organizations evolve. So I think congratulations on that. And thank you for having and sharing these thoughts with me.

Darin Brannan: Thank you. Really appreciate it. Great work.

Thanks for reading THE SILK ROAD NEXUS! Subscribe for free to receive new posts and support my work.

Thanks for reading THE SILK ROAD NEXUS! This post is public so feel free to share it.

What You’ll Learn in This Episode

* How Gap is embedding AI across discovery, personalization, and fulfillment to reimagine the customer experience.

* How AI conversations are becoming the new customer entry point, replacing traditional search and social channels.

* What agentic AI means for product management, customer trust, and operational speed.

* How retailers can balance AI enablement and human judgment to protect margins while improving customer experience.

Subscribe for free to receive new posts and support my work.

The Next Retail Revolution: From Search to Conversation

In my discussion, Hareesh captured the central shift happening in retail today: customer journeys are moving away from typed searches and websites toward agentic, conversational interfaces.

The store, the website, and even the app are no longer the first touchpoint. The first question is now being asked to an AI.

For decades, retail discovery followed a familiar path: consumers searched on Google, clicked through links, and compared products. Then Amazon shifted discovery inside its own platform, becoming the default starting point for millions of households. Now, as conversational systems like ChatGPT and Perplexity gain adoption, discovery itself is fragmenting into dialogue.

Gap’s AI Strategy: Enable, Optimize, Reinvent

Hareesh explained that Gap’s AI roadmap rests on three pillars:

* Enable: Give employees hands-on AI access and training to improve productivity across merchandising, design, and logistics.

* Optimize: Use AI to streamline legacy workflows, such as demand forecasting, HR, and supply chain visibility.

* Reinvent: Rebuild key retail experiences—from product design to personalization—with AI at the core.

“AI is not something you sprinkle across functions,” Hareesh said. “It has to be built into every customer touchpoint—from brand discovery to checkout.”

This inside-out approach contrasts with how many brands are experimenting externally through plug-ins or third-party agents. Gap is treating AI as organizational DNA, not a feature.

Subscribe for free to receive new posts and support my work.

Personalization and Fit: The New Frontier of Retail AI

Fashion retail is a returns-heavy business, and Hareesh believes that’s where AI can create measurable impact. Gap is using machine learning models to build personalized storefronts tailored to each shopper’s preferences, location, and current fashion trends. AI doesn’t just personalize, it anticipates.

He described two innovations:

* AI-curated catalogs that reflect real-time social trends and local climates.

* Fit intelligence powered by product dimension data and a few basic customer inputs like height, weight, and age.

Together, they are narrowing the gap between expectation and experience. Virtual try-ons, powered by improved 3D modeling, bring confidence to the customer while reducing reverse logistics costs.

“The goal is simple,” Hareesh said. “What customers see on the screen should be what they feel when they wear it.”

Checkout and Cart Friction

AI is also helping Gap address one of retail’s most persistent pain points: cart abandonment. According to Hareesh, most drop-offs occur when promotions don’t calculate correctly, loyalty rewards fail to apply, or transactions are falsely declined for fraud. AI-driven checkout systems now help reconcile math instantly and reduce false positives in fraud detection.

“If customers make it to checkout, the math must match the promise,” Hareesh noted.

Reducing friction here improves both trust and lifetime value—a combination that matters more than incremental traffic.

AI and the Supply Chain: Where Retail Economics Are Won

Behind every digital experience lies a physical one. Hareesh emphasized that inventory positioning is one of AI’s most practical applications. Gap uses predictive models to ensure the right products and sizes are available in the right locations, whether that’s college towns or suburban malls. Smarter inventory placement not only prevents lost sales but also cuts markdown costs, directly improving margins.

Subscribe for free to receive new posts and support my work.

The Bigger Picture: AI Conversations Everywhere

While our conversation centered on Gap, similar experiments are emerging across the industry. Walmart’s decision to enable product purchases directly through ChatGPT signals a broader pattern: retailers are preparing for a world where customers may never visit their websites at all. Instead, discovery, comparison, and checkout will unfold inside AI-powered conversations.

Amazon, Shopify, and other players are watching closely. Some are integrating with AI platforms; others are building their own conversational agents. But the underlying pattern remains clear: AI intent is the new retail currency. Whoever captures that moment of inquiry—“What should I buy?”—will define the next era of commerce.

The Future: Building Trust Inside the Conversation

For Hareesh, the long-term challenge is not just deploying AI, but ensuring that it builds trust with customers. Each algorithmic recommendation or personalized fit suggestion must reinforce reliability, not risk it. He believes the winning strategy will balance two forces:

* Speed: Using AI to shorten product cycles and deliver faster personalization.

* Reliability: Ensuring pricing, loyalty, and fit accuracy remain flawless.

Both, he said, are essential to create lasting brand loyalty in an AI-driven ecosystem.

Key Takeaways

* AI is shifting from tool to teammate. Gap’s enable-optimize-reinvent model is an operational blueprint for responsible adoption.

* Returns, fraud, and friction are data problems solvable through continuous learning loops.

* Conversational entry points like ChatGPT are not competitors but catalysts—forcing retailers to think omnichannel at the intent level.

* Trust and transparency will define the winners of this new retail cycle.

Closing Thought

Retail’s evolution is no longer about better websites or smarter ads. It’s about creating confidence inside a conversation. As Hareesh and Gap demonstrate, the next generation of commerce leaders won’t just predict what customers want—they’ll understand how customers ask.

In a world where buying starts with a question, success belongs to those who answer it best.

Thanks for reading THE SILK ROAD NEXUS! Subscribe for free to receive new posts and support my work.

Which is why today’s conversation with Shyam Krishna, Head of Integrations at Roboteon, is super important. In this conversation I decode the nuances of the intersection of AI, logistics and automation.

Subscribe for free to receive new posts and support my work.

Three Strategic Takeaways

* Multi-Agent Orchestration Systems Are the New Software Architecture for AI enabled physical operationsWarehouse automation will not be led by hardware. It will be driven by multi-agent software systems where specialized AI agents collaborate to manage demand, labor, inventory, fleet movement, and exceptions in real time.

* Warehouse Performance Will Be Measured by Flow, Not TasksTraditional metrics such as pick rate or dock-to-stock measure human efficiency in isolation. The next era of performance measurement will be about how smoothly humans and machines move together as one coordinated system.

* Digital Twins and AI Agents Will Form the Operational FlywheelWarehouses will simulate before they execute. Digital twins, paired with AI agents, will predict outcomes, resolve constraints, and feed improvements back into the system continuously.

1. Multi-Agent Orchestration Systems: The Software Architecture for AI Enabled Physical Operations

A MAO platform is the software layer that integrates with and manages the work of a heterogeneous fleet of robots.

The next generation of warehouse intelligence is not being built as a single platform. It is emerging as a multi-agent architecture. Each agent is optimized for a specific function. One agent forecasts demand volatility using marketing signals and historical order behavior. Another evaluates labor capacity against service level requirements. A fleet orchestration agent coordinates AMRs and AGVs across vendors and regions. An exception agent monitors congestion and intervenes before it triggers an SLA failure.

These agents do not operate in isolation. They communicate and negotiate with one another in real time. This enables the warehouse to function as a living network that continuously senses its environment, reasons about priorities, and adjusts accordingly.

This is precisely what Gartner identifies as one of the most transformative forces in physical operations. In its 2024 Supply Chain Technology report, Gartner states:

“Multi-agent orchestration platforms will become foundational to the next era of warehouse software, enabling interoperability across heterogeneous fleets and human workforces. Without MAOs, warehouses will never fully realize the ROI of robotics and automation investments.”

Further, As robot fleets grow, simple point-to-point API integration will not be enough. Companies will instead need an accelerated integration and orchestration capability that can assign work to the right robots or agents based on near-real-time information and the type of activity.

Gartner projects that adoption will shift from early stage experimentation to mainstream deployment by 2027. They also believe by 2028, 80% of warehouses and DCs will be able to deploy the first wave of robotics and/or automation. This signals a fundamental re-architecture of warehouse software, moving from systems that record events to systems that act on intent.

Why MAOs Matter Strategically

* They turn automation into intelligenceRobots perform tasks. Agents make decisions based on cost, service level, and context.

* They unlock interoperabilityMost enterprises now operate multiple robotic fleets across different facilities. MAOs provide the unified brain needed to make them work together.

* They enable perpetual optimizationUnlike static workflows, agents learn from every cycle and improve outcomes without human intervention.

* They power digital twinsMAOs feed data to the digital twin and learn from it in real time, turning simulation into a live decision engine.

It is a shift in how supply chain software will be built and purchased over the next decade. Just as the cloud unbundled infrastructure, MAOs will unbundle warehouse logic and then re-coordinate it at a higher level of intelligence.

Subscribe for free to receive new posts and support my work.

2. Why Flow Efficiency Will Replace Task Metrics

For the past two decades warehouse management has focused on task-based outputs. Lines per hour, units per operator, and on-time shipments were effective in a human-centric model. In a warehouse where humans and machines work together these metrics no longer reveal the true constraint.

Legacy metrics measure performance locally. They do not capture system-wide friction such as waiting for a robot, congestion in a pick zone, or delayed handoffs.

Imagine two operators. One has a high pick rate but routinely waits for a robot to arrive. Another has a slightly lower pick rate but never waits because orchestration ensures perfect timing. The second operator is far more valuable to the business, yet legacy metrics would reward the first.

This is why new flow-based metrics are emerging.

Examples of next generation metrics include:

* Robot dwell time

* Handoff latency between humans and machines

* Congestion events per hour

* End-to-end flow time from task creation to completion

Once warehouses begin optimizing for flow, rather than individual productivity, the return on robotics and AI becomes exponential. The warehouse starts behaving as a coordinated system rather than a set of independent workstations.

3. Digital Twins and AI Agents: From Projects to Flywheel

Digital twins are not visual representations. They are dynamic simulations that reflect the real-time state of the warehouse. They ingest data on inventory, labor, robots, task queues, and congestion points. This allows operators to test scenarios before making operational decisions.

Multi-agent AI systems operate within this twin. Demand agents adjust forecasts. Labor agents propose staffing changes. Fleet agents test routing configurations. These agents evaluate trade-offs and identify the most efficient flow.

This marks the shift from static automation to continuous improvement. The orchestration platform does not simply execute. It learns. It discovers new optimizations over time. The result is a compounding performance curve that traditional warehouse systems were never designed to achieve.

Conclusion

The race in warehouse automation will not be decided by who buys the most robots. It will be determined by who masters orchestration. Hardware can be purchased. Software can be copied. But flow intelligence compounds. It improves every day. It becomes a strategic moat.

Warehouses built around orchestration will unlock:

* Faster scaling of automation across networks

* Lower marginal cost per unit moved

* Higher service resilience during demand shocks

* Greater workforce engagement through intelligent upskilling

The future warehouse is not just automated but also orchestrated.

PS: This episode is an educational deep dive into Multi-Agent Orchestration (MAO) — exploring how humans, systems, and robots work in sync to shape the next era of warehouse intelligence. It’s a longer conversational podcast, by design, aiming to unpack a complex topic that rarely gets discussed beyond surface level. All examples are illustrative, and the views shared by our guest, Shyam Krishna, are his own — offered purely for industry learning and discussion, not as representations of any specific company or client.

Thanks for reading THE SILK ROAD NEXUS! Subscribe for free to receive new posts and support my work.

This week, I spoke with Matt Cohn, Senior Technical Leader at Shopify in the Office of the President, and our conversation shed light on a pivotal shift underway in commerce. We are rapidly moving into an era where AI-driven conversations will replace traditional search as the novel entry point into online shopping. Instead of users typing keywords and browsing pages, they will start with a natural-language question — and the agent will do the work of finding the right product, brand, or merchant.

Shopify sits at the center of this transformation. Historically, its mission has been to empower merchants to reach customers. But the future will require a reversal of that flow: building an ecosystem where customers, through intelligent agents, can easily discover the merchants that best match their intent — even when they are not entirely sure what they are looking for.

Subscribe for free to receive new posts and support my work.

In my essay “Personalization vs Customization” I had defined personalization as:

Personalization, on the other hand, is far more sophisticated. It is the system’s ability to proactively understand a customer’s mood, preferences, and behavioral signals to present a curated experience — ideally better than what the customer could design themselves.

This definition is now ever more important and this conversation breaks into creation this experience of personalization.

Key take away’s from my conversation with Matt:

* The Future Is Conversational CommerceIn the last essay I discussed Walmart and ChatGPT, noting that commerce is shifting from transactional-based browsing to conversational discovery. The core idea is inversion: instead of merchants competing to find customers, conversational AI enables customers to express needs — even when those needs are vague — and have the right merchants and products find them.

* MCP Servers Are Foundational to Building AI-Native CompaniesThe fundamental goal of commerce is to reduce cost and increase conversion. Model Context Protocol (MCP) servers transform Shopify storefronts into intelligent, reasoning systems. They enable contextual understanding, action-taking, and agent-to-platform orchestration, forming the backbone of AI-native commerce infrastructure.

* Agentic AI Will Redefine the User JourneyFeatures like Sidekick represent a shift from “answer engines” to “action engines.” These tools reduce cycle time across marketing, merchandising, and operations by assisting with actions on behalf of users, reshaping the entire commerce flow from search to fulfillment.

Subscribe for free to receive new posts and support my work.

Conversational Search: Perplexity + Shopify as a New Entry Point for Commerce

When consumers begin a purchase journey today, they often start from a vague idea or a problem they are trying to solve. Traditional search engines force users to manually refine queries, click through ads, and navigate marketplaces where discovery is dominated by the biggest advertisers.

Conversational AI flips that model. Instead of browsing through mass-market listings, the customer can ask natural-language questions and be guided toward the exact product that addresses their need, including niche offerings from independent merchants.

Perplexity becomes the first entry point, a discovery layer that precedes channel selection. By connecting Shopify’s structured catalog data to this conversational surface, the partnership delivers three strategic advantages:

* Captures early intent before the buyer decides where to shop, increasing the likelihood that Shopify merchants are included in the initial consideration set.

* Compresses the purchase funnel by using dialogue to clarify needs and present a curated set of relevant options, reducing search friction and abandoned journeys.

* Protects brand equity by ensuring accurate product representation through direct integration, as opposed to marketplace algorithms that commoditize brands.

This is a direct response to the changing economics of advertising. Impression-based models rely on thin signals. Conversational discovery generates thick signals, rewarding brands with strong product data and narrative clarity.

Model Context Protocols (MCP): The New Search and Action Layer

MCP servers turn storefront from a passive website into an intelligent search. It is the way of giving developers a consistent way to build agents that read, reason and act thus enabling an interactive surface for customers. The reason they become super important for Shopify is:

* Platform lock-in - When agents are deeply embedded in core platform functions — catalog, checkout, fulfillment — it dramatically increases switching costs for merchants.

* Network effects - MCP allows developers to build once and deploy AI agents across thousands of stores. This creates a two-sided network: merchants gain access to sophisticated AI capabilities, and developers gain a scalable distribution channel.

* Defensible market position - Shopify is defining the AI orchestration layer of the internet. In this world, buyers do not begin their journey on a marketplace. They begin it in conversation with an agent, and that agent is connected to Shopify via MCP.

Commerce is undergoing a foundational shift—from keyword-driven search and auction-based advertising toward intent-driven conversations mediated by intelligent agents. Shopify’s partnership with Perplexity and its investment in MCP servers are not incremental product updates; they are strategic moves to define the operating system of AI-native commerce.

Perplexity captures the earliest signal of intent. MCP servers convert that intent into intelligent action. And Agentic AI compresses the distance between desire and fulfillment.

Thanks for reading THE SILK ROAD NEXUS! Subscribe for free to receive new posts and support my work.

Thanks for reading THE SILK ROAD NEXUS! This post is public so feel free to share it.