A prototype built with Claude Code still has to survive your ERP | Ep. 7

Episode 7 - A prototype built with Claude Code still has to survive your ERP - Transcript

Matt: Welcome to Data vs. Commerce, where we explore the messy middle between database and doorstep. I’m Matt Johnson.

Floyd: And I’m Floyd Blaikie. Let’s dig in.

Matt: Okay, hey guys, welcome back to Data vs. Commerce. Matt Johnson here, joined with my partner in crime, Ryan Smith. He’s an account executive with Pivotree. We are missing Floyd Blaikie — we miss her a lot, it’s just not the same without you, Floyd. Come back soon. But guys, I hope you’ll hang with us today despite Floyd not being here, because we have some fun things to talk about.

Ryan is fairly new to the Pivotree organization, and he has just been a really amazing teammate. We recently went to the AI for Distributors event in Chicago, an event specific to applying AI in the distribution business. You had distribution executives, leaders in IT, leaders in product data management, leaders in sales and marketing, and supply chain getting together, looking at some of the bleeding-edge technology and tools that are flooding the market in the distribution space.

We learned a lot, and we thought we’d jump on here and talk about it and share some of the things we saw. Even if you’re not in distribution — if you’re a manufacturer, this really does apply to you as well. And even if you’re in retail, I think there are some things to learn about what’s happening in B2B commerce that you can take away as insights for your business.

So Ryan, welcome to Data vs. Commerce. Excited to have you, man. Tell the audience a little bit about you and your journey so far with Pivotree.

Ryan: Thanks, Matt. I appreciate the invite and having me on. Definitely some big boots to fill with the absence of Floyd here, but I’ll do my best. I’ve been at Pivotree for a short tenure, a couple months so far. Really enjoying the people I’ve met on our side of the fence, but really excited about the people I’ve been able to meet and the issues they’re having as they transform their data and their digital space — their tech stacks.

The conference itself was an amazing event, very well put on. Some big players were there. It was awesome to get to know what people were focusing on, but I’ll be honest — it was a little intimidating to hear what they were looking for, because it seemed like everybody wanted that new shiny tool. As somebody who doesn’t sell software alone or have that new shiny tool, it was a check to say, “Hey, what are we going to be able to offer?” Now, I knew we had tons to offer, but when somebody comes in with that headspace of looking for that new shiny tool, steering them away from that for reasons that will actually benefit them is a little more difficult than you’d first think.

Matt: Yeah, exactly. This was the fourth year that Distribution Strategy Group has hosted this event — DSG, as we like to call them. If you want to learn more, just search Distribution Strategy Group. They put on a great event, and it’s been fascinating watching the evolution of adoption in the distribution space. The first year was really just an interesting step into AI — a lot of traditional software companies had some AI features they were starting to roll out.

Today, four years later, we have fully automated, AI-specific applications for everything from order management to product data enrichment, to even bespoke AI-centric ERP solutions. There’s some really amazing evolution happening in the software space. But to your point, Ryan, we don’t sell software. That’s actually one of the things that’s unique about us — we’re software agnostic. So what was our message to these distributors looking to implement AI, wherever they were in their journey? Some were just kicking tires, seeing what’s out there. Some had already done a lot of work and maybe even had people dedicated full-time to AI initiatives. Maybe you could talk through some of the conversations we had at the show.

Ryan: One thing that was really interesting, talking to some of the companies attending — not the exhibitors — was hearing them say, “That young man over at that booth was here last year. He didn’t work for anybody, he was asking everybody a ton of questions, and this year he’s here with an AI tool.” It just shows how quickly these off-the-cuff companies, that may not even have experience in distribution but understand the AI facet of it, can gather information and put together something that resembles a company and a tool attendees can use to make their business more efficient.

But a lot of the conversations — and the message on our part — were with the returnees who’d been there last year and bought a tool. For the vast majority of them, something went wrong in that process. I’d say a large portion of that had to do with the foundation that tool was plugged into not being ready for it — meaning the data wasn’t cleaned and governed the way it should be in order for the AI to do what it needs to do. Whether that’s part of ordering, part of onboarding new clients, part of mergers and acquisitions where two ERPs, two PIMs, two MDM systems are all coming together with duplicate data everywhere.

A lot of the message was, “Hey, why isn’t this working?” And I think it made people nervous about buying the next tool that might help a different part of the company. Say they did something with ordering and e-commerce before, and now they’re looking for something with PIM — they’re a little gun-shy about pulling the trigger. So it comes back to taking a step back and saying, “Maybe I need to look at this from start to finish. When data comes into my company, and when data leaves — that happens before resources and a product leave. Maybe I need to take care of that first.” That was definitely our message: everybody’s jumping to step two. Step one is the foundation, which is your data. Let’s take a look at how this looks inside your company.

Matt: Yeah, it’s so obvious, right? There’s this gold rush toward B2B distribution and manufacturing AI automation products — so much investor money out there for smart ways to take the manual labor out of a very labor-intensive market. So I think a lot of the distributors there were looking for quick wins. Where can I apply AI without necessarily disrupting the way my team does business today — say, in sales — and start to see some lift, get traction with my internal teams, and hopefully create a better experience for my customers?

The message that resonated so well, and why our table was busy and we had a lot of meetings booked, was because I got up there the first day and said, “Everybody in that room is trying to sell you step two.” And to your point, Ryan, step one is the governance, the change management, the security, the enterprise-ready data that makes all of these tools work. I think the misconception out there is that you can plug AI technology on top of a poorly governed, undisciplined data practice and hope it’ll just work around the mess you’ve created over the years.

But really, you have to go back to the foundation. The smart, veteran decision-makers we talked to heard that and said, “Absolutely — that’s why the experiment didn’t work. We didn’t have rules in place. We didn’t have the context the platform needed to execute the right way.” And it’s not that I’m throwing shade on any of those vendors — their technology is phenomenal. It’s just that to get the most return out of that investment, there’s the human side, which is the messiest part, and frankly, AI can’t fix that. That’s where we were saying, let’s fix the human side — the management, the governance — first, then apply the technology the right way.

Ryan: Yeah, absolutely. Part of what we saw there too was not just the data issue, but when you plug that new tool into systems that have been around for 40 years, how do those two systems talk to one another? What does that integration look like? Maybe that young guy who was there last year asking questions built a phenomenal AI tool that does exactly what he says it will do — but that’s in a vacuum. It’s not tested against business case A, business case B, and business case C, all with different types of systems built on legacy infrastructure that’s been around a long time. They’ve got to integrate the right way, and sometimes that’s the majority of the battle.

Matt: Exactly. And when you say integration, it reminds me of the other thing that blew my mind, honestly, because I grew up in the distribution space, Ryan. I remember when there were no OMS solutions, no PIM platforms, no automation — everything was printed out on departmental printers and shuffled around the building. So what’s funny to me is that, for the first time in my career, I ran into distributor roles dedicated specifically to AI development. Granted, the folks I saw were with larger distributors who probably had a huge line item in their budget for AI, and they said, “You know what we need? We need people who are going to code and develop bespoke solutions for our business.”

So these people are literally vibe coding unique solutions for company-specific problems. I remember when the IT department would just create custom workarounds in the ERP, but this takes it to a whole other level. Ryan, what was the thing we were talking about with the one person in that role, and the challenge they were having?

Ryan: Yeah. From the top down, when leadership says, “Hey, we need to focus more on AI,” we had some companies where the person basically said, “I don’t really know what AI is, I’ve just been told we need to utilize it more.” So you had an AI engineer vibe coding different things to make people’s roles easier. You had people coming in who were interested in what the future looks like, trying to wrap their heads around what AI means for the future of distribution. Then you had people coming in looking for tools that were already built — AI-ready, AI-run, AI-led — that they could just plug into their system. And then you had people who said, “I know what AI is, I’m going to help people at my job use it to make their roles more efficient.”

So you had it in the job-description-efficiency way, the company-data-transformation way, and then people just going, “What is this? I need to wrap my head around it. Where are we going to be in five years? How do I prepare for it?” It was cool talking to people from all those different perspectives, but it did seem like a lot of people were trying to just plug holes and say, “This is what we’re working on.” And that brought about a whole bunch of new issues around security risk. If you’ve got multiple people creating the same type of agents within a company and they don’t know about each other — say there are 2,500 people at a company — is this guy in e-commerce building almost the exact same agent as somebody in another department? There needs to be cross-referencing, governance, guardrails. A lot of people were scratching their heads on how best to design that so they’re not duplicating effort and, more importantly, spending revenue in the same place twice.

Matt: Yeah, exactly. One of the crazy things too was that a couple of the folks I talked to weren’t there shopping so much as doing reconnaissance to figure out what they could rip off from a software company. I’m laughing about it because we’re not a software company, but if I were, I’d be a little freaked out by that. You have these companies that traditionally rely on you to provide software as a service, and now they’re starting to build their own software as a service. But it’s easier said than done. We live in a world where you can use Claude Code and spin up a really beautiful prototype. But taking that prototype and productizing it — making it a multi-tenant, secure, enterprise-integrated solution — is a whole other thing.

And that’s something we’ve been seeing at Pivotree recently, so it lines up with real-life examples. We have customers who come to us and say, “I want to implement this e-commerce platform,” or “this data management platform.” Traditionally, gathering requirements has been half the battle — building out all the use cases, all the details about how we want the solution architected, how it needs to run, and all the different angles we need to consider before we put hands on keyboards. Now our customers are coming to us with ready-made prototypes they’ve coded, saying, “Here you go, that’s what I want — now go do the integration, make it system-ready.” It’s a fun thing we’re starting to see from our end as a system integrator.

But the other thing we heard a lot, Ryan, was this future focus on AI and innovation, and yet many of the companies we talked to weren’t even on step one — maybe more like step zero. What did you see there in terms of digital maturity?

Ryan: Yeah, I think the common theme when everybody came up to our table was to say, almost in a bashful way, “We’re way behind. What these guys are talking about with their tools — we’re not there yet. We still have tons of manual spreadsheets, we still have all of this going on.” Short of a couple of companies, everybody felt like they were way behind where they should be in that process. For me as an account executive, that was great, because I could start to say, “Okay, let’s plan out what your next three to five years look like.” That’s step zero — let’s look at where you are, where you want to be, and how we get there. Then step one, data. Then step two, making sure you’ve got the right systems in place so they’re AI-ready. That was a big part of it.

Matt: One thing we saw was this focus on the generational shift — people talking about the boomers and millennials starting to leave the business. You had a really good point on that. I’d love for you to share your thoughts.

Ryan: Yeah, there were a lot of companies there — some of the AI companies whose tools we saw were built around the retirement cliff, the tribal knowledge in their company that’s about to walk out the door. What we didn’t see a lot of was people asking, “Who’s coming in underneath to fill those roles? Who are these younger people who grew up being able to click to buy a Tesla, who are used to the Amazon experience, who order food at the click of a button and track it until it’s at the door?”

That’s not a data transformation — that’s a business transformation. It’s an entire shift that requires a new point of view on how to fulfill the needs and wants of these buyers, not just direct-to-consumer, but B2B — because they’re going to be the heads of companies, the heads of procurement. It’s a huge competitive advantage to be one of the first in your marketplace offering that type of buying experience. In my opinion, we didn’t see enough people addressing that issue. You hear “digital transformation” everywhere at these conferences, and I happen to think it’s a bit of an overused term, because the change management that needs to happen isn’t a digital transformation — it’s a business transformation. It’s a 180-degree shift in where companies need to set their sights in order to capture the buyers coming through, and whoever does it the cleanest and the quickest is going to take market share at a rate we haven’t seen since the adoption of online commerce.

Matt: I totally agree — that’s such a good point about business transformation. It reminds me of some of what I talked about in my breakout session. This was the first time you heard me give this talk — I started with print catalogs as common ground we can all relate to, and ended with AI catalog management, digital catalog management. What were your takeaways from that, in terms of creating a better digital experience?

Ryan: Yeah, my big takeaway — and what I heard from people I talked to afterward — was really about providing that ease of use. It lets people working on the catalog focus on the SKUs that are performing well, and on other things that can be an advantage to the company in terms of revenue and supplier partnerships, instead of the nitty-gritty work of putting a catalog together. And it’s so true what you said in that breakout — you couldn’t send something to print unless you’d revised it fifteen times, because you had to know it was perfect; there were no do-overs. I think sometimes people take for granted that once something’s online, they can go back and change it later. How many searches have the two of us done where we’ve seen a listing that’s clearly what someone typed up three years ago and it hasn’t changed since — and people searching are never going to find that product because of how it’s listed or what attributes it’s missing?

Having a catalog you’re constantly revising and working on — both the SKUs that aren’t getting the click-through, purchase, and fulfillment rates you want, and the ones that are doing really well and deserve more focus, since they’re your biggest earners — I think that’s just as important as filling in the blanks.

Matt: Yeah, exactly. I love that there’s this connection between catalog management and business transformation, because the common denominator is change management. Catalog management, in and of itself, really is change management — and that’s where most B2B companies, especially distributors selling hundreds of thousands of parts across thousands of brands on their website, get hung up. It’s an overwhelming task. Something is always changing. Some parts are pretty evergreen — an industrial tool or component might be the same for the next five to ten years. But products are always getting discontinued, products get updated, a manufacturer creates a brand-new data feed that has to be picked up, translated to a distributor’s unique schema, and published consistently.

We like to think of all that as a technology problem, or think we just don’t have the people to do it — the human capital. But really it comes down to the discipline of having a process, following that process, and knowing what the next right thing to do is. That message really resonated with people, because there’s AI fatigue across every industry right now — not just distribution. We’ve all been hammered by the promise of AI. I’d ask anyone listening: has AI saved you a ton of time yet? It may have just helped you do more, but at the end of the day you’re still grinding, still doing the same work — we’re just not quite over the cliff yet where we’re seeing real net gains. That was another big takeaway: we’re not quite at the point of seeing ROI on these AI investments, and my contention is that comes back to the human management of the business.

Ryan: Yeah, it’s a great point. I have a 50/50 relationship with AI. I use it every single day, and half the time I plug in some information, ask it to do something, and the output is excellent — copy-paste ready. The other half of the time, I go to run some type of report, and it puts somebody from my side of the fence, at Pivotree, in as the prospect’s president or CEO — and I have to go back and reread everything, way more content than I needed, figure out where it made mistakes, ask it to fix it, reread to make sure it didn’t mess anything else up. By the end, I feel like I’ve been in a bit of a scuffle with an AI program at my desk. It’s a love-hate relationship.

We’re not where it needs to be to revolutionize the industry we deal with yet, but you can definitely see the breadcrumbs leading toward it. That’s also why it’s so important to have that human element — when you pick up the phone, you can call and say, “Hey, what’s happening here?” Because if you buy an AI tool that needs to integrate with an old system and it’s a small business where three people run that AI tool, the odds that someone in customer service understands your architecture and your legacy systems well enough to walk you through the issue are slim to none. So a lot of the time, they’ll just say, “That’s an issue with your ERP,” or “That’s an issue with your e-commerce site, not our tool” — and then you’ve got no choice but to call the company behind that system, and they’ll say the same thing back: “Everything was working fine until you installed this new tool, so you need to call them.” It’s important, at this point, to have that human element — to make sure the guardrails are set, the QA is double-checked, and that it’s experts doing that, not just someone at work you’ve told, “Hey, we’ve got this tool now, double-check it.” You want people who’ve done that kind of work their whole lives.

Matt: Yeah, I love that. It really comes down to a theme we talk about a lot here at Pivotree, which is RI plus AI — real human intelligence plus AI, driving business outcomes better, cheaper, faster. That’s something we can wrap our minds around and achieve right now. Is there a day when AI and agents are able to do the work of human beings? Maybe. But what we’re starting to see, at least in our business and with our customers, is that the power of AI is to be a multiplier for the domain experts you already have. You have somebody in your organization who’s sharp about your supply chain, who understands your manufacturer product lines, who understands your customer base — how do you take what they already know and translate that into a digital experience that reflects the way you serve customers offline?

I think we’re starting to see that. A couple of resources as we wrap up: if you haven’t seen it already, one of our earlier episodes was with our CEO and CTO, Bill Di Nardo and Joel Farquhar — you’ll love that conversation, it’s about the way we think about AI plus RI, real human intelligence. Great episode to check out. The breakout I referenced on AI catalog management is also on our YouTube channel. And finally, Ryan — since you’re an account executive, we’d be missing the mark if we didn’t say: reach out to somebody like Ryan if you’re in distribution or manufacturing, so he can work through step zero, step one, or step two with you, wherever you happen to be in that AI journey. Somebody like Ryan is that one throat to choke — one person who can coordinate all the complexity of AI and what it looks like in your business.

All right, Ryan, thank you so much for coming in, and for being my travel buddy for the week and participating in the conference. It was such a pleasure to have you, man.

Ryan: Yeah, it was a ton of fun. Thanks for inviting me on the podcast. I listen all the time, share it all the time on LinkedIn as well. It’s always been fun watching, and it’s cool to be a part of it.

Matt: Awesome. Catch you next time.

Ryan: See you, bud.

Matt: Thanks for tuning in to this episode of Data vs. Commerce. New episodes drop weekly. So if you’re responsible for any part of how products get from a database to a doorstep, subscribe now on Apple, Spotify, or wherever you listen.