Transcript

Floyd Blaikie: Good afternoon, everybody, and welcome to Product Data Maturity: Why “AI Ready” Isn't What You Think. Really excited for today's presentation — there's some original research in here that I think is fascinating, and if you're a nerd like me, you'll probably agree. I'll let our speakers introduce themselves today and I'll get out of here, starting with the CMO from Inriver, Jay Roxe.

Jay Roxe: Good afternoon, good morning, or good evening, everybody, depending on where you're listening. My name is Jay Roxe. I have the honor of being the Chief Marketing Officer of Inriver, which is one of the leaders in product information management and product experience management. And with me is Willem.

Willem Van Dijk: Good to be here, thanks, Jay. Willem van Dijk. I look after data platforms and data services at Pivotree. I've been in MDM and PIM for a very long time, and good to talk with you, Jay.

Jay Roxe: Good to talk with you, Willem, as always, and looking forward to today's conversation. So as Floyd said, for nerds like us, there's a lot of stuff that we want to drill in on with what's actually happening in AI, and particularly what's really happening in AI in the manufacturing space. So we'll spend today talking about some research that Inriver did and what we've learned from it, insights that Pivotree can bring from practice in the field with customers, and what people who are listening to this webinar should choose to do next.

The research started with a very simple premise: how ready are people actually for the move to AI? You can't pick up a newspaper, a Harvard Business Review, or have a conversation with my in-laws without being asked about AI today. So we went out and we ran a survey of about 400 people across a wide variety of industries and a number of different segments. It was about three-quarters manufacturers and the rest distributors. One of the things to know about this is it's actually a pretty high-level respondent on the survey. Everybody was screened to be a C-suite, a VP, or a head of for the given function. And some of the questions we asked were understanding how the organization runs today, what they're doing, what's working, what's not, and how ready they think they are for AI.

Pro tip: the confidence is real. Everybody thinks they are on their way to being able to do this. The number of people who actually met the bar of being ready for true agentic commerce when we tested it — and we said, how fast can you actually deploy a new SKU? How are you supporting new channels? What does product data really mean in your environment? Nobody met the bar when we actually tested, despite almost everybody saying yes, product data has a high impact on my revenue. No kidding. I think anybody that's listening to this webinar knows that as we move into an agentic world, the ability of agents, or even the ability of your own people, to operate rapidly on product data is going to be one of the defining characteristics. But I look at the survey and start asking the question of, did we just create an ego test? At the same time —

Willem Van Dijk: Yeah, it's very new, right? And I think the challenge sometimes is that people say they're ready, and then when push comes to shove you find that they haven't thought through a bunch of different things. Like what does it mean to push AI into workflows? What does it translate to when it comes to costing? Is my legal team on board with insertion of AI? Those are the kind of things that we hit up against all the time. When it gets real, you realize there are a bunch of barriers that you've not thought through. And the biggest one of all, of course, is data itself. Is my data truly ready? Is the context of the data there? Can I trust that it does the right thing?

Jay Roxe: In some ways though, I think this finding should make people feel better. You log on to LinkedIn and you get the feeling that everybody has figured out how to delegate their entire job to agents, and they are just sitting there watching an agent swarm do everything for them. The actual truth is a lot messier.

And that's when you start getting into some of the fun parts of the survey. I'm not going to try and explain a maturity model just as part of presenting the slide — I'd invite people to go get a copy of the report where they can find this. But we didn't start out to build the maturity model. We started to ask questions, and then we went back and said, okay, based on the results we've seen, what do we know? And a lot of people are still sitting there in a fragmented tier. They've got no closed-loop measurement, they've got no integrations. But really, in this day and age, that's not as common. The majority of people have some automation. They've got a chunk of their SKUs published and ready at any time, and their integrations fail probably quarterly. But by the time you get into people that have more than 85% of their SKUs published and ready, you're getting into really rarefied air, where only 0.7% — three people — actually met that bar and have closed-loop measurement. So there's a lot of space for organizations to continue to improve as they look at what it means to be ready for an era of agentic commerce.

And so, Willem, why is product data hard to get right?

Willem Van Dijk: Well, it seems straightforward enough, right? You describe a product — how hard can it be? If you haven't lived it, it seems easy. But think about it this way. You describe one product, fair enough, you can do a decent job. Describe 20 products and now try making sure that you're consistent. Much, much more difficult. You need to start thinking about, well, do I have the attribute profile for each of those 20 products? Do they even fit together? Can I describe them the same way? And that's on the manufacturing side. If you're a distributor or a retailer, you're getting those data points from other organizations that all have their own thoughts about how to describe something. So it becomes much, much more complicated. Even a manufacturer has product management teams and engineering teams that tend to just focus on their own work — their content management or data descriptions, they're relatively not interested in that.

So yeah, it's a complicated issue, and then agentic sort of further complicates it, because you're not talking to a person, you're talking to something that needs data in a very structured way. It's changing a little bit, but you need to be very granular, very accurate. I will say it's like talking to Spock rather than talking to Scotty. You need to be granular and very precise. So that complicates the matter further. Very difficult getting it right, and the people who've lived it know that it is.

Jay Roxe: I like the fact that you went with an original series reference there as opposed to a Next Generation one. And I think there's a lot of places where there's a plus side, where AI can actually be the thing that helps to clean the data and take it from something that Scotty would understand into something that could be expressed for Spock. That's the opportunity for a lot of manufacturers and organizations as they start on this journey.

Willem Van Dijk: Yeah, it used to be unpacking in a rule-based way. Now AI can just look at a label and extract the data, right? Or look at a PDF and pull out the relevant attribution that you need to describe products. Huge difference.

Jay Roxe: It's actually pretty cool. We demoed this recently. We took a PDF from a paint manufacturer — just their actual catalog — and dropped it in and said extract. And the degree to which you can take image and specs... and you're always going to want people in the loop. That's one of the key findings. There needs to be a human in the loop just to validate. And let's be honest, your legal team wants a human in the loop as well. But we're moving towards a world of orchestration where that human has fewer and fewer detailed tasks to do, and they get to do more and more at the strategic level.

Willem Van Dijk: Yeah, we're actually turning it around sometimes, where data extraction through AI can review and validate whether what humans have done is accurate or needs another look. So it's interesting how humans can check AI, and AI can also check humans.

Jay Roxe: We know from other research that about 40% of the listings online are inaccurate in one way or another. And finding some of those challenges at the source, and also being able to test it at every phase in the cycle, is one of the exciting pieces even before we get into full agentic commerce.

But the research really showed five things. And I like what you said earlier — there's a lot of things where people just don't know what they don't know. And that's part of organizations being on this journey. I think a lot of the people on this webcast may be more sophisticated than average, but there's hopefully a couple of nuggets here that can be used as people go to explain the journey to others in the organization. So the things we'll talk through, and we'll cover each of these: how we're seeing the explosion of channels, how people are bringing systems together, what are the ways people are doing it today, who actually owns the problem, and what are they measuring versus what should they be measuring? So why don't we go ahead and dive in?

I was actually a little bit surprised that some of these numbers were so low. 79% report that their digital channel count grew over the past two years. I think there's a time boundary on that that's important, because when we've looked at this through other channels, you look at organizations having increased in the two years prior to that by 114%, as they tried to go into COVID speed very quickly and deal with everybody shopping through new channels. So that 79–80% is on top of an already accelerated base that people are trying to adapt to.

And we're seeing new channels come out that people haven't even looked at before, where as the B2B or B2C buyers get younger, the number of people that are seeking even B2B information through social channels, or validation of that information through social channels, means that manufacturers and distributors have a huge catalog of things that they are still accountable for looking at. We were speaking to one apparel manufacturer who's a fashion-forward brand, and they were saying that within 18 months — 18 to 24 months — they expect to get no meaningful traffic from traditional Google search. It's all gone.

Willem Van Dijk: Yeah, and maybe the numbers are low as well, Jay, because there's a lot of fragmentation, right? People often don't have a full picture of everywhere that data needs to travel. So oftentimes we come in and try to deploy and ask the question, well, where does all this data need to go? And it's over multiple days with different groups that you start getting the full picture. And you put it together, and people are just amazed at the volume of channels they need to serve. They don't all know it all, because they have somewhat blinkered views of the stuff they do, and not a comprehensive view of the entire organisation.

Jay Roxe: And many of them are still amazed at the number of people that print catalogs. We have another customer where that is one of their primary use cases, because they sell to nautical shops where that's how business has been done and will continue to be done. And adjusting to all of the channels, be they bound by trees or tokens, is the challenge that organizations are going to continue to face, because this pace of change isn't slowing down.

Which brings us into one of the bigger challenges: how do agents want to consume data, and how do people want to consume data as they are looking at their PLM, their ERP, and their other sources of truth in the organization? How well data flows between those is a core determinant of how well it's going to be represented in these new channels like LLMs. One of the things that we've spent a lot of time looking at — everybody's now very concerned with how am I represented in ChatGPT or Claude or Perplexity or Gemini. And one of the primary reasons that companies, listings, and products get disqualified is that LLMs are very sensitive to when information is different in different sources. If it says one thing on your website and a third-party site has something else, and it can't determine which one is the credible site, it's very much harder to end up in that core recommendation list. So this gap of how things move throughout the value chain has become quite challenging for organizations that have a more legacy infrastructure, and quite important as they're looking at how they're getting discovered in the modern world.

Willem Van Dijk: Yeah, so the interesting thing that I think is happening is people are really talking about passing the bar, right? You need all those pieces of content to be there. They need to be complete, they need to be accurate, they need to be consistent. But there's going to be multiple companies — your competitors are going to pass the bar as well at some point, and then all that content doesn't just need to be there, complete and accurate. It actually needs to sell better to the Spocks of this world than the competition does, right? So there are sort of multiple levels of where you need to be, how accurate you need to be, how complete and how good that data is. And again, it's a combination of people with AI that can really increase your capabilities in that space.

Jay Roxe: The piece that I think gets uncovered as a lot of organizations go through this process — and you were talking about the multiple rooms and the multiple meetings over multiple days — a lot of this is still happening because Bob emails Mary at 2 p.m. every afternoon with the updated Excel sheet on how it happens. Or they have a legacy system that's already been implemented and continued, and the workarounds are just in place to the extent people don't even know their workarounds anymore. They're just how the job gets done. And for almost 90% of organizations, there's a human who's doing that step to move product data. When we start talking about the sync gap and the speed issues that we were looking at previously, that is a huge risk to the organization.

Willem Van Dijk: Yeah, workflow is really important. I mean, the handoffs just go away, right? The minute you approve something it lands with somebody else that can act on it, and that is really helpful. The first person to pick it up, it disappears from everybody else's view. Escalations can be put in. The challenge that we see a lot is that people tend to over-engineer the workflows. They want every edge case resolved inside of the workflow. It gets extreme. But if you keep workflow simple and you focus on the big moves — the gates that a horse needs to travel through to get to the race, right? There's not just one gate, there's a bunch of different gates. And if you define those well, there's huge benefits to going from a manual process to an automated, workflow-driven process. And then again, AI can help with the gate checks, making sure that data is consistent, and if it's not, pushing it back for a person to review it.

Jay Roxe: We've been talking a lot about product information orchestration, because the future here isn't humans emailing things to each other. The future, and the place where a lot of people want to get to, is an orchestrated dance of humans and agents, moving towards a different balance of people and bots over time. And so getting that orchestration that you've described figured out but not over-engineered is sort of the first step in taking the manual backbone and making it something that's a little bit more usable for the organization.

But this next one surprised me, and I'll be curious on your take as you reflect on the customers you've worked with. I guess I shouldn't be surprised based on everything, but only 8% of organizations actually have true cross-functional ownership of the product data, of the journey that that data goes on. And then you look at who actually says that they own the problem, or that they have leadership on the problem, and it's kind of all over the place. You ask the CIO, he or she has leadership. You ask marketing — guess what? They say they have leadership on it. And you end up in a mess where you've got a history, you've got a path of deployment tickets and approvals and people building independent systems, that takes you back to what we had on the previous slide with a manual backbone, because it's all about who actually knows whom to get the job done.

Willem Van Dijk: Yeah, multiple owners is a problem, right? Somebody I know had this saying: the fastest way to kill a horse is to tell two people to look after it. But I think it's an interesting question. One of the things I'm thinking is that IT is sort of the servant here in some ways, right? They need to make sure the business is able to make the right decisions, to route the product data through the organization and deliver it to the channels that need it. If it's not the business, and it's IT driving that and owning that — that's usually how things start, but over time you sort of outsource that into the business and make it the responsibility of the business. You saw that with readying data for e-commerce. For the longest time it just sat inside of an IT function, where really it probably wasn't the best place for it to reside. And then it moved out more into the merchandising and the product management space. I think with AI, that'll be the same way.

Jay Roxe: I think it's true, and I think it will accelerate. I'll be very curious to see how organizations handle the ownership of various different systems. It's pretty common for IT or data to run your ERP, or maybe engineering or IT to run your PLM. I think we're seeing, for a lot of the marketing systems like a PIM, where the product experience and the customer experience is what has to lead. So I'm curious how fast marketing can take ownership of it in an AI-first age.

And then this one absolutely floored me. Maybe it's because of what I do, where a healthy percentage of what shows up in my inbox every morning when I come in is how we're doing on some of the AEO searches, what we're measuring, and how that measurement is impacting awareness and brand. Less than 12% are actually monitoring how they rank in AEO. And this is reflected in what we see every day, because one of the things we've been doing is AEO assessments for customers and prospects and other people we're talking to. And for a lot of people, it's eye-opening. It's sort of the first time they've started measuring it. And some of them have the “okay, what do I do?” reaction. Some of them have the “hey, this is terrible, let's never talk about this again” reaction to it, because they see their own results. And the fact is, if you're not measuring it, you're not going to move it.

Organizations talked for years about digital shelf analytics, and that's become a well-defined science. There's continuous investment, but the new version of DSA is actually understanding how your AEO — your answer engine optimization — plays into that. Where are you being represented? How are you being represented? Can they actually find your product? There's a really interesting stat in Shopify's recent earnings announcement where they said that they saw the number of searches coming in from answer engines triple on a year-over-year basis. But even more importantly, a large number of those searches went to the long tail of products. So it was not “what's the best car seat for my baby?” It's “what's the best car seat where I can fit three of them across in the back of a 2022 Honda Pilot?” People are getting to that level of specificity. And I used a consumer-facing example — the same thing would apply to a branded manufacturer or any other group as well. So there's a huge gap as people need to start looking at how do we operationalize what we're seeing and what we're finding here.

Willem Van Dijk: Yeah, I completely agree with you. It is stunning, especially because there's a whole range of solutions out there now that give you that feedback. And tying that into a sort of virtuous loop of “okay, this is where things are not right, and how do I improve it” is not something that's impossible to implement. So very much needed. If you don't measure it, you won't make it better.

Jay Roxe: And actually starting to do that on a closed-loop basis, because to your point, the measurement isn't a one-time activity. It's not once a quarter, it's measuring it quite frequently. And look, ten years ago, fifteen years ago, at the dawn of SEO, you saw a host of agencies spring up, all of which had their secret sauce for doing this. And it eventually came down to how you have accurate representation of your product, of your content, of the problems you solve — that more or less drove a lot of the SEO work. Lots of other science there as well. I think AEO is still a bit of the Wild West, but my prediction is it's eventually going to come back to that single source of truth idea.

Which will be important, because we've seen traditional search decline. This stat says it's down by 23%. And that's just the searches that end in a click, so you're now seeing fewer than one in three searches click anywhere. I'm actually surprised it's not a greater degree of decline than that, because when you start looking at what a modern Google page looks like, you've got your search generated experience up at the top, and then you've got the paid ads. And since January of this year, Google has doubled the number of pixels given to paid ads. Which means if you are just a regular blue link, you're being pushed further and further down the page, and people are answering their questions up at the top. Forrester has made the point that you're going to start to see that 94% of B2B buyers use AI in their buying process.

I don't know who the other 6% are, but they must know exactly what they want to buy and from whom. It's now where most people start. My favorite example of this: I invite everybody to go home tonight, open your refrigerator, break one of the little clips that holds the shelf to the refrigerator — I may have done this the other day — and try and find the replacement part. In a world before AI, you could end up down a very deep rabbit hole of exploded parts diagrams. And in a B2B world, you're not necessarily finding that part, but you may be trying to repair an HVAC unit or a combine or something else, where getting the important part can get a $3 million machine back online more quickly.

And then Gartner has gone out on a limb, or at least did last year, and said 90% of buying will be intermediated by AI agents. I think they're pretty aggressive on that one. I'm not sure I see that much intermediation happening over the next two years. But the core takeaway for people is it's going to happen, and it's going to happen quickly.

Willem Van Dijk: Yeah, 2028 — I'm with you. I think that is aggressive. But the challenge there is it's a bunch of different things, right? You need a GTIN, you need your content to be there, you need the content to be granular, complete, accurate and consistent. And then you need other things. You need your return policy to be correct — or to be available, first of all, but also to be on par with the competition. You need your checkout protocols to be correct. So there's a lot of things you need to do in order for that agent to be able to come in and confidently say, okay, this is not only the right product, but based on all the peripheral stuff around it, I can go and make this purchase in a trusted fashion. So maybe that in and of itself tells you that 2028 is a pretty aggressive goal, just by virtue of the fact that you need to put a lot of things in place to get that right.

Jay Roxe: It is an aggressive goal, and to our conversation, I'm not sure it happens on that timeline. But it's also true that if you're not moving on it, your competition is. So starting to ask the core questions of how things integrate and how ready you are for agentic leads into three different questions I'd encourage everybody to just ask, to see where they actually are on the maturity index.

If you start looking at the active SKUs — the things you want people to be able to buy — what percentage of them are actually available right now? What we found is that about a third of companies don't even have 85% of their active SKUs available.

We also know, when we start looking at how people are connecting things, there's a huge number of just: ask when the most recent integration failure occurred. How much are we dependent on two people emailing each other Excel spreadsheets? And if one of them wins the lottery and moves to the Canary Islands, we're really out of luck. And then, how many of the AI claims that the team is making would actually survive a “show me that in production” conversation? So just three pretty simple questions that leaders can ask of their teams, or that you can see people asking of their own systems, that I think start to lead into some of the more interesting conversations.

Willem Van Dijk: I'll add another one, Jay. Thinking back maybe, what is it, fifteen, twenty years ago — should I have thought about e-commerce more seriously at that point, right? And if you think yes, then maybe now's the time to start seriously thinking about agentic commerce.

Jay Roxe: I think that is something where, if you weren't thinking about e-commerce back in the day, we may not know who you are right now. So it's a good time for people to be starting to make their plans for it.

Well, how should organizations think about delivering this?

Willem Van Dijk: Yeah, so this is somewhat of a segue to talk about solutions, to talk about how you deliver, right? We've at Pivotree made a serious pivot in how we deliver, and AI is really front, center and backstage on all of that.

Starting with needs gathering. So we use call transcripts a lot. We have smart people asking good questions, of course, and that stuff gets captured immediately in a call. Those call transcripts are fed into an AI platform that then creates user stories, and it also comes back to us. This is a pretty comprehensive system — it comes back and tells us where we've been amiss, where we've not asked all the questions that we should have asked. We go back, fill in those gaps, further tease out the user stories. Those user stories are then fed into different agents that build the prototype that is initially put in front of a customer. And there's a lot of automation in those builds: reviews, feedback back into AI, iterate again through an AI build. So there's a lot of areas where in the needs gathering and the build, that AI platform comes into play. As we build, we also create both documentation and test scripts.

So the entire process, all the way back from what do you need to here it is, and here's how we're testing it, is automated. And then on the back end, there's a lot of assistance from AI in just triaging things that are coming back. Not everything will be perfect. There will be mistakes, there will be things that go wrong. But the triaging of those issues is also using AI, and that's fed into a platform that has the context of the MDM or PIM solution that we implement. And the entire process is just faster, and as a result of it being faster, it is lower cost as well.

Jay Roxe: I think that's where we start to see the advantage of working with experts that have done this many, many times, and the advantage of deploying tools aggressively so that it can be faster, more efficient, and quicker time to value.

So I think the takeaway here is that even if you're confident about where your program is today, there's a lot of questions that are still worthwhile going to ask. And we at Inriver did this research because we do work with a lot of organizations with complex go-to-market and complex product data needs. We see the challenges as they are looking to onboard, enrich, distribute, and optimize the content to an increasing number of channels. So if anything that you've seen here has sparked your interest, or you have additional questions or needs, I'd encourage you to reach out.

Willem, I think this is the point in the presentation when the audience gets to play Stump the Chump. Do we have any questions?

Floyd Blaikie: Well, I'm back, by the way. Hello. Thank you so much, Jay and Willem. That was awesome, and loved all the nerdy references — but you knew I would. And thank you everyone who joined us for the live stream. Because of the way that LinkedIn live streams are set up, we can't take live questions. You can leave them on the event page. But why don't Jay and Willem give them some ways to get in touch with you, if you have any questions following this, or if you're watching it after the fact?

Willem Van Dijk: Best way to reach me is email: willem@pivotree.com.

Jay Roxe: And I'll go with the same. Mine is jay.roxe@inriver.com. So I look forward to hearing from people, and happy to continue the conversation either electronically or live at any one of the conferences that are coming up. September is a busy conference month.

Floyd Blaikie: All right. Thanks so much. Thanks everybody. We'll see you next time.

Willem Van Dijk: Thanks a lot.

Jay Roxe: Thank you now.