Despite rapid breakthroughs in artificial intelligence, shopping remains one of its most stubborn frontiers—not because of limitations in AI itself, but because of the fragmented, inconsistent product data it depends on. In this TechBullion interview, Rob Gonzalez, co-founder and Chief Innovation Officer of Salsify, explores why AI-powered shopping continues to fall short of its promise. Drawing on his experience helping global brands and retailers manage product information across the digital shelf, Gonzalez explains how poor data foundations undermine even the most advanced AI systems, why conversational commerce is exposing these weaknesses faster than ever, and what organisations must do now to make AI-driven shopping truly useful, reliable, and competitive.
To start, can you introduce yourself, share your background, and explain what Salsify does?
I’m Rob Gonzalez, Co-Founder of Salsify. Salsify invented the Product Experience Management (PXM) category in 2012. We are a software platform that helps thousands of the world’s top brands — L’Oreal, Coca-Cola, Nestle, and many others — to make every product experience matter. The core of our platform offering is a system of record for product information and content, a workflow system powered by automation, and a delivery network that sends channel-optimized data to every retail partner.
From your perspective, what is the biggest disconnect between how AI shopping is being built today and how people actually shop?
Shoppers are omnichannel, not single channel. They jump between modes of purchasing depending on their specific needs. They’ll shop in-store for some things, online for others. Search online and buy in-store. See something on Instagram and buy from a Shopify site. It’s fluid.
There seems to be this belief amongst AI maximalists that agentic commerce will somehow take over all other forms of commerce. This is nonsense. It’ll just be another way to shop. Sometimes it will make sense, sometimes it won’t.
So the interesting question, to me, is when and why and how many shoppers will start with AI vs. other means? For groceries, people will tend to continue to go in-store. For consumer electronics, maybe they’ll default to AI. For household essentials, maybe they will default to subscription or Amazon “Buy Now”. It’ll be a mix.
And, pragmatically, this means that “agentic commerce” will remain a minority part of the mix for quite a long time.
Why does incomplete or fragmented product data cause shopping systems to break down?
The fundamental issue has to do with search. The more product data — and the more accurate it is — that a search algorithm has (and, increasingly, AI-based search), the more likely it is to surface a given product as an answer to a search query.
So if data is missing, incomplete, inaccurate, or not optimized for search, it won’t be found. And if it isn’t found, it won’t be bought. So: product data has a direct impact on sales.
What are some basic shopping questions that systems still struggle to answer, and why do those gaps persist?
So many!
One major one is product compatibility; can this windshield wiper work on this model of car? Another one is based on taste / matching; what pants work with this blazer?
There’s a whole class of questions around projects. For example, you’re moving. You need boxes and tape, but also maybe a new rug, and also maybe new home insurance, and so forth. How do you organize your shopping and spending and budgeting around this whole project vs. each individual piece?
Then there’s a whole class of personalization. That shirt looks good on that model, but he’s 6’0″ and I’m much shorter, will it work on me? Or, I liked this book and that book but NOT this other book that most people recommend. What should I read based on that nuance?
How has the rise of AI changed what good product data looks like compared to traditional ecommerce feeds?
On the one hand, nothing has changed. People still shop through traditional ecommerce far more than through AI. It will continue to pay for quite some time to do traditional ecommerce well. Also, AI agents scrape existing product detail pages for data, so by getting those right, you help win with agentic commerce too.
On the other hand, agentic shopping systems ask for new data that isn’t traditionally used in ecommerce. OpenAI’s public data feed, for example, has fields like “Intended Purpose / Role” and “Event Context / Use Case” which is not information that is typically stored and used in traditional ecommerce. This information is unstructured text and is about describing a products’ contextual relevance in the world. That is new.
Google and others are moving toward more conversational shopping experiences. What does this shift mean for how brands should prepare their product data?
This is an AND not an OR in the sense that brands will have to do everything they do today AND they will start having to create and organize the contextual relevance data I talked about in the last question.
Another thing brands will have to start doing if they aren’t already is lean into how much their product information appears in social media and third party sites like Reddit. “What’s your Reddit strategy?” should be a newer question most brands should be asking themselves.
As AI becomes a common starting point for shopping, how does this change visibility on the digital shelf, and what risks do brands face if they don’t adapt?
The key part of that sentence is “a common starting point for shopping” vs. “the common starting point for shopping”. People will still shop in-store. They’ll still shop using traditional ecommerce. They’ll still discover new products on social media. So for any shopper on the digital shelf, visibility is, from one point of view, unchanged.
What’s key is visibility on the agentic shelf; when a shopper is using AI for discovery (I’m skeptical it will be used for transactions at any volume anytime soon) how do you make sure your products are visible? OpenAI’s product data feed gives us strong hints as to what information matters, and it’s the contextual data: when is the product used, what it is used with, what is it used for, what is it used instead of, etc. That will help impact it. (And of course there are already like a billion “AEO” professionals who claim to know how to do this well…)
Looking ahead, what shifts in shopping or product data should business and technology leaders be paying attention to over the next year, and how can they avoid falling behind?
Bill Gates said, “People tend to overestimate what can be done in one year and to underestimate what can be done in five or ten years.” Your question is about what changes over the course of one year. The answer is: very little. We’re >30 years into ecommerce (Amazon was founded in 1994), and ecommerce is around 25% of addressable retail spend in the US. In tech, “the cloud” has been a thing for decades, and, by many estimates, only around 30% of enterprise workloads are in the cloud. IBM still has a large mainframe business.
“Old” companies are also quite capable of adapting across technology lifecycles. I recommend the book Leap, which talks about how Proctor & Gamble, John Deere, and others have ridden several massive technology revolutions over well over 100 years (while some of their competitors did not). Walmart is a trillion-market cap company, while Sears is gone. Microsoft has reinvented itself, but Sun Microsystems is gone. D2C brands had a lot of growth in the 2010s…and then stalled out as older brands mastered the channel. The technology itself is not what determines winners and losers; the technology is, after all, available to both upstarts and incumbents.
So what to do in the short run, the next year given this?
First, focus on standard ecommerce data — especially on what shows up on the Product Detail Pages! The AI systems are scraping the data off all of them, so by doing these well, you optimize for current and future shopping. It’s a “no regret move”.
Second, have a team tasked with being on top of the Agentic Shelf, with a test-and-learn budget, with the ability to run tight experiments and report back learnings to the enterprise. I would keep this investment small. After all, the world moves slowly, the gross dollars from AI are small, and there aren’t any clear “early mover” advantages to be had.
