I’ll be at the Digital Shelf Summit Europe on the 7th October. This year's theme is “Welcome to the Agentic Commerce Era”, so ahead of the day, here is where my thinking is.
There’s a lot of conversation at the moment about AI commerce and what it could mean for the way consumers discover and buy products. It's also raising questions about product data. If consumers are increasingly using AI to research products, ask questions and get recommendations, do brands have the right information in place for their products to be understood and surfaced?
Before we get to AI though, there's a more basic question worth asking: have we actually got the fundamentals right yet?
Salsify’s 2025 Consumer Research found that 53% had abandoned a purchase due to incomplete or poorly written product titles or descriptions, while 54% had done so because product information differed across websites. That field work predates most of today's AI shopping behaviour, which is rather the point. These are problems we already had.
None of that will surprise anyone working in product data, and the reasons are rarely about effort. The same attribute means different things in two retailer taxonomies, so it gets maintained twice. Mandatory fields differ by category and change with little notice. Enrichment gets prioritised by launch date rather than by what a channel actually needs. And the richest product content is usually written by copy and creative teams, for campaign, and never finds its way back into the PIM.
AI adds another layer to that.
Traditionally, we focused on giving consumers the information they need to understand a product and make a purchase, while meeting the requirements of retailers and marketplaces.
But what happens when the consumer doesn't search for a product or browser category in the usual way? They might ask:
“I need a lightweight moisturiser for holiday that will sit well under makeup. What do you recommend?”
For a product to be relevant to that request, it isn't enough to know that it's a moisturiser, its size, the skin type it's designed for and its ingredients. Something has to answer: Is it lightweight? How does it feel on the skin? Can it be worn under the makeup? When might someone use it? What makes it different from other moisturisers in the same range?
Two things make that harder than it sounds.
First, a lot of that context is claim territory. “Lightweight”, “non-greasy” and “sits well under makeup” aren’t neutral descriptors in beauty. They need substantiation, and what is approved in one market may not be approved in another. So, this isn’t only an enrichment problem. It’s a governance one, and it involves regulatory and legal colleagues who usually aren’t apart of the digital shelf conversation.
Second, that information often does exist, but only within product descriptions or marketing copy. AI can read and interpret this content, so the issue isn't whether the information is structured. The difficulty is that the content may have been written for one market, one campaign or one retailer, with no guarantee it's present, consistent or approved everywhere the product appears. Those Salsify figures show what that inconsistency already costs. It’s hard to see why it would matter less when AI is doing the reading rather than a shopper.
This is where I think the AI readiness conversation earns its place, as long as it doesn’t turn into a separate programme of work.
It doesn't necessarily mean brands need to rethink their product data strategy overnight. It does give us a reason to look again at the foundations we already have: our data models, taxonomy, attributes, product relationships, governance and enrichment processes.
Because if getting complete and consistent product information onto the PDP is still a challenge, AI isn't going to make that problem go away. It's going to ask more of the data we already have, and it’s going to do it across surfaces we don’t control and can’t always see.
So perhaps the question isn't simply whether our product data is ready for AI. It's whether we've built a product data foundation that can adapt as the way consumers discover products changes.
AI might be what prompts us to ask these questions, but most of them aren’t really AI questions at all. They’re product data questions, and plenty of us have been asking them for years.
I’m looking forward to hearing how other people are approaching this at the Digital Shelf Summit. If you are going to be there, come and find me. I would love to hear more about how your company is thinking about this and how ICP can help support you.