AI is already deciding who makes the shortlist
This episode previewed a Shoptalk Fall session called Turning Data Into Dollars in the Age of Agentic Commerce. Brian opened with the pace of the underlying shift: by 2030, the number of searches conducted through AI or answer engines is projected to exceed traditional search, only a few years away. He cited a striking figure from Forrester’s Buyers’ Journey Survey of nearly 18,000 global business buyers: 94% of global business buyers used AI at some point during their most recent purchase process, with usage spread across every stage of the funnel rather than concentrated at the top.
Brian followed with a number that framed the entire conversation. According to MarketScale, only 44% of B2B marketers currently have the data infrastructure in place to support AI initiatives, meaning a majority do not. Andy said he expected the figure to be even lower. Since most companies are not specifically prepared for AI, and AI increasingly decides who shows up in a buyer’s shortlist, the practical question for anyone listening becomes whether their own data gap is costing them revenue, and how to close it.
Andy Didyk: the funding is roaring ahead of the data infrastructure
Joining the conversation was Andy Didyk, CEO of Ntara, a systems integrator that helps companies selling physical products through multiple channels implement product information management, digital asset management, and master data management systems, ensuring products show up consistently and in the right context everywhere they are sold.
Didyk shared original research Ntara conducted earlier this year with WBR, inRiver, and Bynder, surveying about 100 senior manufacturing leaders on their data infrastructure and AI initiatives. The results were, in his words, more stark than the numbers already discussed. Seventy nine percent of respondents said their organization has C-suite directives to implement AI, either for internal use or for customer-facing and business development purposes. But only 7% said their systems were fully integrated and ready for AI to reliably take advantage of them.
Why structured product data is what AI actually needs
Didyk explained that demand for PIM and DAM systems is surging, with software partners like inRiver, Bynder, Akeneo, and Aprimo all seeing significant growth. The gap is not in adopting these systems individually, it is in connecting them to the rest of the organization. Companies are standing up product information management and digital asset management to create structured data, but many are not commissioning the integration work needed to give AI engines the additional context they need to function well. The result is AI that returns confident-sounding answers that are frequently incomplete, a problem Didyk said he and Andy witnessed together firsthand during a recent workshop, where answers degraded noticeably once you probed one or two levels deeper.
Andy asked directly whether this was a matter of will, capability, or simply someone taking initiative. Didyk called it a combination of all three, but pointed to something more structural underneath it.
You can’t walk into many organizations and say, show me who’s in charge of your product information. Give me the one person whose job it is to ensure that things are showing up appropriately across all the channels. Companies are just now starting to wake up and think about organizing around the fact that their product data is their storefront across any digital channel.
Andy Didyk, Ntara
He added that this is fundamentally a human and organizational challenge more than a technology one. Every executive he talks to agrees product data needs to be consistent and fed into these answer engines, but coordinating the teams to actually make that happen is where most companies get stuck.
Structured data versus real-time access: enterprises need both
Brian asked whether the real fix is aggregating scattered data into one central place, such as a data lake, or accessing it in real time wherever it already lives. Didyk said the growth of platforms like Snowflake and Databricks shows companies are pursuing both unstructured and structured data strategies simultaneously. Ntara’s focus is specifically on the structured side: how product attributes, product families, and taxonomy are organized. He explained that an answer engine like Gemini does not inherently understand a product; it draws relationships based entirely on the information it is given. Feed it well-organized structured data from a PIM or DAM, along with context about how a product is actually used, and it can make useful, if non-deterministic, recommendations. Across both B2B and B2C, structured data drives the outcome roughly 90% of the time.
Why the stakes are higher in B2B
Brian asked whether inaccurate product data carries higher stakes in B2B than in B2C. Didyk said the consequences can be more catastrophic in B2B, using an aircraft tire’s pressure tolerance as an example versus a T-shirt arriving in the wrong size, though he noted consumer categories like food and beverage carry their own serious risks around allergens and material disclosures. He also cited a data point from Ntara’s own survey: even among the subset of respondents whose systems were at least partially integrated, more than half reported a data governance or security issue in the past year.
The bigger structural difference, Didyk said, is what AI is actually doing in each context. In B2C, an answer engine is typically making a single product recommendation. In B2B, buyers are using these tools as part of a broader procurement process, where the AI is effectively producing a shortlist, always delivered with confidence. Incomplete or inaccurate product data at that stage does not just produce a worse recommendation, it can mean a company never enters the consideration set at all.