Applied AI for Distributors: the room agreed on the bottleneck

I spent three days at Applied AI for Distributors in Rosemont, and gave a short talk of my own on the state of B2B product data. This is a quick field report — the thread that ran through every keynote, plus the one point I wish had gotten more airtime.

The room has moved past "should we"

A year ago, the question in distribution was whether AI belonged in the business at all. This year that question was simply gone. Every keynote — Graybar, Grainger, Distribution Strategy Group — took adoption as a given and spent its time on execution.

Graybar's Ed Fenton put it bluntly: "The technology is not the problem. It's going to be your data, your processes and your people." Grainger's Jonny LeRoy told the room to "rethink your processes before sprinkling them with AI pixie dust," and to treat today's models like an intern — enthusiastic, energetic, and fallible. Jonathan Bein and Brian Hopkins argued the real moat isn't the large language model everyone can license; it's your own institutional knowledge and data.

Strip away the logos and it was one message: the model isn't the hard part. Your data is.

I agree completely. But "fix your data" is where most of these talks stopped, and it's exactly where the real work starts. So here are the two things I'd add.

"Don't wait for perfect data" and "garbage in, garbage out" are both true

There's a tension every distributor in that room felt, even if nobody named it directly. Fenton is right that you shouldn't wait for a perfect catalog before you start — "you just need some of what's in there to be meaningful enough." Also true: point AI at a messy catalog and all you get is wrong answers faster, at scale.

Both are correct. What reconciles them isn't cleaner spreadsheets. It's the layer underneath — your taxonomy, your attribute definitions, your category templates, your validation rules. That's the steering wheel. AI is only the engine: a very fast executor of instructions. Vague instructions, vague output. Give it your rules and it can enrich a hundred thousand SKUs against your standards. Skip them and it just gets you to the wrong place faster.

You don't need perfect data to start. You need the rules that let AI make data good — continuously, on every new SKU load, instead of as a one-time cleanup project that's stale in a quarter.

Your buyers aren't only human anymore

Here's the part that barely got said out loud, and it's the one I'd most want a distributor to leave with.

Product data used to fail in two places: your PIM, and your storefront. There's now a third. LLMs and AI agents are becoming a buying channel of their own, and they read your catalog to decide what to surface and, increasingly, what to buy. Thin or wrong product data doesn't just cost you a conversion on the product page — it makes you invisible to an entire class of buyer that never loads your website at all.

Everyone at that conference is racing to deploy AI inside their business: in search, in customer service, in forecasting. Far fewer are asking whether their catalog is legible to the AI that's about to be shopping it. That's the real shift in stakes, and it lands hardest on product data specifically — not "data" in the abstract.

What the fix actually looked like

One example I shared on stage. A national building-materials distributor: 100,000+ active SKUs, hundreds of suppliers, multiple DCs. The fix wasn't a heroic cleanup project. It was a continuous pipeline that treats every incoming SKU the same way — source it from wherever it lands (a 14-tab spreadsheet, a 2019 PDF catalog, a supplier portal), normalize it to their taxonomy, validate it, and score it on completeness, correctness, and consistency.

The numbers that moved:

• New-SKU time-to-live went from 18 days to same-day.
• Attribute completeness went from 35% to 88%.
• Same data-entry headcount — now doing enrichment and QA instead of retyping spec sheets.
• Product-page conversion up 12%, on the back of more complete, unique content.

What mattered wasn't a one-time scrub. It was that every future supplier load runs through the same pipeline, so the catalog stays good instead of decaying back to where it started.

The takeaway

The conference's real conclusion wasn't any single keynote. It was the consistency across all of them: execution is replacing experimentation, and the thing standing between a distributor and real returns is almost always the state of the data underneath.

I'd sharpen it one notch. For distributors, that data is your catalog. The fix is rules plus a continuous pipeline, not a one-time cleanup. And the clock is now being set by AI buyers, not just human ones.

If your catalog is on your mind after the show, that's what we do at Anglera. Send us a sample and we'll show you the specific gaps in your own data — no commitment. You can reach me at amay@anglera.com.