We’re buying the engine before the fuel

Master B2B released its 2026 State of B2B eCommerce report this month, and Brian Beck and Andy Hoar asked me to react to one finding on the webcast. Rence Winetrout of GracoRoberts, Megan York of SAP and Daniel McIntyre of Coveo covered the rest. Watch my segment above. This is the argument, written down.

Two numbers from the survey.

AI is the number one technology investment priority for B2B eCommerce teams, at 55%. Data hygiene sits at 35%.

Same survey, same respondents: bad product data is the number one barrier to growth for the second year in a row. Most teams grade their own data a B or a C.

So the industry has correctly identified the problem and then funded something else.

Why this keeps happening

For decades, product data in B2B never had to stand on its own. A sales rep explained the difference between two SKUs. Customer service caught the wrong spec. Someone fixed the price on the phone. There was always a human bridging the gap, so nobody assigned independent value to the data itself.

That human is gone. Buyers ask ChatGPT, Claude and Gemini about your products before they call anyone. The catalog is now out in the open, competing for attention with no interpreter. Incomplete attributes and a price a machine can’t read don’t get a follow-up question. They get skipped.

Companies know this, more or less. What they get wrong is the mental model. Product data is treated one of two ways, and both fail. Either it’s a cost center you fund reluctantly, or it’s a project with a start date and an end date. Clean it once, declare victory, move on. Then you buy more AI, run it on the same broken process, and expect a different result.

The report puts the cost of poor data maturity at up to 20% of revenue, through mispricing, inefficiencies and lost opportunities. If a fifth of your revenue was walking out the door because someone kept breaking into the building, you’d have a security system by Friday.

The one thing I told them

The AI model you deploy this year is obsolete in twelve months. The data layer is the durable investment, and it’s the only place your company’s DNA can live. If every distributor runs the same commodity models on the same quality of data, every distributor gets the same average answer. The only way out is to put forty years of expertise, the knowledge that made you the supplier people call, into the data itself, so the models can read it and recommend you.

GEO starts in the PIM, not in prompts.

Pull fifty random SKUs from your live catalog. Two checks: is the attribute set complete, and can a machine read the price without logging in? That number says more about your AI readiness than your AI budget does.

Then look at your data quality initiative. If it has an end date, remove it. Give it an owner, a cadence and a metric instead.