Your AI Plan Is Really a Data Plan: What to Fix in Your ERP First

If you’re weighing AI for your distribution or manufacturing business, the most useful thing you can do this quarter isn’t picking a tool. It’s getting your ERP data in shape. Research published over the past few weeks keeps landing on the same point: data quality, not the technology, is what’s holding most companies back.

I’ve been reading a lot of AI coverage lately, and this is the thread that keeps showing up.

Is AI actually paying off for companies like mine?

For some, yes. In NAW’s inaugural 2026 AI Adoption Index, 47.4% of distributors who assessed business impact said AI is already having a moderate or significant effect on back-office efficiency. That’s a real signal.

But the same research shows a gap. According to MDM, 68.1% of respondents ranked data quality and integration among their top three barriers to adoption. Only 39.4% reported a formal data inventory, and 32.2% have formal AI governance.

Why does data quality matter so much for AI?

Because AI works from what you feed it. If item descriptions are inconsistent, units of measure don’t match, or approved substitutes live in one person’s head, an AI assistant will give wrong answers, and it may sound sure of itself.

A study covered by MDM this month shows how that plays out. Reshape Automation tested leading AI models on 100 industrial parts questions across 14 manufacturers’ catalogs. The best score, with web search turned on, was 52%. An estimated 36.5% of the wrong answers stated a specific fact, figure, or part code without any qualification. (Reshape wrote and graded the questions itself, so the direction matters more than the exact numbers.)

It’s not only a distribution issue. In the 2027 Third-Party Logistics Study, released in early October, “data quality and availability” was the top AI barrier at 61%, ahead of legacy systems and integration complexity at 54%.

Do we need a new system before we can use AI?

Not necessarily. Acuvera Tech’s implementation consultants often find the bigger win is cleaning up and governing the data already in Epicor Kinetic or Prophet 21, so the AI capabilities Epicor is building into its ERP have something reliable to work with.

MDM’s coverage of the NAW Innovators Summit made a similar point: targeted AI projects can run alongside larger ERP or modernization efforts and deliver nearer-term returns.

What should we do now?

Based on the research and what our consultants recommend:

• Start with one business case. Pick a problem with a number attached, like quote turnaround, pricing accuracy, or stockouts. As one Summit takeaway put it, AI without a clear business case is “experimentation.”
• Inventory your data. List where item, customer, vendor, and pricing data live, who owns each, and where it tends to go wrong.
• Clean the item master first. Descriptions, units of measure, cross-references, and approved alternates feed quoting, purchasing, e-commerce, and any AI you add.
• Test on your own catalog. MDM suggests testing tools against your own products and your hardest customer questions, requiring traceable sources, and deciding when a specialist must review an answer.
• Set simple governance. Decide who can turn on AI features, what data they can use, and how you’ll track accuracy, returns, and rework, not just speed.

The bottom line

AI is starting to deliver for distributors and manufacturers, but it rewards companies whose data is ready. A few weeks of focused cleanup now makes every AI feature you add later more useful and more trustworthy.

If you’d like a second set of eyes on your Kinetic or P21 data before your next AI step, book 30 minutes with Joe.