
Someone building AI quoting for wholesale distributors shared what actually went wrong. Two lessons stand out for any owner being pitched AI: decide which steps it never touches alone, and judge its accuracy in the unit you pay people to handle.

You've probably thought about an eventual exit. Most owners meet a buyer's questions for the first time in diligence, when there's no time left to fix anything. Those same questions, asked now and worked through in small steps, make the business clearer to run today, whether you ever sell or not.

AI-powered planning tools don't fix bad data. They amplify it, generating wrong purchase orders faster and with more apparent confidence than the manual process they replaced. Below roughly 85% BOM and inventory accuracy, the tool stops helping and starts working against you, and teams don't catch it right away. A couple months in, once the pattern of wrong orders is undeniable, they quietly start keeping a parallel spreadsheet again.

A researcher who talked with thirty plant managers found that manufacturing exception data wasn't the problem: every system already caught the defect and flagged it. What was missing sat one step further, a designed process for who owns the fix, on what deadline, with automatic escalation if it stalls, instead of a manager personally chasing it down by hand every time. The piece argues this gap is a process design failure, not a technology gap, and stays that way even in a fully automated plant, since deciding what to do about a flag is still human work.


Most ERP migrations don't fail on the software. They fail on the workarounds nobody wrote down, the ones that only surface after go-live as a "coverage gap" instead of a bug. This piece walks through how to find those workarounds before they find you, and the one question that tells you whether to drop them or build them in for good.