95% of AI pilots show no profit impact. The skeptics in your company have a point.

You’re right to be skeptical. MIT studied corporate AI pilots and found 95% produced no measurable profit impact. RAND ran its own numbers and landed near 80% of AI projects failing to deliver business value. Those are not rounding errors. That is most of the money.

I’ve watched the same pattern in distribution. The demo dazzles in the conference room, then the rollout dies inside the ERP because nobody scoped the data cleanup, the workflow handoffs, or who owns the output when it’s wrong.

The detail buried in the MIT research is the part worth acting on. More than half of generative AI budgets went to sales and marketing tools, and the measurable payback showed up in back office automation instead: order entry, document processing, invoice matching.

For a distributor, that is the bulk of the day. A CSR keying in sales orders, someone matching BOL’s to shipments by hand. Three people touching every invoice.

My advice hasn’t changed: pick one workflow buried in paperwork, keep a human approving every output until it earns trust, and measure it for 90 days before you spend another dollar. Unlike in Sales, being boring here is good.

If you ran an AI pilot this year, are you still using it in production today?