Ninety-five out of a hundred companies get no measurable return. The reason is rarely the technology.
An MIT report on 300 deployments found that 95% of organizations investing in generative AI were getting zero return. Not a small return. Zero. It is the single most useful number in this industry, and almost nobody selling AI will show it to you.
The instinct is to read that as a technology problem. It is not. The same report found the barrier is not infrastructure, regulation or talent — it is that most systems never learn the specific, messy logic of the business they were dropped into.
The expensive version of the mistake
A vendor demonstrates something impressive. It handles the general case well. Nobody has yet asked which part of your operation is actually losing money, so the tool gets pointed at whatever is most visible rather than whatever is most expensive. Six months later there is a pilot nobody can kill and nobody can scale.
That is the pattern the audit exists to interrupt. Not because an audit is inherently virtuous, but because the alternative is buying a solution before anyone has priced the problem.
What to ask before anyone builds anything
Which workflow costs the most right now, measured in hours and lost jobs rather than in opinion. What it would cost to fix. What the fix returns. Whether the people who do that work every day agree with the diagnosis. If a vendor cannot answer those four, the demo is not evidence.
Figures in this piece come from The GenAI Divide: State of AI in Business 2025, MIT NANDA, July 2025, a preliminary report rather than peer-reviewed research. It is worth reading in full.
Photograph: Kindel Media / Pexels
