The Series
First I told you your AI is rented. Then I told you your traffic is rented too, and the rent just went up.
This one's about the money you're spending to fix all that. Because the industry has an answer for you, and the answer is a pilot. An AI pilot. Everyone's running one. Almost nobody's getting paid.
The Numbers
MIT's Project NANDA studied 300+ enterprise AI deployments and put a number on it that should be printed on every vendor's business card: 95% of generative AI pilots produce zero measurable P&L impact. That's from August 2025 — I'm dating it for you, because vendors never date anything.
RAND ran their own post-mortem across 2,400+ enterprise initiatives: 80.3% of AI projects fail to deliver their intended business value.
And S&P Global found 42% of companies abandoned at least one AI initiative in 2025 — up from 17% the year before — at an average cost of $7.2 million per failed enterprise project.
Read those three together. Different researchers, different methods, same shape: most of this spend is a bonfire.
The Honest Caveat
Now, the MIT number caught flak, and you deserve the fine print. Critics pointed out the 95% measured custom enterprise pilots chasing rapid revenue impact — while personal AI use inside the same companies was quietly working fine. Employees loved the tools. The official projects still didn't pay.
Sit with that for a second, because it's not a rebuttal. It's the whole diagnosis.
The tools work. The deployments fail. Which means the problem was never the AI — it's where businesses point it.
The Part That Should Stop You Cold
Here's the finding that kills the standard response: investment levels don't correlate with returns.
The companies spending seven figures fail at roughly the same rate as the ones spending five. So when the pilot disappoints and the vendor says "you need the enterprise tier," understand what's being proposed: doubling down on a bet whose odds don't improve with stake size. You've seen this table at the casino. It's the one with no maximum bet.
More money doesn't fix it. More AI doesn't fix it. I know, I know — I sell AI. Stay with me.
The Difference
So what are the 5% doing? Because they exist, and they're not smarter, and they're definitely not richer.
They started at the other end.
The 95% bought a capability and went looking for a use case. A licence first, a workshop second, a "digital transformation roadmap" third, and somewhere around month nine, someone asks what it's actually for.
The 5% started with a specific, expensive problem — one they could put a dollar figure on — and worked backwards to the shortest path between AI and that money. No roadmap. No pilot phase. A problem, a tool, a number to beat.
Ask yourself which order your last AI project ran in. Be honest. Nobody's looking.
The Shortest Path
Every business we've ever audited has the same specific, expensive problem sitting in plain sight: the database. Thousands of contacts, every one of them paid for — ads, referrals, time — most of them never converted, none of them being spoken to.
That's not a marketing observation, it's an accounting one. It's the largest pre-paid asset on the books that isn't on the books.
Which is why it's the profitable entry point to AI. Not because database reactivation is glamorous — it's the least glamorous thing we do. Because it's the one AI project where the money is already sitting there, the cost of starting is near zero, and the result shows up in a calendar, not a slide deck. One client: $1.4 million in pipeline before they paid us a cent. Another: 224 booked sales calls in a day, off a list they'd written off.
And the cash it collects funds the next rung — and the next. AI that pays for itself before it asks for a budget. Everything in the other order is a pilot. You've read the pilot statistics.
The Question
Last time the question was: if Google sent you zero clicks next month, what survives?
This one's simpler, and it's the one I'd make every vendor answer before a dollar moves:
What, exactly, will this pay us — and by when?
If the answer contains the words "transformation," "capability," or "roadmap," you've found the 95%. If the answer is a dollar figure with a date on it, you might be looking at the 5%.
We built a one-page test for this — seven questions that tell you whether AI will make you money this quarter or just burn your time. Takes two minutes. Stings a little.
