Last week I asked which of five levels you are operating at.
The most common honest answer inside a $20M–$100M manufacturing business is Level 2. You approved a pilot. Someone sponsored it. It ran.
And then it did not take.
What happened next is the subject of this edition. It is the most expensive misread I encounter in the field, and almost nobody recognizes it as a misread while they are making it.
Seth Godin wrote something this week that put a finer point on it than I had managed on my own.
His subject was rejection letters. The reasons people give when they turn down your proposal, your project, your resume. His observation is that the stated reason is almost never the real one. It is a courtesy sentence, written to close the conversation politely. You can answer it perfectly and change nothing, because you answered the wrong thing.
I have read him for years and he still does this to me.
Here is where it landed. A failed AI pilot ends in a rejection letter too. The business writes it. The business signs it. And it reads exactly like the ones Godin is describing.
The pilot ended. A conclusion formed. It usually sounds like one of these:
AI is not there yet for our kind of operation. Our data is too messy. Our people would never adopt it. We are not big enough for that.
Every one of those is a verdict on the technology, the data, the team, or the size of the business.
Not one of them is a verdict on the thing that actually failed.
Godin's point is that the person rejecting you is not telling you the truth.
Mine is worse. When a business rejects a technology, the letter is addressed to itself. And it is the most convincing rejection letter you will ever read, because you wrote it to be believed.
The Base Rate Nobody Handed You
MIT's Project NANDA published a finding last year that reframes this entire conversation. In The GenAI Divide: State of AI in Business 2025, they report that 95% of organizations are getting zero return on generative AI. Only about 5% of custom enterprise AI tools ever reach production. Against thirty to forty billion dollars invested.
Read that the way it is meant to be read.
If your pilot produced no return, you did not underperform. You performed at the rate.
That should land as relief before it lands as anything else. What went wrong was not a failure of your judgment, your team, or your plant. Something structural is producing this same outcome at nearly every business that attempts it.
And structural causes do not respond to trying harder.
You Fired the Pilot. You Kept the Airframe.
When an aircraft goes down, the investigation does not stop at the pilot. It asks what the airframe demanded of the person flying it.
A tool dropped onto an unarchitected business makes the same impossible demand. It is asked to produce an outcome that depends on information it cannot reach, inside a process nobody has defined, with an owner nobody has named, measured against a scoreboard that does not exist.
It will fail. It has to.
And when it does, the tool takes the blame, because the tool is the only part of that arrangement anyone can see. The missing architecture is invisible by definition. You cannot point at the absence of a decision.
This is what I mean when I say a business is building a perfect propeller.
The propeller is not the problem. It was never going to be the problem. It is a fine propeller, bolted to an airframe designed for a different century of flight. You can machine it to tighter tolerances every year and never once close the distance to the business that changed the engine.
Forty-two years of watching technology transitions produced one pattern that has never broken: the businesses that survive a transition are not the ones that bought the new thing first. They are the ones that decided what the new thing was for before they bought it.
Your Failed Pilot Is the Best Intelligence Asset You Own
Here is the part that costs the most, because it gets thrown away every time.
A pilot that fails is a diagnostic instrument that already ran.
It went looking for information and found out exactly where the information stops. It needed a decision and found out exactly who is not authorized to make one. It produced an output and found out exactly what nobody downstream was prepared to do with it.
That is a map of where your Digital Exhaust is disconnected. Your operation generated it. You paid for it. Most businesses shred it in the postmortem and record the whole thing as a lesson about AI.
The successful pilot teaches you far less. It tells you one thing worked. The failed one tells you where the architecture is missing, and that is the only question worth having answered before you spend the next dollar.
You Are Not Behind Because You Are Small
Same MIT report. Page seven. This one almost nobody has read.
Mid-market top performers go from pilot to full implementation in roughly ninety days. Enterprises take nine months or more.
And MIT defines mid-market as $20M–$100M in annual revenue. That is your band, by name, in a report about the largest technology transition of your working life.
Sit with what that means.
The businesses beating the enterprise on implementation are not the ones with the largest budgets, the deepest benches, or the most sophisticated data science. They are the ones with the shortest distance between a decision and the floor.
You can make a ruling on Monday and watch it change how Tuesday runs. There is no steering committee between you and the plant. No quarterly governance cycle. No four business units negotiating whose definition of a customer wins.
A two-billion-dollar competitor would pay almost anything for that and cannot buy it.
Being small is not why your pilot failed.
Being small is the advantage you did not deploy, because nobody told you it was an advantage, and the vendor selling you the tool had no reason to.
What the Rejection Is Actually Costing
The failed pilot cost you the pilot's budget. That number is knowable and it is survivable.
Rejection costs quarters.
Every quarter spent at we tried that is a quarter your competitor spends compounding. This is what I call the Catch-Up Penalty, and the reason it deserves a name is that the cost is not flat. The gap does not wait politely at the width you left it. It widens while you are not looking at it, and it widens faster the longer the architecture goes unbuilt, because architecture is what makes each subsequent capability cheaper to add.
Which means the business that decided two years ago is not two years ahead of you.
They are two years ahead and adding faster.
The Question to Take Into Your Next Leadership Meeting
Not should we try AI again.
What did the last one actually tell us?
Pull the postmortem. If there was no postmortem, that is the finding. Then ask, of the pilot that did not take:
What information did it need that it could not reach? What decision did it require that no one was authorized to make? What did it produce that no one downstream was prepared to act on? Who owned the outcome, by name, and did they know it?
Those four answers describe the architecture your business does not have yet. They are worth more than the pilot cost.
And you already paid for them.
One More Thing
I am forty-two years into this and I still read every morning.
Not because the fundamentals move. They do not. The pattern that decides who survives a technology transition has not changed once in my working life. I read because the fundamentals get said in a new way by somebody who was not in the room when I learned them, and that new way reaches a part of the problem my own language had stopped touching. Godin did that to me this week. He was writing about rejection letters and he had no idea he was writing about your pilot.
That is the entire discipline, and it is the one I am asking you to turn on your own operation. The information is already yours. You paid for it. Somebody has to look at it again with fresh eyes and refuse the first explanation.
A business that cannot do that is only ever as smart as it was yesterday.
Before You Buy Another Tool
Everybody is selling AI tools. That is tactics.
The work is the strategy and the architecture that runs them, so AI becomes your competitive advantage instead of another expense. And I can tell you, in dollars, what every month you wait is costing you.
You have a failed pilot and the wrong conclusion drawn from it. That conclusion is currently costing more than the pilot did, and it will keep costing until something replaces it.
The Strategic AI Intelligence Diagnostic is the working session where we replace it. We examine what your operation actually revealed, where your Digital Exhaust stops, and what the architecture underneath the next decision has to look like before anything gets deployed on top of it. It is the door to the practice, and it is deliberately the only one.
If your last pilot did not take and you have never been shown why, write to me directly: [email protected]
Not a form. Not a calendar link. A note from you to me.
The Decision Room opens this week: Strategic AI for the Mid-Market Leader. Ten decisions that determine whether AI becomes an operating advantage or an expensive layer of activity, on demand, with a seat for the person you would tell at the kitchen table. Places are being held now at carlpeterlin.com and everyone holding one receives an edition of Your Forfeited Capital Brief that is not published anywhere.
Carl J. Peterlin Jr. is the Strategic AI Intelligence Architect™, with 42 years of pattern recognition across pharmaceuticals, defense, energy, and high-growth startups. He is the author of Death To Excel! and Your SMB AI Revenue Ratchet.
SOURCE NOTES
- Rejection letters and the stated reason. Seth Godin, "Give me one reason," seths.blog, August 2026. Argument paraphrased, not quoted. Godin writes about interpersonal and commercial rejection. The application to a business rejecting a technology is mine, not his.
- 95% zero return · 5% reach production · $30–40B invested. MIT Project NANDA, The GenAI Divide: State of AI in Business 2025 (Challapally, Pease, Raskar, Chari), pp. 3, 6–7. Figures are directional, interview-based, and self-reported. Attribution is MIT Project NANDA and never McKinsey. Canon
NANDA-95-FAILURE, verified 2026-07-11. - ~90 days mid-market vs 9+ months enterprise. Same report, p. 7. Top performers, not cohort averages. MIT defines enterprise as >$100M revenue and mid-market as the $20M–$100M band. Canon
NANDA-90DAY-MIDMARKET, verified 2026-07-11.