95% of AI projects fail. Here's what the 5% that actually move EBITDA have in common

A few months ago I posted here looking for search funds to work with. Here's what changed, and what we've learned about AI in operations since. Back then we were building an AI lab focused on small businesses. Our focus has since sharpened, and I wanted to come back and share what we have learned, because most of it is relevant to anyone here thinking about value creation after close. Quick context. My co-founder Mike was at Accenture as a managing director leading supply chain and operations, after years running operations at Mars. I am a serial founder with a few exits, my last AI company was acquired by a US BPO. We have a team of senior AI engineers behind us. Today we build practical AI agents inside the operations of mid-market companies, roughly $50M to $500M revenue, mostly PE-backed. The pattern we keep seeing: almost everyone selling AI right now is either an engineer who does not understand the operation, or a consultant who understands it but cannot ship. The first brings a hammer and hopes you have nails. The second hands you a deck and a bill. Operators feel this instantly, which is why 95% of AI projects failed (MIT, 2025), and the 5% that worked were the easy cases. What actually works in the mid-market, from our deployments: 1. Diagnose before you prescribe. We spend two weeks mapping the operation into###-###-#### cases scored by value and ease before building anything. Prescribing without diagnosing is the single biggest reason these projects die. 2. Get paid from realized value, not hours. Two-week diagnostic, first case live two weeks after, and we earn from the measurable bottom-line improvement. If it does not land, you do not pay. 3. Speed beats scope. A case shipped in two weeks that moves EBITDA beats a six-month transformation that dies on contact with reality. Why I am posting. This community is full of operators and MBAs who are one handshake from the PE firms and mid-market operators we want to learn from. One thing would be hugely valuable: intros to PE operating partners or operators at mid-market companies ($50-500M, US or UK) where this might create real value. If that is you, or you know someone, I would love a conversation. And to pay it forward, since the AI noise here is loud: if you are weighing an AI play inside a business you are buying or running, DM me and I will give you an honest read on what is real and what is theater. No pitch.