The first time someone running a 40-person business asked me "so what should we actually do with AI?" I gave them a 20-minute answer. It was org-chart thinking — the kind of answer I'd have given at the last three companies I was inside. Discovery cycles, use case matrices, cross-functional working groups, a phased 12-month roadmap.
They let me finish. Then they said: "yeah, but which one thing on Monday?"
That's the decision gap. Not lack of ambition, not lack of vendors — the exact opposite. Every AI conversation at their scale produces more options, not fewer. Nobody hands them a Monday.
The shape of stuck
I stopped counting when I hit ten variations of the same problem. A managing partner with 30 browser tabs open, one per vendor demo they'd sat through this quarter. A CEO whose head of ops kicked off three AI initiatives in a year and had all three go quiet by month three. An owner who couldn't get their leadership team to agree on what "AI" even meant — half were thinking chatbot, a quarter were thinking spreadsheet macros, and one guy was still talking about self-driving cars.
None of them had a technology problem. They had a decision problem. And nothing in the market was helping them make one.
The obvious fixes I watched fail
Hire a consultant
The honest market for "tell me what to build and then build it" is small. Most of what I watched people buy was either 90-day strategy work that shipped a slide deck, or a vendor pitching their own product wearing a consultant's jacket. Neither one leaves you with a working system, and both eat a real budget you now don't have.
Train the team
Training teaches people to use AI. It doesn't teach a business where to put it. I've seen owners drop $8K on a prompting course for their leadership team and still not know what to do first. The training was fine. The problem was they went in with the wrong question.
Just pick something
This is the advice individuals get, and for individuals it works. For a business, picking wrong isn't a wasted weekend — it's months, tens of thousands of dollars, and a team that now believes AI didn't work for them. Which quietly kills the next attempt for a year.
What actually works
When someone asks me the Monday question now, I don't give them the 20-minute answer. I ask three things: what are you already trying to do more of, what part of that is manual, and who on your team already believes AI could help. If those answers overlap on anything specific, that's the first thing to build.
The conversation takes 15 minutes. If it lands, we take a week or two to pressure-test it — score against data readiness, integration complexity, and time-to-value; write a one-page business case a CFO can stress-test in ten. That's the whole thing. Not a 90-day strategy. The smallest engagement that leaves you with a decision.
The half that doesn't make it
A meaningful share of the time, in my experience, the honest answer at the end of this is "not yet." The data's too messy. The leadership team isn't aligned. The specific use case they want is the exact one AI is worst at right now. That's a real output — probably the most valuable one. "Not yet" saves an SMB the $50K–100K they were about to spend proving it wouldn't work, and hands them a specific list of what to fix before trying again.
The businesses I've seen get unstuck don't have more ambition than the rest. They just stopped trying to evaluate every option and started asking a smaller question: what is the one thing on Monday.