Gate:Problem framingLens:Conflicted AssumptionsSeat:Executive & BoardType:Buy-Side Decision Audit

Where AI is going wrong before you even start

Vendor category → ??? → your actual problem

Take any AI initiative your institution approved this year and find the first document that named it. If that document was written inside your institution, in words your own people use, this piece is wrong about you.

If it wasn’t, nobody needs to have been careless. The words a problem arrives in come from whoever is already selling a solution to it, and the same words circulate across vendors, analysts and conference agendas — so comparing several vendors does not put distance between you and the framing. Inside the institution, the sponsor who carries the initiative has a position in it too: budget, standing, a name that has already been approved.

Trace the initiative back to its first sentence.

Find the document that first named this as a priority. Is it a whitepaper, an analyst report, or a vendor deck’s “use case” language — or is it a sentence one of your own people said out loud before any of those existed?

Delete every vendor’s name from the problem statement. Does it survive?

Rewrite the problem in one sentence using only words your frontline staff would use. They know the problem; they are rarely the ones who write the problem statement. If the sentence falls apart without a category a vendor supplied — “workflow orchestration,” “agentic automation,” “digital transformation” — you’re describing their solution, not your problem.

Who had a position on the answer?

List everyone who touched that sentence between its first appearance and the approved business case. Next to each name, write what they gain if the initiative goes ahead as named. The list is not an accusation. It shows how many hands the sentence passed through without one of them being free of a stake in it.

Vendor-defined or built in-house, the trace is the same. A vendor’s solution arrives with its category attached. An internal build inherits one too, through the models, frameworks and analyst language the team builds with.

Can you state the problem your team has in one sentence, without a category anyone was selling? If not, that says less about your team than about where the sentence came from: it arrived already named, and no one who handled it was free of a stake. The problem is still unnamed. Name it before the business case is built on the category’s version.

“Because whoever names the problem also shapes which solutions qualify — and nobody who usually handles that sentence is free of a stake in it.”