When the Vendor's Definition of "Done" Doesn't Match Yours

The project met the spec and missed the point. That gap was written into the contract long before anyone felt it.

Most mid-market organizations buy AI capability through a vendor relationship. That is not a criticism — it is usually the right call. You do not have a machine-learning team sitting idle, and building one to run a single initiative would be its own kind of waste. So you buy.

But buying has a consequence people underweight. When you buy the capability, you also, in part, buy the vendor’s definition of a successful outcome. It gets encoded in the statement of work — the scope, the acceptance criteria, the milestones that trigger payment. And a vendor’s definition of done is an honest, reasonable thing that is not the same as yours. It is the version that is buildable, scopeable, and defensible in a status meeting.

So the contract gets signed. The implementation begins. Everyone is professional and the milestones get hit. And roughly six months in, the leadership team looks at what was delivered and realizes it did exactly what the SOW said — and does not do the thing they actually needed.

That is the quiet failure mode. Not a blown deadline, not a vendor who underperformed. A vendor who performed precisely to a definition of done that was subtly, expensively, not the same as the outcome the business was picturing. Everybody can point to the contract. Nobody got what they wanted.

The reflex fix is to write a tighter SOW. More detailed acceptance criteria, more milestones, more specification. In my experience that mostly produces a more detailed version of the same miss — because the problem was never the precision of the spec. It was that the two sides never sat down and agreed, out loud, on what success actually looks like in the operation before the language got locked.

Getting to a shared definition of done is not a procurement tactic to run before signing. It is the work. It is the single most leveraged conversation in the entire engagement, and it costs nothing but the discomfort of admitting, early, that you and your vendor might be picturing two different things. That conversation is cheap in month zero and nearly impossible to have in month six, once the contract is the thing everyone is defending.

Worth asking, before the next SOW gets signed: do you and your vendor actually agree on what “done” means — or have you just agreed on what gets you to a signature?