AI Procurement Is Broken — Here's How to Fix It

Your procurement process was built to buy software that does the same thing every time. AI vendors do not sell that, and the mismatch produces two failure modes: procurement blocks a good vendor for failing to answer questions that do not apply, or waves through a bad one because the questions it does ask are irrelevant. The standard questionnaire asks about uptime, data residency, and support tiers. All still relevant. None of them touch the things that actually determine whether an AI vendor is safe to depend on: what happens when the model changes underneath you, what your data is used for, and whether you could reproduce today's output in twelve months. Three questions belong in every AI vendor evaluation, and almost none contain them. What is your model deprecation policy and notice period. Is our data used for training, by you or by your upstream provider. And can we pin a model version — at what cost. A vendor who cannot answer the first question crisply is telling you your system can change without warning.