Buying AI is not the same as buying a familiar software category. The language is less settled, capabilities move quickly and demonstrations can make weak products look polished. A sound sourcing process therefore starts before the supplier list. It starts with a clear description of the decision, the workflow and the outcome the organisation actually needs.
At AIPVA, we prefer to turn that need into a small set of testable requirements. What data does the solution need? Which systems must it work with? Where does a human need to review or approve an output? What would make the result commercially useful rather than merely interesting? These questions make it easier to separate real capability from broad claims about artificial intelligence.
Supplier evaluation should also create evidence that can survive internal challenge. A useful scorecard combines functional fit, implementation effort, security, data handling, operating model impact and commercial value. For AI products, it should also examine how outputs are generated, how the system behaves when information is missing and what controls exist around accuracy. A live use case or structured proof of value is often more informative than a long feature checklist.
Procurement still has an important role after the technical evaluation. Pricing models can be difficult to compare because vendors charge by user, task, token, document, workflow or outcome. Contract terms around data use, model changes and service performance also deserve close attention. The cheapest headline price may not be the lowest total cost once usage grows.
The goal is not to slow AI adoption. It is to help teams move with confidence. A well designed sourcing process gives stakeholders a shared language, makes trade-offs visible and creates a record of why a supplier was selected. That matters when the technology evolves, when usage expands and when the organisation needs to revisit the decision later.
In practice, the most useful sourcing exercises also include a short internal calibration before suppliers are scored. Evaluators review one sample response together, discuss what good evidence looks like and agree how to treat gaps. That small step can prevent large scoring differences later. It also gives procurement a clearer record of how the evaluation framework was applied, which is valuable when stakeholders ask why one supplier progressed and another did not.
That discipline also makes later supplier reviews easier because the original criteria and evidence remain understandable months after the award.