Step 1: Separate advisory from delivery
The first cut is the most important. Many firms that sell AI help produce a strategy deck and a roadmap and then hand you back the hard part: actually getting something into production. That is advisory. A delivery partner is judged by what runs in your environment after the engagement, not by the quality of the slides. Ask any candidate directly what they will leave behind that is live and used. If the honest answer is a document, you are buying advice, which may be fine, but do not mistake it for delivery.
Step 2: Ask for a use case they took to production
Request a concrete example: a use case the firm took from idea to a workflow handling real decisions, with a result someone in the client's finance function would recognize. Push past the demo. What was the baseline, what did the AI change, how was the value measured, and how long did it take. A partner who has actually shipped will have these numbers and the scars that come with them. One who deals in potential will redirect to capability statements and frameworks.
Step 3: Test whether they govern what they ship
Shipping AI that nobody can control is a liability, not a win. A delivery partner worth hiring builds the controls in as they build the workflow: who approved the use case, what data it can touch, what it is doing in production, and an auditable record of its decisions. Ask how they would let you prove to a regulator or a board how a given AI decision was made. If governance is described as a later phase or someone else's job, the production system they leave you will become your problem the first time it misbehaves.
Step 4: Check who owns the result when they leave
A good engagement makes you more capable, not more dependent. Establish up front what you own at the end: the use case, the control layer, the operating knowledge, and the ability to run and extend it without the partner in the room. Be wary of an arrangement where the partner owns the platform, the data plumbing, and the expertise, leaving you renting access to your own AI capability. The durable position is that you own the use case and the controls and can swap the underlying model or the partner without starting over.
Step 5: Match the engagement to your first real outcome
Scope the first piece of work around a single use case you can get into production and prove, not a year-long transformation. A focused delivery, one governed workflow live and measured in weeks, tells you more about a partner than any proposal. It also gives you an attributable result to take to the budget conversation. Expand the relationship from evidence, not from a master plan. If the partner pushes for a large multi-stream program before proving one outcome, that is a signal about whose risk they are managing.
Frequently asked questions
What is the difference between an AI consultant and an AI delivery partner?
A consultant produces strategy and a roadmap. A delivery partner is measured by what runs in your environment afterward. Ask what will be live and used at the end of the engagement; if the answer is a document, you are buying advice, not delivery.
What should an AI delivery partner leave you with?
A governed use case in production that you own and can run without them: the workflow, the control layer, an auditable record of its decisions, and the knowledge to extend it. You should be able to swap the model or the partner without rebuilding.
How big should the first engagement be?
Small and outcome-focused. Scope it around one use case you can get into production and prove in weeks. Expand from a measured result rather than committing to a large program before any outcome is demonstrated.