← Free resources
Twenty questions for an AI vendor
Most AI vendor pitches survive the demo and fail in month three. These twenty questions are what we would ask if we were buying, grouped by the failure they are designed to surface.
- Time to use
- 45 minutes in a first call
- What you end up with
- A vendor conversation that surfaces operational reality instead of capability claims.
On what actually exists
- What of this is running in production today, for whom, and since when?
- Can I speak to a client who has been live for more than six months?
- What did you build, and what are you reselling under your own name?
- Show me the thing failing — an error state, a bad input, a rate limit.
- What did you ship last month?
Listen for specifics with dates. "We work with leading enterprises" is a sentence that survives having no clients.
On ownership and exit
- Who owns the data you process, and where does it live?
- Who owns the model outputs, and the prompts?
- If we leave in month four, what do we keep and in what format?
- What is hosted in your account versus ours?
- Which third-party vendors are in the path, and what do they cost?
Question eight is the one that reveals lock-in. A vendor who has thought about your exit has thought about your interests.
On the work when it goes wrong
- What breaks most often, and how do we find out?
- Who is on the other end of an escalation, and in which time zone?
- What is the approval gate before an agent writes to a real system?
- What happens when the model changes underneath you?
- What have you got wrong for a client, and what changed after?
Question fifteen is the most useful question in the list. A vendor with no answer either has no operating history or is not going to be straight with you.
On cost and scope
- What drives the price up — seats, volume, integrations, or people?
- What is explicitly out of scope?
- What does month thirteen cost, once setup is done?
- Which of your costs are pass-through, and do they float?
- What would you tell us not to buy from you yet?
Our own answer to seventeen is on the cost breakdown for AI lead generation: scope drives the number, and a price without a scope is marketing.