Problem Interview Examples for AI SaaS Teams (Worth Stealing)

Below are sketched examples that look like real AI SaaS problem interviews. Names and details are illustrative. Steal the patterns.

Example 1: A Founder Researching AI Tools for Legal Ops

Founder asks: "Walk me through the last contract review you ran. What did you do, and where did things get slow?"

Interviewee: "Last Tuesday. Vendor agreement, 23 pages. We have a checklist of 14 standard terms we have to flag. I read through, mark up, send it to legal counsel for the actual decision. The reading-and-marking part is what takes me three hours."

Founder follows up: "Have you ever tried an AI tool for the reading-and-marking part?"

Interviewee: "Tried two of them. One missed an indemnification clause we always flag. Cancelled. The other was fine but did not explain why it flagged things, so I had to read everything anyway to verify. Cancelled."

Founder follows up: "Where would AI be welcome in this workflow?"

Interviewee: "If it could pre-tag the standard terms with citations to the contract text I can verify quickly, it would save me two of the three hours. The actual flag-or-not decision still has to be mine. The legal-counsel decision still has to be theirs."

Notes: Trust boundary - AI welcome for pre-tagging with verifiable citations, not for the flag-or-not decision. Vocabulary: "missed an indemnification clause," "did not explain why," "had to read everything anyway." Cost - 3 hours per contract. Failed tools - two of them, both for explainability reasons.

Example 2: A Founder Researching AI Tools for Customer Support Managers

Founder asks: "What happened the last time a ticket came in and your team was not sure how to handle it?"

Interviewee: "Yesterday actually. Edge case on a refund. Junior agent escalated to me. I had to dig through past tickets to find precedent. Took twenty minutes of my time."

Founder follows up: "How often does that happen?"

Interviewee: "Two or three times a day. Different agents, different edge cases."

Founder follows up: "Have you tried an AI tool for the precedent search?"

Interviewee: "Yeah, the one bundled with our help desk. It surfaces tickets but it lies about the resolution. It made up a precedent once and the agent almost shipped a wrong refund. Now we ignore the AI suggestions."

Notes: Trust boundary - AI welcome for surfacing past tickets, not for inferring resolutions. The hallucinated-precedent incident broke trust. Vocabulary: "lies about the resolution," "made up a precedent." Frequency - 2-3x daily across the team. The tool exists and was rejected for a specific failure mode.

Example 3: A Founder Researching AI Tools for B2B Marketing Operators

Founder asks: "Walk me through how you wrote your last campaign brief."

Interviewee: "Honestly, I used GPT for the first draft. Then I rewrote about 70 percent of it."

Founder follows up: "What did you keep from the first draft?"

Interviewee: "The structure, mostly. Headers, sections, that sort of thing. The actual copy was too generic. It did not know our customer."

Founder follows up: "What would have made it more useful?"

Interviewee: "If it had read our last twelve campaign briefs and our voice guide, the first draft might have been keepable. As it is, I am paying for a structure I could write myself."

Notes: Trust boundary - AI welcome for structural scaffolding, not for voiced copy unless it has access to context. Failure mode - generic output without brand context. The customer is paying for something they do not actually need (structure) and not getting what they would value (context-aware copy). Product positioning information.

What These Examples Share

Three patterns. The founder asks about specific past events, never future hypotheticals. The founder explicitly asks about prior AI tool experience, including the failure stories. The founder maps the trust boundary explicitly - where AI is welcome, where it is not, and why.

Notice what is not in any of them. The founder never opened with "how do you use AI." The founder never demoed their own product. The founder never argued with the customer about AI's capabilities. The conversation stayed in the customer's actual workflow.

Stealing the Pattern

Same shape every time. Workflow opening. Story prompt. Prior-AI experience question with the failure stories. Trust boundary mapping. Notes that include exact phrases.

Run five calls in this shape and the texture starts to repeat. The same kinds of trust boundaries, the same kinds of AI failure modes, the same kinds of vocabulary. That texture is your map.

What Bad Examples Look Like

Bad example: founder opens with "tell me how you use AI." Customer recites position on AI. Founder writes down "interested in AI." Walks away validated. Builds an AI tool. Customer never adopts because the actual constraint was their workflow trust boundary, which never came up.

The difference is conversational discipline. The good examples sound almost ordinary. That is part of what makes them work.