Problem Interviews Checklist for AI SaaS Teams

The working checklist for AI SaaS teams. Skip the reading, run the steps. Designed for engineering or product founders with limited time and a need to actually finish the round.

One-Time Setup

• Define the user with two filters: a job-or-role and a recent behavior. No filtering on "interested in AI."
• Write three questions on a sticky note: workflow story prompt, prior AI tool experience, trust boundary.
• Open one shared doc for all interview notes.
• Decide the no-pitch rule. You will not describe what you are building during the call.
• Decide the no-AI-frame rule. You will not say "AI" until the customer says it first.
• Decide the no-demo rule. No prototype shown during problem interviews.

Sourcing the Calls

• Build a list of 40 candidates fitting the audience filter.
• Send each a three-sentence message: why specifically them, what you are researching, ask for 30 minutes.
• Do not include "AI" in the subject line. Self-selects for ideology.
• Aim to land 15 calls.
• Send messages in batches of 10 over a few days.

Five Minutes Before the Call

• Open the notes doc to a fresh entry. Date, name, role, audience filter match.
• Re-read the three questions.
• Ask permission to record at the start if you intend to.
• Remind yourself: no pitching, no demoing, no leading with AI.

During the Call

• Open with a low-stakes intro. State you are researching, not selling. Do not mention AI in the intro.
• Ask the workflow question first. Listen.
• After every answer, count to three before responding.
• Use the redirect: "Tell me more about that" or "Walk me through what happened next."
• Capture exact phrases when they say something specific or vivid, especially about AI failures.
• If they ask what you are building, deflect to the end of the call.
• When a model or architecture idea hits you, write it silently. Do not voice it.
• Map the trust boundary explicitly. Where do they trust AI output? Where do they not?
• End on time. Thank them. Ask if they would be open to a follow-up.

Within Ten Minutes After the Call

• Add four things to the notes: trigger event, prior AI tool experience, trust boundary, one thing that surprised you.
• Mark direct quotes about AI failures with an asterisk.
• Note any tool, internal hack, or workaround to look up afterward.
• Move private model/architecture ideas into a separate "hypotheses" doc.

After Every Five Calls

• Reread the last five sets of notes back-to-back.
• Write down repeats: same workflow shape, same trust boundary, same AI failure mode, same vocabulary.
• Write down disconfirmations: assumptions that have not held up.
• If a question keeps surprising you with weak answers, replace it.

After Fifteen Calls

• In one sentence each: who is the user, what is the workflow, where is the trust boundary, what is the cost of the existing approach.
• If you can answer cleanly, stop interviewing. Pick a model and ship the smallest version.
• If you cannot, decide what is missing: tighter audience, different questions, or a coherence problem.
• Pick the next move within a week. Do not let the round drift waiting for the next model release.

Anti-Patterns to Watch For

• You opened with "how do you use AI." Reset on the next call.
• You demoed the prototype. Reset.
• You only talked to AI enthusiasts. Recruit some skeptics.
• You wrote paraphrases instead of quotes. Quotes carry the data.
• You did three calls and called it validated. Get to ten.
• You did thirty calls waiting for the model landscape to stabilize. The bottleneck is no longer information.

Print and Use

Pin this list visibly during the round. Re-read it before each batch of calls. Most AI SaaS interview failures come from forgetting one of these steps in the moment, particularly the no-AI-frame rule. The checklist exists for the moment.