Problem Interviews for AI SaaS in Plain English

Strip out the framework names. Strip the methodology jargon. Here is what AI SaaS problem interviews actually are, in language a founder can use without studying first.

What They Are

A conversation. You sit down with someone whose workflow you might AI-augment. You ask them to tell you stories about that workflow. You listen. You write things down. That is it.

Why You Do Them

Because building an AI SaaS product is expensive in token spend, engineering time, and customer trust. The cheapest way to find out if you are about to misfire is to talk to a few people first. Five hours of your life learning whether the next three months of building are aimed at a real workflow.

What You Ask

Three things. Walk me through the last time you did X. Have you ever tried an AI tool for this? Where in this workflow do you check the output by hand?

The third one is AI-SaaS-specific. Everything else is follow-up to those three.

What You Do Not Ask

How do you use AI in your work. Would you pay for an AI tool that does X. What features do you want.

All three put the customer in fantasy mode. Stick with stories about specific past events. Real data.

Where You Find People

LinkedIn for B2B audiences. Niche communities for specific roles. Cold email when you can demonstrate you understand their world. Three-sentence message: why specifically them, what you are researching, ask for thirty minutes.

Do not include "AI" in the subject line. Self-selects for ideology.

What You Do With the Notes

After every five interviews, read all five back-to-back. Look for things that came up more than once. Same workflow shape. Same trust boundary location. Same AI failure mode. Same vocabulary. Those are your patterns.

When You Stop

When you can answer four questions in one sentence each. Who is the customer? What is the workflow? Where is the trust boundary? What is the cost of the existing approach?

If yes to all four, ship the smallest version. If not, do five more interviews.

Plain Summary

Talk to people doing the workflow. Ask about specific past incidents. Map the trust boundary. Take notes. Look for repeats. Stop when you can summarize the customer in three sentences. Then ship the AI capability that respects the boundary.

That is the whole practice. Anything more elaborate is decoration.