What Working Problem Interviews Look Like for AI SaaS

AI SaaS adds a specific complication to problem interviews. Customers have strong, often confused opinions about AI in general - what it can do, what it cannot do, what it should cost. A working interview cuts through that and gets to the actual workflow underneath. Here is what that looks like.

It Starts Without the AI Frame

Bad AI SaaS interviews open with "tell me about how you use AI in your work." You get an ideology answer. The customer recites their position on AI. None of it is data.

Good AI SaaS interviews open with the workflow. "Walk me through what happens when you do X." The AI question never appears in the prompt. If AI tools come up in the answer, great - you learn how they actually use them. If they do not, that is also data: this workflow is not AI-touched today, which is itself a useful signal.

It Surfaces Where AI Already Failed

Most professional users have tried AI tools, found them not-quite-right, and quit. Working interviews drag this out. "Have you ever tried an AI tool for this?" almost always produces a story. The story includes which tool, why they tried it, what the failure mode was, whether they would try again.

This is some of the highest-signal data you can collect. The customer has already done the hard part - tested AI in their workflow - and they are telling you exactly where the existing offerings broke. Their disappointment is your roadmap.

It Distinguishes Hope From Use

AI SaaS interviewees often describe what they hope AI will do for them in the future. That is not what you came for. A good interviewer redirects: "I am more interested in what you actually do today. Can we go back to the workflow?"

The future hopes are a side stream. They are useful for understanding what the customer thinks AI will do, but they predict almost nothing about what they will adopt. Adoption is predicted by present pain, not future excitement.

It Captures the Trust Story

AI SaaS specifically: every customer has a trust story. Where do they trust AI output and where do they not? In what step of the workflow do they always check the output by hand? When did the AI produce something wrong and what was the consequence?

Good interviews surface this. "Have you ever caught an AI output that was wrong? What happened?" The trust boundary the customer drew that day is exactly the boundary your product has to respect. Cross it and the product fails. Stay inside it and you are useful.

It Asks About Cost the Right Way

Asking "would you pay for an AI tool that does X" produces fantasy data. Asking "how much do you currently spend on the workflow this would replace" produces grounded numbers. Hours, contractor invoices, existing tool subscriptions, employee time. The answer is what your AI product can compete against.

Working interviews stay in the past tense for cost. The customer remembers what they spent. They do not reliably predict what they will spend.

It Tolerates Skepticism

Some customers are AI-skeptical. The temptation is to argue or convert. Working interviews do neither. The skepticism itself is data. Why are they skeptical? What evidence would change their mind? Have they ever been pleasantly surprised?

The skeptics often turn out to be the highest-value customers if you cross their bar. Capture their objections in their own words. Those objections are exactly what your landing page has to answer.

It Captures AI-Specific Vocabulary

"Hallucinated." "Cooked the result." "Made up a citation." "Got it almost right." "Sounded confident but was wrong." "Better than I expected for the easy stuff." These are the phrases customers use about AI tools. They are gold for your messaging.

The non-AI version of vocabulary capture matters too, but for AI SaaS this is sharper. The audience already has a shared dialect about AI failures. Use it. Your landing page should sound like a customer talking to another customer about AI, not like a vendor.

It Ends With a Demo Invitation, Not a Demo

If the interview goes well, the customer often asks to try whatever you are building. Say yes, get their email, and follow up when there is something to show. Do not pull up a demo during the call. Once you do, the conversation becomes a demo reaction and you lose the discovery shape.

For AI SaaS, this is even more important than for other categories. Demos warp the conversation toward feature reactions. Working problem interviews keep the conversation about the customer's actual workflow until the very end.

What These Interviews Produce

By the end of fifteen well-run AI SaaS problem interviews, you have: a workflow map, the trust boundaries, the existing-AI-tool failure modes, the cost data, the vocabulary, the skepticism patterns, and a list of fifteen people who would try a beta. That is the spec for the next product. Build around it.