How do you validate an AI-integrated SaaS idea before building too much?

I recently started working on an AI-SaaS product, and like many of us here, I wanted to avoid building something no one really wants.

So here’s the process I followed (and still refining) — curious how you do it too:

1. Start with a painful workflow

Instead of brainstorming a “cool AI use case,” I picked a workflow I personally hate doing manually or one that takes hours to complete.

2. Mock the solution before building it

Before writing a single line of backend code, I mocked the whole experience:

• Landing page with clear problem/solution.

• Fake UI using Figma or Webflow. Recorded a demo as if the product exists.

Then shared it on LinkedIn, indie hacker groups, and a couple of Reddit subs. This gave me:

✅ Early interest

✅ Skepticism (which helped a lot)

✅ The exact objections I need to solve

3. Let users try to pay

The biggest validation for me wasn’t signups, it was:

Are users asking for a trial* or trying to subscribe even when there’s friction?

* Are they frustrated there's no live version yet?

I offered no free plan, just a simple pricing table — and I got people to ask "where can I try this?"

4. Track one key thing: urgency

If people say “this looks interesting” but don’t take any action → it’s a sign to go deeper into the problem.

But if even 10 people express urgency, ask about launch dates, or DM for access → you’re onto something.

5. Don’t overbuild the AI part yet

I use GPT, Claude, etc., but kept the AI layer very shallow in the MVP.

Why? Because:

• Most users don’t care how* it works.

• They care if it saves them time or money.

• Build the wrapper, test the UX, then scale the AI logic.

Curious to hear from you 👇

If you’ve validated an AI-SaaS, or in the middle of it — what worked for you? What signals did you look for?

Let’s swap notes