I Spent ~$500 Promoting an AI SaaS. It Got 30 Signups and 0 Sales.

Hey IH,

I’m a solo developer building ImagineMyHouse, an AI home design product that lets users redesign rooms, visualize floor plans, remove furniture, stage empty spaces, and experiment with different interior styles.

The product is still very early, but I wanted to share the numbers because the gap between “people used it” and “people paid for it” has been more interesting than the building process itself.

So far, I have spent roughly $500 submitting the product to several paid AI directories and launch platforms.

The approximate results:

• 100–200 visitors
• Around 30 registered users
• A little over 20 users tried the generation tools
• 0 paying customers

The sample size is still small, so I’m not treating this as a final verdict on the product.

But it is enough to show that several of my original assumptions were wrong.

Usage Is Not the Same as Purchase Intent

Every registered user currently receives three free generations.

One of the most interesting sessions came from a user working with a black-and-white floor plan.

They:

1. Generated a colored version.
2. Tried a second variation.
3. Edited a specific area in a third generation.
4. Downloaded all three results.
5. Opened the payment modal more than once.
6. Left without purchasing.

From a product perspective, this looked like a successful session.

The user understood the tool, completed the workflow, found the results useful enough to download, and reached the paywall.

From a business perspective, it produced nothing.

That session forced me to separate three signals I had previously grouped together:

• The user can operate the product.
• The user gets value from the output.
• The problem is important enough for the user to pay.

The first two may be true without the third.

It is also possible that three free generations were enough to complete the user’s entire one-time task.

Directory Traffic May Not Be Customer Traffic

Most of the first visitors came from AI product directories.

They successfully produced visits and registrations, but I’m increasingly unsure whether that audience matches the people most likely to pay.

A directory visitor may be:

• Another founder researching competitors
• Someone casually testing new AI products
• An AI enthusiast trying every new tool
• A person with one image and no recurring need

That traffic is not useless. It helped expose bugs, test onboarding, and produce the first real usage sessions.

But I probably made the mistake of treating “AI tool traffic” as equivalent to “home design buyer traffic.”

They are not necessarily the same audience.

Someone remodeling a kitchen, staging a property listing, or producing client presentations has a very different motivation from someone browsing an AI directory.

“AI Home Design” May Be Too Broad

I originally thought of the product as one broad category: AI home design.

Actual usage has been much more specific.

People do not arrive thinking:

I need an AI home design platform.

They arrive with a job:

• Colorize this floor plan
• Remove the current furniture
• Turn this space into a bathroom
• Preview a kitchen renovation
• Stage an empty listing
• Change the lighting without changing the room
• See how one area looks in a different style

These jobs may share the same underlying image-generation infrastructure, but they have different purchase intent and different definitions of success.

This has changed how I think about the product.

Instead of asking, “How do I market an AI home design platform?”, I am now asking:

Which specific workflow has enough urgency and repeat usage to become a paid product?

More Features Are Not Automatically the Answer

As a developer, my default response to weak conversion is to build more:

• More room types
• More styles
• More editing tools
• More landing pages
• More models
• More languages

That feels productive because I can control it.

But adding another feature does not explain why someone downloaded three outputs and still did not pay.

At this point, I think the higher-value work is understanding the gap between download and purchase.

What I’m Testing Next

My next experiments are less about model quality and more about the funnel:

1. Track conversion separately for each workflow and traffic source.
2. Identify which tools lead to downloads, repeated generations, and payment-modal opens.
3. Test whether a small one-time package fits occasional homeowners better than a subscription.
4. Focus positioning on specific jobs rather than the broad “AI home design” category.
5. Localize only workflows that show real demand instead of translating the entire site at once.
6. Continue improving results, but stop assuming better image quality alone will solve monetization.

The question I’m struggling with is what to test first.

Would you:

• Reduce the free allowance from three generations?
• Keep the free allowance but introduce a very small one-time purchase?
• Stop optimizing the paywall and first acquire a much larger user sample?
• Narrow the product toward a repeat-use audience such as real estate agents or interior designers?

I’d especially appreciate feedback from people who have built credit-based AI products.

At what point did usage turn into reliable willingness to pay?

P.S. The product is ImagineMyHouse. Brutal feedback on the pricing and free-to-paid funnel is more useful to me than compliments on the UI.