How to Validate a SaaS Idea Before You Build It With AI

In September 2026 I finally said out loud what our own customer list had been showing for months: most of the companies paying for KashBoss were distributors, companies with sales reps who load goods onto a truck, sell, collect cash and settle at the end of the day. Yet our website, plan pages, articles and ads all spoke to small retail shops. We had been writing for the customer we imagined, not for the one who paid.

It was not the only signal I had misread. New companies signed up, and it felt like demand, but a signup said very little about who would stay. And the first time I asked my assistant how many paying customers we had, it counted test accounts, my own account and companies that had never issued an invoice.

That gap, between the customer you imagine, the people who say yes, and the people who actually pay, is where most software ideas go wrong.

The short answer

To validate a software idea, collect proof that a specific group of people has a painful problem, already spends time or money on it, and will pay you to solve it. Compliments, signups and "active" accounts are not proof. Money, time, effort and repeat use are. The most honest method is a 30-minute conversation about something that already happened to them, not a question about what they would do.

Why AI makes validation more important

AI made building cheap. You can now build the wrong product in three months instead of three years, and it will still be the wrong product. So validation is the step worth spending your hours on before the assistant writes a line.

You would not open a restaurant by asking friends, "Would you eat here?" Everyone says yes, because saying yes is free. Every idea has to answer three questions instead:

1. Who exactly? Not "small businesses." Something like "food distribution companies with five to thirty sales reps."
2. What do they do today? Paper notebooks, Excel files, an old desktop program, a competitor, or an employee whose whole job is to fix the mess. That is your real competitor.
3. What would make them switch and pay? People switch when the pain of staying is bigger than the pain of moving.

Signals that lie and signals you can trust

When we followed eight days of new signups through each step in April 2026, each step lost people: confirming the email, coming back after the first visit, creating a first invoice, and paying. The people who did reach the product created their first invoice in about 17 minutes on average, so the product was fast enough. The drop happened before people reached it. A signup is a click, not a decision.

How to run the interviews

A problem interview is a 30-minute conversation with someone who might buy. You do not pitch. You listen. Ten to fifteen of them tell you more than a thousand survey answers. The trick is to ask about the past. "Would you pay for a system that does X?" invites a polite lie. "Tell me about the last time a sales rep's cash did not match his invoices. What did you do? How long did it take?" invites a story, and stories contain the truth about how often the pain happens, what it costs and what they have tried.

Your AI assistant can prepare the script and keep you from selling:

After five or six conversations, ask the AI to look for patterns, and ask for the uncomfortable answer, because assistants tend to agree with you:

Competitors are good news, not bad news. A crowded market proves that people pay. Read their pricing pages and their bad reviews, and require a source for every claim the AI makes about them.

Pay for conversations, not clicks

We learned the signup lesson again with advertising. An early Facebook campaign sent cold visitors straight to a self-service signup page and brought registrations, but as the signup data showed, a cold signup is not a customer. A different kind of ad, one that opened a chat with us, brought a steady flow of real conversations and several companies asking about annual plans. A business system is not an impulse purchase; people need to ask questions about their own business before they trust you with their accounts.

A conversation is also the cheapest interview you will ever get. Every message someone sends is a free sample of how customers describe their problem. Save them and read them before you write your marketing.

Find out who actually pays, then write for them

We rewrote everything in one day, starting from the distribution cycle: goods go onto the rep's stock, the rep sells, collects payment, takes returns and settles at the end of the day. Now the customer sentence is checked against the list of who actually pays, not against what we imagined. We also stopped saying "shops." A distributor with a large team does not want to be called one.

The same discipline applies to counting. Our first "paying customers" count used subscription status alone. Counting properly meant peeling away layers: status "active" removes trials that never converted, a real agreed price removes test accounts, and real use in the last 30 days removes companies that barely work in the system. Decide what "a real customer" means before you launch, write it down, and measure it the same way every month.

Treat requests as evidence, and never sell what does not exist

Once you have customers, each request is a data point, not an order. Ask: would the next ten customers in your target group want this too?

And never describe a feature as existing before it does. In May 2026 we found that our settings screen had switches for expiry dates and batch numbers, which a "pharmacy" business type turned on automatically, but no screen used them. We held back the pharmacy marketing page until expiry tracking was properly built later that month. It is perfectly fine to say, "We are building that now; can I show you when it is ready?" It is never fine to say it exists when it does not.

Chapter 6 of the book includes the full checklist.

Key takeaways

• A signup, a compliment or an "active" account is not demand. Money, time and effort are.
• Ask people about what already happened, never about what they would do.
• Competitors prove that people pay. Their bad reviews show you where to start.
• Your real customer may not be the one you imagined. Check who actually pays, and rewrite everything for them.
• Building is now cheap, so the expensive mistake is building the wrong thing well.

Frequently asked questions

How do I validate a software idea without building it?

Hold ten to fifteen 30-minute conversations with the kind of person who has the problem, asking about the last time it happened and what they do today. Collect real files, read competitors' pricing and bad reviews, and ask someone for a deposit or commitment. If no one will give money, time or their own data, the idea is not validated.

Are signups a good sign of demand?

Not by themselves. A signup is a click, not a decision. When we followed new signups step by step, each stage lost people, and the loss happened before they reached the product. Measure what happens after signup: first real use, return visits and payment.

What should I ask in a customer interview?

Ask about the past: the last time the problem happened, what they did, how long it took, what it cost and what they already tried. Do not mention your product and avoid "would you" questions, because people answer those politely and untruthfully.