Feedle.ai: Lessons in Market Validation

The Challenge

I set out to prove I could build and sell a SaaS product completely on my own. The idea was a personalized "For You" feed that any website could enable by pasting a single JavaScript snippet. If it worked, readers would stay longer, publishers would show more ads, and I would turn the first paid pilots into a real company.

Approach

I spent spring 2024 coding nights and weekends with two freelance engineers. Within six weeks we had an MVP and installed it on science outlet N+1 (nplus1.ru). Early A/B testing looked encouraging: whilу only 1% of visitors clicked the floating button (with really bad design), yet average session time still grew from 11 min 13 s to 11 min 29 s. My confidence stayed high, so I doubled down.

Analytics revealed stark differences between user segments

During the summer I redesigned the interface, refined the ranking algorithm, and set an audacious goal—triple every metric. I launched $200 worth of Reddit and Google ads and personally emailed 250 editors through Apollo. One thousand people visited the landing page, 3 requested demo, nobody bought. That was the first serious doubt about product-market fit.

Money was running out by August, so I stopped paying contractors and switched to solo development. Using Cursor and Claude 3.5 I rewrote the parser, added "Likes", built a modest analytics dashboard, and killed dozens of bugs. Shipping alone with AI felt great, but the market stayed silent. No matter how I repositioned the value proposition, publishers preferred spending nothing to earn "maybe" 3% more time on site.

By October the numbers were clear: engagement uplift plateaued at 2–3% overall, zero paying customers, and roughly US $4,000 of personal savings gone. I paused development to rethink the business.

Results

The widget correlated with greatly lengthened sessions for the tiny group who used it — people who clicked once stayed six times longer. Yet the overall lift across all traffic never broke 3%, and that was not compelling enough for editors to reach for a credit card. Commercial traction was therefore exactly zero sales.

Key Learnings

• Sell first, build later. A wait-list with real email addresses is a better validation than optimistic spreadsheets.
• Optimism is not a strategy; reality rarely triples metrics just because you wish it.
• Do not open a legal entity before revenue—an invoice with deferred payment is enough for the first deal.
• Coding with AI was exhilarating, but without customers it remained a hobby.