Mistake 1: we launched with template selection. The first version of mani had a template gallery. You picked a template, the AI filled it in. This was the easiest thing to build and the worst thing for users. Templates created a false expectation: the output would look like the template, just with their content. When the AI adapted the template to their Brand DNA, the output looked different from the template they picked, and they felt misled. We killed the template gallery in week 3 and switched to fully generative output. Conversion dropped 15% (the template gallery was a familiar UI pattern that users trusted) but satisfaction increased 40% (the output matched their brand instead of matching a template).
Mistake 2: we over-engineered the Brand DNA review process. The initial extraction produced a 28-field profile that users had to review and approve before generating anything. Most users abandoned the review because 28 fields felt overwhelming. They wanted to generate, not to fill out a form. We simplified to 6 key fields (colors, fonts, tone, audience, product, competitors) with an "advanced" toggle for the remaining 22. Completion rate on the review step went from 34% to 78%. The 22 advanced fields auto-populate from the scan and most users never touch them.
Mistake 3: we showed generation credits prominently. The Starter plan has 200 generations per month. We showed a big counter on the dashboard: "173 credits remaining." This created anxiety. Users rationed their generations instead of using them freely. They approved fewer creative pieces because each approval "cost" a credit. The scarcity framing turned a generous allocation into a constraint. We moved the credit counter to the settings page and replaced the dashboard display with "Generate more." Usage increased 60%.
Mistake 4: we priced the Pro plan at $49/month initially. The logic was: $19.99 is accessible, $49 is mid-range, $199 is premium. But $49 was not enough to differentiate from Starter. The feature set was the same (every feature on every tier), and the volume difference (200 vs. 500 generations) did not justify the 2.5x price jump. Pro conversion was 4% of signups. We raised Pro to $99 and increased the volume to 1,500 generations. Pro conversion went up to 11%. Higher price, more value, better conversion. Counterintuitive but consistent with SaaS pricing research: the perceived value must scale with the price, and 500 generations felt like "slightly more than Starter" while 1,500 feels like a real upgrade.
Mistake 5: we did not build multi-brand from the start. The initial version supported one brand per account. Agencies and portfolio operators, our highest-value users, needed multiple brands. Adding multi-brand after launch required refactoring the database schema, the generation pipeline, and the billing system. Three weeks of engineering that would have been three days if we had built it from the start. The lesson: your highest-value customers often have needs that your average customer does not. Build for the highest-value use case from day one, even if most users do not need it yet.
Mistake 6: we launched without platform-specific generation. The first version generated generic "social media" creative. Users had to manually adapt the output for each platform. This was the number one complaint in the first month. "Why does my Instagram ad look like a LinkedIn post?" Because it was a generic post that did not belong to either platform. Adding platform-specific generation rules took two weeks and immediately became the most praised feature. Users did not want generic creative that worked everywhere. They wanted specific creative that worked on each platform.
Mistake 7: we assumed founders would write their own copy and use mani only for design. We built a powerful visual generation engine and a basic copy engine. Users wanted the opposite: great copy was harder for them than great visuals. They could use Canva for design but could not write headlines that converted. We flipped our investment priority: 60% of engine improvement now goes to copy quality, 40% to visual quality. The market told us what it valued. We listened.
Every one of these mistakes cost us time, users, or revenue. None of them were fatal. The aggregate effect of fixing them all is a product that is dramatically better than what we launched. But the fixes only happened because we tracked user behavior closely, listened to complaints seriously, and changed direction quickly. The mistakes themselves are not the story. The speed of correction is.
Mani publishes all significant product changes on the changelog, including the reason for the change and what was wrong before. We are not proud of the mistakes. We are proud of the corrections. And we believe that sharing both is more useful than sharing neither.
The meta-pattern across all seven mistakes is speed of correction, not avoidance of error. We did not avoid more mistakes than our competitors. We corrected faster. The template gallery lasted 3 weeks. The over-engineered DNA review lasted 2 weeks. The credit counter lasted 10 days. Each correction was informed by real user data, not by our intuitions about what was wrong. The process, ship fast, measure fast, correct fast, is more important than the individual decisions.