Six months in: refusals, growth, and what we got wrong

Six months ago we shipped Unrestricted with one long essay and a small waitlist. Today we have a working product, real revenue, and a list of things we got wrong. This is the honest retrospective — the part usually written in private.

What worked

The thesis. People really do want a chatbot that answers their questions. The top piece of user feedback — across survey, support, social — is some variant of "it actually helped." That is a low bar that most of the market is failing.

Privacy-as-architecture. We expected this to be a branding feature. It turned out to be a purchase feature. Enterprise inbound cited it more often than any other attribute.

The blog. This is meta, but: the posts we wrote — benchmarks, mission, comparisons — moved the needle more than any paid channel we tried. Writing honestly about refusals worked better than advertising the fix.

What broke

Onboarding. New users didn't know where the floor was, so a meaningful fraction spent their first ten minutes probing for a refusal we weren't going to give. We've rewritten the first-run state twice.

Pricing. We launched too cheap, signaled budget, and attracted the wrong early cohort. Retention looked worse than it was because the first thousand users were price-shoppers.

Abuse detection. Because we don't store conversations, early signal on a small category of abuse (automation trying to script against us) took longer to catch than it should have. We've since added anonymous, aggregated abuse counters without changing the privacy model.

Every early-stage startup fails in at least one way that reads as a strength in the pitch deck. Ours was "no storage."

Three decisions we'd reverse

First: launching without a team page. We wanted the product to speak for itself. What happened is that a meaningful number of potential users assumed we were sketchy because we were anonymous. A team page is live now. It should have been there at launch.

Second: calling the free tier "free." Users interpreted "free" as "experimental." We'll rename it to "included" and add tier names that imply seriousness.

Third: not publishing our refusal floor earlier. The narrow floor we keep was in our head but not on the site. Some users assumed "unrestricted" meant anything. The floor is now on the About page, explicitly.

What's next

API access in the third quarter. Native mobile apps after that. A public benchmark leaderboard so you don't have to take our word for anything we claim about refusals. And the blog keeps going — one essay a week, in the same direct voice.