After ten-plus AI product launches, patterns repeat: PMF appears when one job gets measurably faster for a defined user, not when demos go viral. Here is what worked, what flopped, and what we recommend founders measure. Ten launches taught us that PMF whispers in cohort charts before it shouts on Product Hunt — demos are loud, retention is honest. Failed launches taught us to kill features faster — sunk cost in demo code should not block roadmap truth.
Signals of real PMF
Users return weekly without prompts. They refer peers unprompted. They pay or expand usage before you perfect secondary features. Unprompted peer referral is the strongest signal — users stake reputation recommending tools to colleagues. Paying before feature completeness indicates job urgency, not novelty curiosity. Navamsha retention improved when daily insights referenced user dasha — generic horoscopes churned.
AI-specific: low edit distance on outputs, high citation click-through, support tickets drop on covered topics. Vanity: high trial, zero D30, social shares from non-users. Low edit distance on AI outputs correlates with retention better than NPS alone — users who rewrite everything leave. Citation click-through shows trust building over time. GTM tools retained when tied to CRM workflow — standalone scoring dashboards did not.
Vertical vs horizontal
Astrology, healthcare CDS, agri advisories, and GTM scoring reached trust faster than generic chat-with-PDF. Domain data moats and vocabulary matter. Generic chat-with-PDF lost to vertical tools with domain vocabulary and pre-built integrations — astrology, agri, clinic notes won trust faster. Horizontal products need embedding in existing daily workflows or extraordinary distribution. Client launches that pivoted after golden-set failure recovered faster than those that scaled broken demos.
Horizontal copilots need existing workflow density — embed in CRM, inbox, EHR — or die as novelty. Domain data moats compound — mandi prices, classical texts, closed-won GTM outcomes improve models competitors cannot copy quickly. Investor updates with cohort charts beat feature lists — PMF narrative needs numbers.
Implications for roadmap
Double down on one metric and one persona before expanding languages or agents. Kill features that score well in demos but not in cohort retention. Expand languages only after one persona shows D30 retention — premature localization spreads eval thin. Kill demo-popular features that cohort analysis shows zero repeat usage. Vertical focus simplified eval — horizontal products needed ten times more golden cases for same confidence.
Build own products first so client advice is evidence-based — not theory. Internal products prove advice — we recommend patterns Navamsha and client launches validated, not blog theory. Build-own-products-first philosophy exists because we saw PMF patterns repeat across categories.
What to do next
Start with one measurable outcome — latency, accuracy, adoption, or revenue — and design your stack around that constraint. GenOrbit ships production AI systems across healthcare, astrology, fintech, and SaaS; if you want a second pair of eyes on architecture or delivery, we are happy to talk.
Related: /services/product-strategy-mvp · /blog/ai-features-user-retention · /blog/why-build-own-products-first