Turnfunnel

Traditional funnels assume users move between pages and click events. In conversational products, a user can state an intent, clarify constraints, invoke tools, correct the agent, abandon temporarily, return later and complete a task without changing screens. The supplied research confirms two early conversation-intelligence products: one clusters production conversations and proposes prompt changes, while another tracks timelines, intents and outcomes for products above 1,000 conversations per month. It found no confirmed product centered on product-manager funnels across conversation state. Turnfunnel ingests only authorized, purpose-limited events and minimizes raw conversation retention. Every proposed intent, confusion, repair, abandonment and task state records the supporting turns, classifier version, confidence, alternatives, reviewer correction and expiry. A label is an analytic hypothesis, not a fact about a user, emotion, competence or vulnerability. Silence may be a pause, success elsewhere, channel switch or failure; it is not automatically drop-off. Task completion should come from a verified destination event or explicit user confirmation whenever possible, not a model judging its own answer. Instrumentation, inferred state, reviewed state, funnel aggregation, product hypothesis, approved change, deployment, exposure, measured outcome and causal conclusion remain separate. The product prohibits individual employment, credit, health, eligibility or manipulation decisions, and never opens or deploys a product change without accountable human review.

A product, growth or conversational-experience leader operating an AI-native application whose core user journey occurs through dialogue and tool use rather than screens.