HackerAI Retention Analysis

Determine whether comparable subscribers leave more often after a change, what the data can establish now, and what evidence or follow-up is still missing.

Workflow

1. Establish the question, intended environment, population, comparison, and time window. Verify the analytics project's timezone and instrumentation coverage. Inspect repository instrumentation first; use the connected analytics and billing tools for current evidence within the authorized scope.
2. For cohort design, outcome definitions, billing reconciliation, and readout wording, consult the relevant sections of the measurement guide. Prefer valid historical randomized assignment. Otherwise use pre-exposure comparability and label the result observational. Preserve original assignment and report actual exposure, crossover, and missing outcomes separately.
3. Produce an immediate experience readout where available, then distinguish mature cancellation decisions, renewal retention, and invoice-linked payment recovery. Use equal follow-up, appropriate user/subscription/invoice units, and uncertainty that accounts for repeated runs per user. Keep subscribers who cancel before their baseline renewal date in the renewal comparison.
4. Lead with early warning, mature increase/decrease, inconclusive, or not measurable. Include both groups' numerators and denominators, absolute differences, uncertainty, limitations, next valid readout, and a concrete recommendation. Payment-failed endings do not rule out dissatisfaction.
5. Save reproducible bounded queries and aggregate results in a dated task artifact. Keep volatile results out of this skill. Update external tracking or schedule follow-ups only when authorized by the user's task.

Boundaries

This is an analysis workflow, not authorization to change routing, prices, allowances, feature flags, customer billing, or source configuration. Missing tables or credentials are evidence gaps, not permission to repair a deployment. Follow the repository's environment and credential boundaries.

Keep customer content out of analytics queries and reports. If the user requests content-based qualitative research, use the separate HackerAI user research workflow and its authorized gateway. Do not silently expand aggregate retention analysis into conversation inspection.