Turn engagement data into clear retention probabilities fast. Adjust coefficients, test scenarios, and benchmark cohorts. See results instantly, then export for team decisions weekly.
Sample cohort records to illustrate typical inputs and outputs.
| Cohort | Tenure | Sessions | NPS | Tickets | Recency | Adoption | Annual | Retention % |
|---|---|---|---|---|---|---|---|---|
| New onboarding | 14 | 2.0 | 6.5 | 2 | 5 | 0.30 | 0 | 42.10 |
| Core users | 180 | 7.0 | 8.2 | 0 | 1 | 0.78 | 1 | 88.40 |
| Price-sensitive | 120 | 3.5 | 7.0 | 1 | 4 | 0.55 | 0 | 61.30 |
| At-risk dormant | 60 | 0.8 | 5.2 | 3 | 18 | 0.20 | 0 | 12.90 |
| Loyal advocates | 365 | 9.0 | 9.1 | 0 | 0 | 0.92 | 1 | 94.70 |
This tool uses an interpretable logistic scoring model to estimate retention probability from multiple signals.
Retention probability converts scattered behavioral signals into a single forecastable metric. When applied to a cohort of 1,000 users, a shift from 55% to 62% expected retention changes the retained count from 550 to 620. That delta affects revenue projections, capacity planning, and customer success coverage. By using a consistent horizon (such as 30, 60, or 90 days), teams can compare periods and segments without mixing definitions.
The calculator blends tenure, engagement, satisfaction, friction, recency, adoption, and plan commitment. Weekly sessions and feature adoption typically lift odds, while days since last login and support tickets can reduce them. A 0–10 satisfaction score provides a standardized sentiment input, and price change captures sensitivity. These signals are common across product analytics stacks and can be computed from event logs, surveys, and billing systems.
Tiers translate probabilities into decisions. High likelihood cohorts can be targeted with expansion prompts and new-feature onboarding. Medium likelihood cohorts often benefit from nudges, in-app education, or proactive outreach. Low likelihood cohorts should trigger recovery workflows: win-back offers, guided activation, or friction reduction. Because thresholds are configurable, you can align tier counts with team capacity and expected contact rates.
Presets provide sensible starting weights, but manual coefficients enable experimentation. For example, if your data shows recency dominates, you can increase the magnitude of the recency coefficient and observe the probability impact. Coefficients act per unit, so a recency weight of −0.10 means each additional dormant day multiplies odds by e−0.10. Scenario testing is valuable during pricing changes, feature rollouts, and onboarding redesigns.
CSV export supports downstream analysis in spreadsheets and notebooks, while PDF provides a lightweight report for stakeholders. Exports include inputs, coefficients, log-odds score, and both raw and horizon-adjusted probabilities. This structure helps analysts validate assumptions, reproduce results, and document how forecasts were produced. For production deployment, fit coefficients on labeled outcomes and calibrate probabilities to match observed retention rates.
It is an estimated chance that a cohort remains active through the selected horizon, based on a logistic scoring model and your selected or manual coefficients.
No. It is a practical scoring approach for quick forecasting and segmentation. For time-to-event modeling, consider survival models and hazard-based features.
Longer horizons usually lower expected retention. The adjustment applies a mild exponential decay beyond 30 days to keep projections comparable across horizons.
Start with 0.75 for High and 0.50 for Medium, then adjust so tier sizes match your team’s outreach capacity and the business impact of interventions.
It depends on your product. Recency and engagement often dominate, while friction and price sensitivity matter in specific segments. Fit coefficients using your historical labels.
Compare predicted probabilities to observed retention by cohort, then calibrate if needed. Track error metrics over time and re-train coefficients when behavior shifts.
Important Note: All the Calculators listed in this site are for educational purpose only and we do not guarentee the accuracy of results. Please do consult with other sources as well.