---
title: Cohort Retention Analyzer
category: product
entity_type: skill
price: $15
canonical: https://forgehouse.ai/skills/cohort-retention-analyzer/
lang: en
hreflang_alt: https://forgehouse.ai/tr/skiller/cohort-retention-analyzer/
last_updated: 2026-06-20
---

# Cohort Retention Analyzer

> Group customers by kickoff month and turn retention into a curve, fitting decay, projecting LTV and computing NRR, GRR, Magic Number and Rule of 40 with a survivor-bias guard.

Turns the question 'are the customers who started in month X still with us?' into an answer that is a curve, not a single number. It groups customers by their kickoff month, fits retention decay (linear onboarding + exponential steady-state), projects LTV, and computes the SaaS health vitals: NRR, GRR, Magic Number and Rule of 40, with a built-in survivor-bias guard so departed customers stay in the denominator.

## Use cases
- Answer 'which start-month cohort stayed longest?'
- Build cohort-based LTV projection for LTV/CAC
- Report NRR, GRR, Magic Number, Rule of 40
- Render a cohort retention heatmap on a dashboard
- Find the top 20% of cohorts driving most revenue
- Calibrate retention forecasts with a Brier score

## Benefits
- Avoid inflated retention claims with a fixed-cohort-size survivor-bias guard
- Forecast lifetime value with a hybrid decay model instead of naive linear math
- Track whether your forecasts are actually accurate over time, not just optimistic
- Surface the few cohorts that quietly produce most of your revenue

## What’s included
- Extended PostgreSQL cohort retention view with materialized refresh and indexes
- Python analyzer fitting hybrid decay curves with scipy.optimize.curve_fit
- Recharts heatmap component mapping kickoff month against months-since-start
- Linear vs exponential vs hybrid LTV comparison plus NRR/GRR/Magic Number/Rule of 40 formulas
- Brier-score calibration logging that compares each forecast against later actuals
- Client report section template framing the cohort as a shared community journey

## Who it’s for
For founders, RevOps and analytics teams who need defensible retention, LTV and SaaS-health numbers grounded in cohort data rather than survivor-biased averages.

## How it runs
Retention math lies easily, mostly through survivor bias. Anchoring cohorts to first real payment and locking denominators in SQL, the analyzer fits decay curves, projects LTV three ways and scores its own forecasts.
1. Assigns every customer to a cohort by kickoff month, derived from the date of their first completed payment, so cohorts are anchored to real money rather than signup forms.
2. Builds the month-by-month retention matrix (M0 through M23) with a survivor-bias guard built into the SQL: the denominator stays locked at the cohort's starting size, so customers who left are never silently dropped from the math.
3. Fits the retention curve with a hybrid model: linear decay for the first 3 onboarding months, exponential decay for steady state, then computes the cohort half-life, the number of months until half the cohort is gone.
4. Projects lifetime value per customer three ways (linear, exponential, hybrid) and computes the SaaS health set on top: NRR, GRR, Magic Number and Rule of 40.
5. Logs every forecast it makes against the actual that arrives a month later and scores the gap with a Brier score, so the model's calibration is continuously measured instead of assumed.
6. Isolates the top 20% of cohorts by cumulative revenue (the power-law check) and asks what they share, sector, channel, onboarding intensity, turning the best cohorts into a replication target rather than a vanity stat.

## FAQ
### What do I need to feed it to get the cohort curves?
It works from your customers grouped by their kickoff month with their retention or revenue over time, which is standard subscription data. From that it fits the decay and projects forward, so the input is history, not a forecast you supply.

### Why fit a curve at all instead of just averaging retention?
Because a single average hides the shape: early churn during onboarding behaves differently from the slow steady-state decline. Modeling those two parts separately gives an LTV projection that holds up better than a flat number.

### Will the projection be trustworthy if I only have a few months of data?
A projection is only as defensible as the history behind it, so thin cohorts give wide, uncertain curves. It computes the vitals like NRR and the Rule of 40 from what exists, but it cannot invent maturity you have not lived yet.

## Price
$15, one-time, no subscription. VAT included.

Related guide: [AI Google Ads and Meta Ads management](https://forgehouse.ai/guides/ai-google-ads-management/)