Will This Curve Flatten? Forecasting a Cohort's Long-Tail Retention
Objective: Given only weeks 0-4 of a new cohort's retention curve, forecast whether it will flatten by week 12 and give a numeric range, then check that forecast against the real outcome.
You're the retention analyst at Allbirds forecasting whether a newly launched reorder-reminder cohort will settle into a healthy repeat-purchase pattern before the quarter closes.
Use the early-week decay rate to forecast a week-12 range, state it as a number, then compare it to the real result once it arrives.
Before you start
What you'll need
Free path (everything below is enough to finish)
Free, sufficient for a 5-point time series and simple extrapolation
The process
2 steps
Step 01 of 02
The lesson's Quick Summary says the goal is a curve that drops fast early then flattens; that flat tail is the signal of product-market fit, not the raw retention number itself.
Week 0 is 100%, week 1 is 52%, week 4 is 33%. The drop is decelerating (48pts, then 19pts over 3 weeks). Forecast a week-12 range using that deceleration pattern.
Procedure
- Enter W0=100%, W1=52%, W4=33% in a sheet
- Compute the drop rate per week for W0-W1 (48 pts/wk) and W1-W4 (6.3 pts/wk average)
- Extrapolate a decelerating drop rate out to week 12, producing a forecast range rather than a single number
Drop rate: W0-1 = 48 pts/wk, W1-4 = 6.3 pts/wk avg (decelerating) Forecast W12 range: 22%-27%, assuming the deceleration trend continues
Healthy
The forecast is stated as a range (22-27%), not a false-precision single number, because only 2 data points inform the slope.
Unhealthy
Linearly extrapolating the W0-W1 drop rate (48 pts/wk) forward, which would predict negative retention by week 3.
What this means
A forecast from an early, steep decay period must use the decelerating segment, not the initial cliff, or it wildly overstates future churn.
So what do I do about it?
| Symptom | Action | Effort |
|---|---|---|
| Only 2-3 weeks of cohort data exist and a stakeholder wants a week-12 number now | Give a range built from the decelerating segment, state the assumption, and commit to revisiting at week 8 | 5 min |
Step 02 of 02
The lesson lists the 6-month survival curve as the clearest single metric for product-market fit: the cohort percentage still active well after the initial drop-off.
Week 12 has now arrived and the real value is 24%, inside your forecast range. Does this cohort's shape support a stronger or weaker product-market-fit read than last quarter's cohort, which flattened at 19%?
Procedure
- Log the real W12 outcome (24%) next to the forecast range (22-27%)
- Compare 24% against last quarter's flattened W12 value (19%)
- Conclude whether the reorder-reminder cohort shows a stronger long-tail than the prior baseline
Forecast range: 22-27% Actual W12: 24% (within range) Prior quarter flattened at: 19% Delta: +5 pts vs prior baseline
Healthy
The actual result lands inside the forecast range, and it beats the prior quarter's flattened value by 5 points.
Unhealthy
Treating the 24% figure as meaningful without a prior-quarter baseline to compare it against.
What this means
A single survival number means nothing alone; it only means something next to a forecast and a prior cohort.
So what do I do about it?
| Symptom | Action | Effort |
|---|---|---|
| A new cohort's flattened value has no baseline to compare against | Pull the equivalent flattened value from the prior comparable cohort before presenting the number | 5 min |
Final deliverable
A forecast memo stating the week-12 range with its assumption, plus a post-hoc accuracy check once the real value lands.
See a reference example
Casper Sleep, W12 forecast vs actual (excerpt) Forecast (from W0-W4 deceleration): 18-23% Actual W12: 21% (within range) Read: reorder-cohort curve confirms early product-market fit; carry this decay model into next quarter's forecast.
Success criteria
You're done when you can:
- States the forecast as a range built from the decelerating segment, not a linear extrapolation of the steepest early drop
- Compares the actual week-12 outcome against both the forecast range and a prior baseline cohort