Free Tool
Leaky Bucket: Trial vs Repeat
household panel · cohort retention · Cinderhaven series #4
Of the households that tried you, how many came back? Leaky Bucket measures trial against repeat to show whether penetration growth is adoption or expensive sampling. At Cinderhaven, a fictional specialty food brand built from realistic test data, two launches repeated at 14% and 51%.
Household penetration can rise every quarter while the business quietly dies. New buyers pour in from a price feature, a display, a demo program; almost none repeat; reach climbs. Then the spend stops, and the bucket, which was leaking the whole time, empties. A distribution or velocity report cannot see this coming, because it counts transactions and a transaction cannot tell a first-ever buyer from a loyal one.
The worked example is Cinderhaven Provisions, a fictional specialty food brand built from realistic test data, with about $25M a year in wholesale invoices, measured here on a seed-locked synthetic household panel. The verdicts are computed from the panel, not scripted, and the math is exactly what the tool runs on real household data.
Two launches: 14% repeat against 51%
One launch, the snack item, drew a big trial spike: nearly a quarter of panel households tried it. On a penetration dashboard it is the star, the one that moves the quarterly reach number and earns the follow-on display budget. The other, a pantry staple, drew less than half that trial and no headlines.
Count who came back and the ranking inverts. Of the households that tried the snack item, 14% repeated within the window. Of the households that tried the pantry staple, 51% did. The star was a promotion: expensive sampling that collapses the moment acquisition spend stops. The quiet one was a brand.
The cost of confusing them compounds. At $4.50 of trade and promotion per acquired household, a 15% repeat rate prices a retained household at $30; a 35% repeat rate prices it at $12.86. The trial-heavy launch pays 2.3x more for every customer it actually keeps, and the ratio worsens each quarter as the easiest households are acquired first.
Repeat counted only on buyers who have had time to repeat
A naïve repeat rate lies about your most recent buyers: they have not had time to come back yet, and counting them as non-repeaters marks a healthy launch down for being new. Leaky Bucket cohorts buyers by the quarter they first bought and shows retention as a triangle: the missing corner is the time that has not elapsed, left blank rather than smoothed over. Every summary repeat number applies a maturity cutoff: only triers whose full repeat window has passed are counted, and the window is a parameter (8 / 12 / 26 / 52 weeks), because a snack bite and a pantry staple repeat on different clocks.
Three views: the verdict (is growth sticking or leaking), the flow (where buyers go after trial: repeat, lapse, or churn), and the cohort triangle. It is the integrity check on Decompose: #3 in the series counts the buyers, #4 says whether they stuck. And it is the number behind the velocity decay that ends in a rationalization conversation: the item that scans well on promotion and drops off after it ends is trial without repeat, visible from week four when repeat is measured.
What you get
A per-item verdict on every launch in the portfolio: trial-heavy and repeat-light, or modest trial with a real consumer franchise. The interventions are opposites: low trial with good repeat is a distribution problem you solve with doors and visibility; high trial with low repeat is a product or value-proposition problem that more trade spend will not fix.
Rising penetration can hide a failing launch
The full argument (why rising penetration is the number that hides CPG’s most dangerous failure mode, and what a trial-heavy portfolio actually costs) is in Trial vs. repeat purchase: when penetration growth is expensive sampling →
Next step: Authorized, on the shelf, and not moving
Get the number for your own data. Seven business days.
The worked example above is synthetic so the method can be shown in full. The offers below run it on your data. The Snapshot credits in full toward the audit.
Private, expiring upload, never email. Mutual NDA before anything moves. Files destroyed within 30 days of delivery, with a certificate. Methods published, tools open source.
Start in writing.
A few minutes by form, no call. Tell me which launch you are about to re-promote, and I’ll tell you what to pull from your panel or loyalty data to learn whether it earned the second promotion, or whether the first one only bought trial. No deck, no obligation.