The Fixed-Cohort Retention Review Framework

FIXED-COHORT RETENTION REVIEW — original Growthcraft synthesis
Decision and proposed action:
Starting snapshot / ending snapshot / timezone:
Currency and constant-currency conversion policy:
Account grain / identity mapping version:
MRR definition: recurring items, discounts, trials, delinquency, usage:
Segment membership frozen at which time:
Source exports / extraction cutoff / restatement version:

GATE 1 — DEFINE (finance + analytics)
Do both snapshots implement the same definition?
Evidence / unresolved differences / owner:
GATE 2 — RECONCILE (data owner)
One row per account, explicit start and end values, no unknown converted to zero:
Start S / end E / retained gross G / churn C / contraction D / expansion X:
S - C - D + X = E; G = S - C - D
Outside-cohort ending MRR (excluded; may be new or reactivated):
Observed aggregate totals and explained residual:
GATE 3 — INTERPRET (growth + customer success)
GRR = G/S; NRR = E/S; logo retention = surviving starting accounts / starting accounts
Undefined denominators / concentration / intra-period activity hidden by snapshots:
What changed vs why it changed:
GATE 4 — ACT (named approver)
Status: HOLD / INVESTIGATE / APPROVE BOUNDED ACTION
Specific evidence-supported action, accountable owner, review date:
Customer-quality and economic guardrails:
Missing facts that could reverse the action:
No benchmark, causal claim or automatic spending recommendation implied.

Use this before acting on a retention headline

A rising NRR can coexist with substantial customer losses. This original operational framework separates metric definitions, arithmetic reconciliation, interpretation and approval. It is for growth, customer-success and finance teams preparing a retention review. It is not a causal model, an accounting standard or an automated diagnosis of churn.

Use it with two complete account-level MRR snapshots in a single currency and a documented period. Do not use it for usage-only cash receipts, mixed-currency exports without a conversion policy, or a campaign-impact claim. An event-based movement report needs its own reconciliation and can legitimately differ from this endpoint convention.

Inputs and accountable owners

The data owner supplies dated exports and account mappings. Finance owns recurring-revenue eligibility, currency, discounts and reconciliation. Analytics owns cohort membership and formulas. Growth and customer success propose an investigation; the budget or commercial approver owns any intervention. Bring extraction timestamps, timezone, snapshot definitions, missingness checks and the vendor report settings you intend to compare. Anonymise IDs and keep customer-level source records inside approved systems.

Four non-compensating gates

1. Define: freeze membership at the starting snapshot. Record the definition version and treatment of trials, past-due items, discounts, usage charges and account migrations. Hold if definitions differ; a high NRR cannot compensate for incomparable data.
2. Reconcile: require unique account rows and explicit nonnegative endpoints. Missing data is not zero. Confirm the account totals against approved aggregate exports and explain every material residual. Exact equality is expected for the synthetic fixed-precision example; production materiality must be approved rather than invented.
3. Interpret: separate gross retained MRR, ending cohort MRR and outside-cohort MRR. Review logo retention and account concentration alongside revenue. List changes that endpoint snapshots cannot see. Do not infer reasons or causality from the bridge.
4. Act: choose hold, a scoped investigation or a separately approved bounded action. Record owner, review date, cost, guardrails and the evidence needed to reverse the decision. No generic retention percentage is an approval threshold.

Worked example and decision

Synthetic EUR monthly-normalised amounts. Account A changes 100 → 140, B 200 → 150, C 100 → 0 and D 0 → 90. The starting cohort is A/B/C with MRR 400. Churn is 100, contraction 50 and expansion 40; ending cohort MRR is 290. Gross retained MRR is 250. GRR is 62.5%, NRR 72.5% and logo retention 2/3, approximately 66.67%. D's 90 is outside the cohort, not evidence of retention.

The full ending base is 380. Dividing that by 400 would produce 95%, but it mixes an outside account into the retention numerator. Suppose exports reconcile but cancellation context and approval are absent. Mark the arithmetic gate complete and the action HOLD: ask customer success to investigate C's exit and B's contraction, with an owner and due date assigned by the team. Do not prescribe a discount or spend increase. There is no observed explanation yet.

Failure modes and limits

• Changing IDs can mimic both churn and acquisition; reconcile migrations before calculation.
• An account that cancels and returns between endpoints appears retained. This is not uninterrupted retention.
• Freezing plan membership at the end selects survivors; freeze the intended segment at the start.
• Reactivation of an account with zero starting MRR stays outside this cohort.
• Currency changes, price increases and seat changes can alter MRR without the same change in customer value.
• Snapshot gross retention nets within-account changes and is not necessarily identical to a movement-based vendor report.

Method background: ChartMogul GRR documentation, updated 11 September 2026, describes exclusions and segment settings. Stripe Billing analytics documentation, accessed 21 September 2026, documents configurable billing definitions and downloadable subscriber snapshots. This original snapshot method is not a vendor-certified reproduction of either platform.