Retention With the Cohort Intact

Retention becomes useful when it stops being a single percentage and starts behaving like a reconciliation. The operating question is not whether revenue rose. It is what happened to the customers and recurring revenue that existed at the start of the period.

The companion metrics chapter defines the measures. This guide turns them into a monthly or quarterly close that another person can reproduce; the retention bridge is the calculation companion.

Freeze the cohort before looking at the answer

Choose an opening date, a closing date, and one recurring-revenue definition. Export every paying customer active at the opening date with an account ID and opening recurring revenue. That list is the cohort. Do not add accounts acquired later. Do not remove accounts that churned; their closing revenue is zero. Record mergers, account splits, currency changes, and billing migrations in an adjustments column rather than quietly rewriting history.

This discipline matches the method Snowflake describes in an SEC filing: it defines a starting population, compares revenue from that population in a later period, and leaves non-consuming cohort members in the calculation at zero. That is an issuer-defined metric, not a universal accounting standard, but the fixed-cohort logic is sound (Snowflake S-1).

Calculate three different answers

For the same cohort, calculate:

• Logo retention = closing active logos / opening logos.
• Conventional GRR used here = (opening recurring revenue - churn - contraction) / opening recurring revenue. Expansion is excluded.
• NRR = (opening recurring revenue - churn - contraction + expansion) / opening recurring revenue.

Keep one-time services, implementation fees, taxes, and revenue from newly acquired logos out of all three. If an old customer buys a genuinely recurring add-on, that is expansion. If a former customer returns, state a win-back policy and apply it consistently; do not improvise based on whether the quarter needs help. Issuers sometimes use differently composed metrics under similar labels, so compare formulas before comparing percentages.

Worked hypothetical: the healthy-looking leak

Worked hypothetical — not a benchmark. A January cohort contains 10 customers and $100,000 of MRR. By December, two customers representing $12,000 have left, one account has contracted by $8,000, and two surviving accounts have expanded by $30,000. Eight logos remain.

Logo retention is 8 / 10 = 80%. Conventional GRR is ($100,000 - $12,000 - $8,000) / $100,000 = 80%. NRR is ($100,000 - $12,000 - $8,000 + $30,000) / $100,000 = 110%.

Now suppose sales added $70,000 of MRR from new customers. Ending company MRR is $180,000, but cohort NRR remains 110%, not 180%. Passing that $70,000 through NRR would turn acquisition into retention and erase the question the metric is meant to answer.

Use the gap as a diagnosis

NRR above 100% can coexist with weak GRR. Here the 30-point gap says expansion from a few survivors is masking $20,000 of gross loss. Ask which accounts expanded, whether that expansion repeats, and whether the churn came from a segment you still intend to serve. Logo retention adds another clue: losing many tiny accounts is different from losing one dominant account, even when revenue loss matches.

Public issuers do not all calculate retention identically. Workday, for example, describes a version based on annualized subscription and support revenue and includes upsells, cross-sells, price changes, and multi-year renewal increases. Treat every published retention figure as an issuer-defined claim until you read its denominator and cohort rules (Workday annual report).

The close checklist

Before circulating the sheet, confirm: cohort IDs are frozen; churned accounts remain at zero; opening and closing revenue use the same definition; new logos equal zero in the retention bridge; expansion is recurring; contract migrations are documented; and logo retention, conventional GRR, NRR, and the GRR-to-NRR gap appear side by side.

Limitations: a small cohort can swing on one account, usage revenue may reflect consumption rather than contracted value, and annual cohorts hide timing inside the year. Use account-level rows and segment cuts, but do not choose segments after seeing which one flatters the result.