What growth accounting adds to a cohort table
The cohort analysis tutorial groups customers by the period they were acquired and follows each group by age. It is the right tool for asking whether customers stay and how the curve differs by channel, and the retention and payback guide reads the shapes it produces. What a cohort table does not show directly is the movement in the whole base this month: a business can have improving cohorts and a shrinking base, or worsening cohorts and a base that still grows because acquisition is rising faster than the leak.
Growth accounting answers that second question. It reads activity by calendar period rather than by cohort age, and it rests on one identity: the active customers this period equal the active customers last period, plus the new ones, plus the resurrected ones, minus the churned ones. Because the identity must hold exactly, it doubles as a check on the data, and because each term has a different cause, it turns "growth has stalled" into a statement about which movement changed. The academy exercise on balancing a guitar lesson app's flat quarter with growth accounting practises splitting the quarter's change into these movements, checking that it balances and saying what the bridge shows that the headline missed.
The troubleshooting page on a growth plateau asks for the same split in prose, separating new customers from returning ones before looking for causes. The bridge is the formal version of that split, run every month. It also sits naturally beneath a north star built on active customers: in the language of the KPI tree framework, the four movements are the arithmetic one level below the active base, and the growth fundamentals hub explains how that base relates to the north star the team steers by.
Before you start
- A stable customer identifier. An account, a subscription or a customer record that survives device changes and email updates. The new and returning customers guide sets out the decisions behind "who is a customer", including guest checkouts and households.
- An activity source the business trusts. Orders for a retailer, billing records for a subscription, product events for software. For subscriptions, the subscription measurement guide explains why billing is the source of truth for status.
- History beyond the reporting window. Enough periods to report on, plus a warm-up before them, because resurrection can only be recognised for customers whose earlier activity is in the data.
- The cohort table, already built. The bridge reuses its definitions of acquired and active, and the movements are easiest to explain when the cohorts behind them are to hand.
- A place to build. A warehouse and a SQL client, or a spreadsheet on exported data for a first version; the SQL for marketers guide covers the joins and window functions used below. Google Analytics Explorations include cohort and segment techniques that give a quick first look at activity, but the bridge itself belongs on orders, billing or product events keyed to customers.
The steps
Define an active customer and the period
Write down what "active" means before counting anything, because every movement depends on it. A subscription business usually counts a customer as active in a period if they held a paid subscription at its end. A retailer has no cancellation event, so activity is defined by orders, and the churn and retention metrics guide recommends setting the window from the business's own gaps between purchases, so that someone marked inactive is genuinely overdue rather than between orders. Software can use the core action that marks real use; the app activation tutorial shows how such an action is chosen from the product's own data, and the same comparison works on web product events. Choose a period that matches the rhythm of use, usually the calendar month. If activity is defined over a trailing window longer than one period, note that churn will be recognised only when the window closes, so it lags the moment a customer actually stopped.
Pull one row per customer per period
Build a table with one row for every customer in every period from their first activity to the end of the data, with a flag for whether they were active in that period. Orders or events give you only the active periods, so join them to a calendar of periods to create the inactive rows as well; without them the gaps that define churn and resurrection are invisible. Carry the attributes you will split by on every row: the acquisition channel and segment as they were at acquisition, not as they are today, and for subscriptions the plan and whether a lapse was a cancellation or a failed payment.
Classify each customer's state in each period
For each customer and period, look at three facts: active this period, active last period, and active at any period before last. The rules follow directly. New: active now and never active before. Retained: active now and last period. Resurrected: active now, not last period, but active at some earlier period. Churned: active last period and not now, counted in the period in which the customer is first missing. A customer inactive in both periods is dormant and does not appear in this period's bridge, although they remain in the pool that future resurrection is drawn from. A window function that reads the previous row and a running flag for "ever active before" give all four states in one pass.
Count new, retained, resurrected and churned
Group the classified rows by period and state and count customers. Keep the counts in a narrow table with one row per period and a column per movement, plus the active base at the start and end of the period. This is the bridge. Resist the temptation to merge resurrected into new or to net churn against gains; the whole value of the bridge is that each movement has a different cause and a different owner.
Check that the bridge balances
Two identities must hold in every period. Last period's active base plus new plus resurrected minus churned equals this period's active base. And last period's active base equals this period's retained plus churned, because every customer who was active last period either stayed or left. If either fails, the data is wrong, not the business: duplicate customer records, a customer appearing twice in a period, refunds or reversals that make a first order disappear, or a definition applied differently in two queries. Check also that the earliest periods, where history is missing, are excluded from reporting, as the warm-up in the preparation list above requires. Fix the data before reading any trend.
Compute the rates and the balance of gains to losses
Turn the counts into rates with stated denominators. Retention of last period's base is retained divided by last period's active customers, and churn is the remainder, consistent with the churn rateShare of customers at the start of a period who left during it. definition. New and resurrected are best read as counts and as shares of this period's base. Resurrection can also be read against the dormant pool it is drawn from, which shows whether win-back is getting easier or harder as the pool grows. Finally, divide the gains (new plus resurrected) by the losses (churned). Some teams call this a quick ratio; whatever it is called, its meaning is pure arithmetic: above one the base grew in that period, at one it held still, below one it shrank. This tutorial sets no target level for it, because the right expectation comes from the business's own history; the useful reading is its trend and which term moved it.
Split the bridge by channel and segment
A total bridge says which movement changed; a split says where. Split new customers by acquisition channel, churned customers by the channel they were acquired through and by tenure, and resurrected customers by what brought them back. Early churn and later churn usually have different causes: the subscription measurement guide notes that cancellations in the first periods are where acquisition quality shows, while later ones belong more often to product, price or service. For subscriptions, separate failed payments from deliberate cancellations, as the churn and retention metrics guide defines them, because each has its own fix. Read small segments with care; the guide to reading small numbers explains why a movement of a handful of customers in one segment is rarely a signal.
Run the same bridge for recurring revenue
Where revenue recurs, the same classification produces a revenue bridge. Opening recurring revenue, plus revenue from new customers, plus expansion from retained customers who now pay more, plus revenue from resurrected customers, minus contraction from retained customers who now pay less, minus revenue lost with churned customers, equals closing recurring revenue. The customer and revenue bridges often tell different stories: a flat customer count can sit beside growing revenue if expansion is strong, and a growing count beside flat revenue if new customers arrive on smaller plans. Revenue churn and net revenue retention are defined in the churn and retention metrics guide, and the expansion and renewal marketing guide covers the expansion term for account-based businesses; use their definitions rather than inventing new ones.
Read the trend and give each movement an owner
Read the bridge over several periods, not one, and compare like with like where the business is seasonal. When the active base changes, the bridge says which term moved, and the guide to explaining a change in a marketing metric covers separating a real change from a mix shift. Then give each movement an owner. New customers belong to acquisition and to activation, since sign-ups who never become active never enter the bridge; read new alongside the activation rateThe share of new users or sign-ups who complete the activation event, the action marking first real value, within a set window.. Churn belongs to onboarding, product and service, and early churn by channel goes back to whoever bought that channel. Resurrection belongs to lifecycle work; a win-back campaignMessages aimed at customers who have stopped buying or engaging, sent once they pass their usual gap between purchases. is judged against a holdout of lapsed customers, because some would have returned anyway, and the lifecycle marketing guide covers the win-back stage. Put the bridge into the monthly cohort and economics review that the experimentation cadence guide describes, so that each owner sees their movement next to the others.
The growth accounting bridge: a table
| Movement | Definition | Role in the bridge | Usual owner | Read it with |
|---|---|---|---|---|
| Active, start of period | Customers active last period | The opening balance | Growth lead | The cohort table and the north star |
| New | Active this period, never active before | Added | Acquisition and activation | Activation rate; new customers by channel |
| Resurrected | Active this period, not last period, active at some earlier period | Added | Lifecycle and retention | The dormant pool; win-back holdouts |
| Retained | Active this period and last period | Carried over | Product, service and lifecycle | Retention by cohort age |
| Churned | Active last period, not this period | Subtracted | Onboarding, product, service; acquisition for early churn | Churn by tenure and channel; failed payments apart |
| Active, end of period | Customers active this period | The closing balance | Growth lead | Must equal opening plus new plus resurrected minus churned |
| Balance of gains to losses | New plus resurrected, divided by churned | Above one the base grew | Growth lead | Its trend and which term moved it |
The revenue version uses the same rows with expansion and contraction added between retained and churned, and the same rule applies: the closing balance must equal the opening balance plus every movement. A KPIA metric tied directly to an objective. built on the bridge should name the movement and the definition of active, such as "churned accounts in their first months", rather than a blended rate that hides which term changed.
A café rota app reads a flat quarter (illustrative)
Prellumy Rota takes a flat quarter apart
Prellumy Rota, an invented shift-rota web app sold by monthly subscription to small cafés and bakeries, counts an account as active in a month if it holds a paid subscription at the month's end. Active accounts read 2,410 at the end of March and 2,414 at the end of June, and the quarterly report calls the quarter flat.
The growth accounting bridge tells a different story.
| Month | Start | New | Resurrected | Churned | Retained | End | Gains to losses |
|---|---|---|---|---|---|---|---|
| April | 2,410 | 190 | 45 | 230 | 2,180 | 2,415 | 1.02 |
| May | 2,415 | 215 | 38 | 255 | 2,160 | 2,413 | 0.99 |
| June | 2,413 | 240 | 31 | 270 | 2,143 | 2,414 | 1.00 |
New accounts rose every month after a paid social push, and churn rose almost as fast. Splitting June's 270 churned accounts by tenure shows 160 in their first two months, almost all of them accounts that never published a second rota, and most of those came through the new paid social campaigns. Resurrection fell month by month: the returning accounts were mostly seasonal venues reopening in spring, a stream that had largely arrived by June.
The June revenue bridge shows why the finance team had not noticed a problem: opening recurring revenue of £86,900, plus £8,400 from new accounts, £2,100 of expansion from venues adding a second site and £1,000 from resurrected accounts, minus £900 of contraction and £9,300 lost with churned accounts, closes at £88,200. Expansion was covering the leak.
The growth lead gives each movement an owner. Activation, defined as publishing a second rota within the first month, goes to the product and lifecycle team with an onboarding change to test. The paid social lead reviews the campaigns' early churn before raising spend. Seasonal win-back moves to the weeks before venues reopen, with a holdout of lapsed seasonal accounts to show what it adds.
Common mistakes
- Counting returning customers as new. Without a resurrected movement, every win-back looks like acquisition and the channel that touched the customer last takes the credit.
- Reporting the earliest periods. Customers whose history began before the data look new; keep a warm-up and report only after it.
- A bridge that does not balance. If opening plus gains minus losses does not equal closing, the definitions or the data are wrong, and no trend read from it can be trusted.
- An active definition that ignores the buying rhythm. A window shorter than the natural gap between purchases manufactures churn and resurrection that are really customers between orders.
- Netting churn against gains. A net change of zero can hide rising acquisition and rising churn, which need different owners and different fixes.
- Treating the ratio as a target with a benchmark. The balance of gains to losses is arithmetic about the base; set expectations from the business's own history, not from a number seen elsewhere.
- Reading the customer bridge alone where revenue recurs. Expansion and contraction can make the revenue bridge move in the opposite direction to the customer count.
The growth economics hub shows where the bridge sits alongside the unit economics table, the cohort analysis and the payback reading.
