Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

Marketing Frameworks › The KPI Tree Framework

KPI Tree for Retention

How a KPI tree changes when the objective is retention: choosing the root, defining retained, cohort splits, churn types, early-life inputs and guardrails.
Objective guideintermediate7 min readUpdated

Why the acquisition tree does not transfer

The usual KPI tree multiplies its way down: revenue is orders times order value, orders are sessions times conversion rate. Every leaf is produced within the period being reported. Retention breaks that pattern in two ways. First, the outcome is a stock carried from one period to the next, so last quarter's customers are still in this quarter's number and a change at the top can come from decisions made long ago. Second, the population beneath the root keeps changing shape: each month a new cohort joins, and young cohorts leave faster than old ones. A tree that ignores both will report a retention problem every time acquisition grows, and a retention success every time it stalls. The decisions below exist to stop that.

Decisions that change for retention

Tree levelIn the generic treeIn a retention treeDecision it forces
RootA total for the period, such as revenueA stock carried over: customers kept or revenue keptLogo retention or revenue retention, and over what window
Driver splitMultiplicative funnel chainStart of period plus new minus lost, with lost split by kindWhere voluntary and involuntary churn are separated
PopulationEveryone in the periodCohorts by age, compared at the same ageWhich age buckets the tree reads and which it ignores
InputsTraffic, conversion and order valueActivation, first repeat, usage frequency, payment recoveryWhich early-life milestone predicts staying
OwnersAcquisition channelsProduct, lifecycle messaging, service and billingWho owns a leaf that marketing cannot move alone
GuardrailsCost ceiling, refundsDiscount depth, message fatigue, margin on saved customersWhich saves count and which are bought

Choose the root: customers kept or revenue kept

Logo retention counts the customers still active; revenue retention counts the money they still bring, including upgrades and downgrades. They diverge whenever customers change plans, and a business can lose customers while revenue retention holds, because the ones who stay spend more. Pick the root that matches the economics the business runs on. A subscription with a single price can use customers; anything with tiers, add-ons or variable baskets should use revenue and keep customer counts as the first driver beneath it. Whichever is chosen, write the window down, since monthly and annual retention describe different products.

Define what retained means before drawing anything

Contract businesses have an event: the cancellation or the failed renewal. Non-contract businesses, a shop or an app, have only silence, so someone has to decide how long a customer may go without buying or opening before they count as gone. Set that window from the natural purchase or usage cycle, not from a reporting convenience, and keep it fixed; a window that changes quietly will move the root more than any campaign. The churn and retention metrics guide covers the competing definitions.

Split by cohort age, not only by period

The blended churn rate is a weighted average of cohorts at different ages, so it rises whenever the customer base gets younger. Draw the population split as cohort age bands and compare like with like: this year's month-two customers against last year's month-two customers. The cohort analysis tutorial shows how to build the table that feeds this branch.

Separate voluntary from involuntary churn

Customers who decide to leave and customers whose card expired sit in the same churn count and need entirely different fixes. Make them additive siblings. The involuntary branch belongs to billing and payment recovery and is often the cheapest retention gain available; the voluntary branch belongs to product, service and messaging, and is where the diagnosis in the customer retention strategy guide applies.

Push the inputs into the customer's first weeks

Retention is mostly decided early. The leaf that predicts staying is usually an activation milestone: the first repeat purchase, the second session in a week, the first completed project. Find it by comparing the early behaviour of customers who stayed with those who left, then make it a named input. Its owners are spread across channels, which is why the tree should name the channel guide each one works from: email and SMS flows for replenishment and lapse triggers, push and in-app messaging for app habits, and the lifecycle email programme for the cadence that ties them together. The AARRR framework is a useful check that activation is in the tree at all.

Set guardrails that only retention needs

A retention tree can be flattered in ways an acquisition tree cannot. Discounts offered at cancellation keep customers on the books at a loss; pausing a subscription instead of cancelling it hides churn for a month or two; extra messages lift short-term activity and raise unsubscribes. Put the margin on saved customers, the share of the base on pause and the unsubscribe or opt-out rate in the same view as the tree, and compare the lifetime value (LTV) of saved customers with that of customers who never tried to leave.

Decision points in a worked retention tree

Illustrative example — not a real result

A retention tree for an invented coffee subscription, Hearthside Coffee Club

Active subscribers (end) = Active at start + New subscribers − Voluntary cancellations − Involuntary lapses

Hearthside ships coffee monthly at £14 a bag. It starts the quarter's last month with 8,000 active subscribers, adds 900, and closes on 8,140: 600 cancelled and 160 lapsed on failed payments. The blended monthly churn rate is 9.5% (760 of 8,000), up from 8.1% a quarter earlier, and the board asks why retention fell.

BranchCohort ageStartLostChurnA quarter earlier
VoluntaryMonth one to three2,40033614.0%14.5%
VoluntaryMonth four and later5,6002644.7%4.9%
InvoluntaryAll ages8,0001602.0%1.1%
Activation inputSecond bag rated in app900 newn/a52% rated58% rated

Read top down, the tree gives two answers. The cohort split shows voluntary churn slightly better at both ages; the blended rate rose partly because a spring promotion made the base younger, 30% of starters in the first three months against 22% before. The involuntary branch nearly doubled, traced to a card-updater setting switched off in a billing migration. The activation input fell, which predicts next quarter's young cohort will churn faster. Decisions: billing restores the updater and adds a payment-retry email; the lifecycle team owns lifting the rating step back above 55% within two months; the promotion is not blamed for churn, but its cohort is tracked separately. Guardrail check: save offers ran at 20% off for 140 customers, and their margin per bag falls from £5.10 to £2.30, so they are reported as saves at reduced margin rather than as retention.

The example shows the decisions doing their work. Without the cohort split the board would have chased a voluntary-churn problem that did not exist; without the involuntary branch the billing fault would have stayed hidden in the average; without the activation input nobody would have seen next quarter's problem arriving. And the guardrail stopped discounted saves being counted as the same thing as loyal customers.

Measuring the tree: KPIs and cadence

A retention tree mixes fast and slow measures, and each needs its own rhythm. Activation, first repeat and payment recovery move within days and can be reviewed weekly. Cohort churn needs a full cycle to mature and is best read monthly, always at a fixed cohort age. The root moves slowest of all and belongs in the quarterly review, beside the guardrails. For a website or app with signed-in users, the cohort exploration in GA4 can feed the cohort branch; for a subscription, the billing system is usually the more reliable source for both churn branches. Where the tree is used to judge an intervention, such as a new onboarding sequence, compare against a holdout group rather than the previous period (A/B Test), since cohort mix alone can move the comparison. Payback on acquisition (CAC) belongs beside the tree as a reminder that retention and acquisition share one budget. Service inputs on the voluntary branch, such as repeat contacts and contact rate per order, need settled definitions and a guardrail each, which the guide to customer service metrics provides along with a weekly service report.

Where a retention tree misleads

The identity balances whatever the cause, so it will happily attribute a seasonal lull to the cohort that happened to be young at the time. Window choices leak into the root: lengthen the inactivity window and retention improves on paper. Shared drivers such as product quality appear on both voluntary and acquisition trees and are easily double-counted as wins. And a tree that stops at churn reasons recorded in a cancellation form inherits the form's bias, since people choose the polite answer. Read the tree for where to look, then ask customers why.

Frequently asked questions

Should a retention KPI tree start from churn rate or from retained customers?

Start from the stock the business wants to keep, retained customers or retained revenue, and put churn inside the tree as the outflow. A churn rate at the root has no natural driver beneath it and invites a single number to be argued about; a stock with inflows and outflows shows which flow moved and lets each one have an owner.

How often should a retention tree be reviewed?

The early-life inputs, such as activation and first repeat, can be read every week because they move quickly and predict what follows. The outcome and the cohort curves change slowly and are best read monthly, with a fuller re-balancing at each planning cycle. Reacting weekly to the outcome mostly means reacting to noise and to the calendar.

Can one tree cover both a subscription product and one-off purchases?

Usually not. Subscription customers leave by an event, a cancellation or a failed payment, while one-off buyers simply stop returning, and the definition of retained differs. Build a tree for each model and note the shared drivers, such as service quality, on both, rather than blending definitions into a root nobody can explain.

Terms used on this page

Frameworks

Where this page is used

Also related

Sources

  1. [GA4] Get started with Explorations — Google Analytics Help
  2. Digital Marketing and Measurement Model — Avinash Kaushik

Related