What AARRR is
The acronym names five stages in the life of a customer, and the pirate nickname comes from how the five letters sound when read aloud. The framework lets a small team track a handful of numbers that matter instead of everything its analytics tool offers.
What makes AARRR a framework rather than a list is the discipline it imposes. Each stage has to be defined as an event or a query that can be counted for a specific person, and each has to be measured as a rate from the stage before. A team that applies it ends up with a short chain of conversion rates (CVRShare of visits or clicks that convert.) running from first arrival to repeat payment. That chain makes two things visible. It shows where the largest share of people fall away, which is where the work should go. And it shows which stage a proposed initiative would actually touch, which stops "growth" being used as a synonym for "more traffic".
The framework is agnostic about channel and business model. It fits a consumer app, a business software trial, a subscription box and a service that books appointments, provided each stage is given a definition that matches how that business's customers behave. The definitions are where the work is, and the section after the table is about them.
The five stages
| Stage | Question it asks | Typical definition | What it makes visible |
|---|---|---|---|
| Acquisition | How do people arrive, and from where? | A first visit, install or account, attributed to a channel | Which channels bring volume, and at what cost |
| Activation | How many reach the first moment of value? | A specific action the product's value depends on: a first project created, a first order placed, a first appointment completed | Whether the product delivers before people give up |
| Retention | How many come back, and keep coming back? | Return within the period that natural usage suggests, read by cohort | Whether value persists after the first experience |
| Referral | How many bring someone else? | Invitations sent and accepted, or new customers who name an existing one | Whether customers are themselves a source of acquisition |
| Revenue | How many pay, how much, and for how long? | Paying share, revenue per customer, lifetime value (LTVTotal value a customer brings over time.) | Whether the preceding stages add up to a business |
The stages are listed in the order most teams use. Some put revenue before referral, which suits products where payment comes before any recommendation. The order is a modelling choice; the definitions are not.
Defining stages for a business
The framework is only as useful as its definitions, and the definitions have to be written for the business rather than copied from a template. Each one names an event, the population it is counted over, and the period, so that two people pulling the number get the same answer.
Acquisition needs a decision about what counts as a new person: a visitor, an install, a verified account. It also needs an attribution rule, because the stage is only informative when it can be split by channel and priced; the growth economics hub covers acquisition cost (CACCost to acquire one new customer.) and the lifetime value it has to be judged against.
Activation is the hardest and the most important. It is the first moment a customer receives the value they came for, and it should be set at the action after which people are markedly more likely to stay. The only way to find that action is to compare retention between people who did and did not take it; an activation eventThe share of new users or sign-ups who complete the activation event, the action marking first real value, within a set window. chosen because it is easy to count, or because it flatters the funnel, defeats the purpose.
Retention needs a period that matches natural usage frequency. A messaging app is judged daily, an accounting tool monthly, a tyre retailer across years. It also needs a definition of "active" that means something: opening the app is not the same as using it. Retention is read by cohort, never as a blended figure, and the cohort analysis tutorial covers how to build the table.
Referral counts measured referrals, not stated willingness to recommend. Revenue needs gross or net, first payment or renewal, and the unit the business plans in.
Metrics per stage
Each stage gets one headline metric, which is the stage's KPIA metric tied directly to an objective., and a few diagnostics that explain it. Keeping the headline to one per stage is what keeps the dashboard readable.
| Stage | Headline metric | Diagnostics |
|---|---|---|
| Acquisition | New customers or accounts per period, by channel | Cost per acquired customer; share by channel; first-session quality |
| Activation | Activation rate: activated as a share of acquired | Time to activation; activation rate by channel and by onboarding path |
| Retention | Share of a cohort still active after the chosen period | The full retention curve per cohort; churn reasons; retention by activation path |
| Referral | New customers who came from an existing one | Invitations per active customer; acceptance rate; time from activation to first invitation |
| Revenue | Revenue per customer over the planning horizon | Paying share; average revenue per paying customer; payback period against acquisition cost |
The diagnostics are read only when the headline moves. A dashboard that shows all of them at once has reproduced the problem the framework was meant to solve. Note that the headline metrics form a chain: the activation rate multiplied by the retention rate multiplied by the paying share gives the share of acquired people who become paying customers, which is the number the acquisition cost has to be set against. The north star metric and OKRs framework usually picks one stage, most often retention or revenue, as the metric the whole organisation watches; AARRR then becomes the input structure beneath it.
Worked dashboard
A weekly AARRR dashboard for a language-learning app
| Stage | Definition | This week | Rate from previous stage |
|---|---|---|---|
| Acquisition | New installs | 24,000 | – |
| Activation | Completed a first lesson within 24 hours of install | 9,600 | 40% |
| Retention | Active in the second week after install | 3,840 | 40% |
| Referral | Accepted invitations sent by retained users | 310 | 8% |
| Revenue | Paid subscription within 30 days of install | 1,150 | 30% of retained |
The team reads the table from the top. Activation is the weakest step: 6 in every 10 installs never complete a lesson, so no later stage can see them. A redesigned first-run flow raises first-lesson completion to 46% over 6 weeks, which lifts activated users to 11,040 on the same install volume. Retention holds at 40% of activated, so week-two actives rise to 4,416 and paid subscriptions to about 1,325, with no change in acquisition spend. Acquisition cost was £2.10 per install; the same spend now produces roughly 15% more paying customers, and the payback period shortens from 11 weeks to 9.
The example shows the framework doing its job: the stage with the lowest rate and the largest population above it is where a fixed amount of effort moved the most. It also shows what the dashboard cannot say. It does not explain why installs who skip the first lesson leave, and it does not show whether the newly activated users retain as well as the earlier ones; both need cohorts and qualitative research.
Limits
AARRR is linear, and that is its central limit. It draws customers moving one way through five gates, when in many products retained customers drive acquisition through content, invitations or marketplace supply. The growth loops framework is built for that case, and a team whose growth depends on a loop will find AARRR describes the stages accurately while missing the mechanism.
It also hides time. Revenue arrives weeks or months after acquisition, so a weekly dashboard that reports all five stages for the same week compares people who arrived at different times. Cohorts fix this, but the framework does not insist on them.
The stage rates are counts, not causes. A falling activation rate says that fewer people reached first value; it does not say whether the onboarding broke, a new channel brought colder traffic, or a competitor launched. The framework points at a stage and stops.
The definitions can be chosen to flatter. An activation event set early, or a retention period set short, produces a healthy-looking dashboard and a business that is quietly losing customers. The protection is the retention comparison described above, applied honestly.
Finally, the framework starts at arrival. Everything before it, awareness, consideration and the work of getting someone to the site at all, is the territory of the marketing funnel framework, and businesses with few customers and long sales cycles will find the stage rates too noisy to read week to week.
Related frameworks
| Framework | Reach for it when |
|---|---|
| Marketing funnel | The question is about the stages before someone arrives |
| Growth loops | Retained customers are themselves the main source of new ones |
| North star metric and OKRs | One stage has to become the metric the whole organisation commits to |
| KPI tree | The arithmetic between the stages needs to be written down and decomposed further |
For business pipelines, AARRR for business-to-business lead generation redefines activation as sales acceptance, retention as nurture re-engagement and referral as more contacts from the same buying committee. Online shops need different definitions again: AARRR for ecommerce sales treats the first order and its margin as activation and a second order within the buying cycle as retention.
Common mistakes
- Borrowed definitions. An activation event copied from another product measures that product's customers, not yours.
- Blended retention. A single retention figure across all cohorts hides whether recent cohorts are better or worse than old ones.
- Reporting all five stages for the same week. The people in the revenue row arrived weeks before the people in the acquisition row.
- Treating the weakest rate as the only lever. A small rate on a large population usually matters more than a smaller rate further down.
- Counting stated referrals. Survey intent to recommend is not a referral; an accepted invitation is.
- Stopping at the stage. The framework locates the problem; cohorts, research and experiments explain it.
- Volume without its cost. Acquisition by channel reported without cost per customer makes the cheapest-looking stage the most expensive one.
