You can buy your way to a bigger top of funnel and watch revenue refuse to move, because the number you grew was never the one holding you back. AARRR tells you which number that is.

The quick version

  • AARRR = Acquisition, Activation, Retention, Revenue, Referral, the five stages of a user's journey from stranger to advocate.
  • It's a diagnostic, not a scoreboard: the win isn't a bigger count at every stage, it's reading the conversion between stages.
  • Find the one leak that, plugged, lifts everything below it, then fix that lever alone.
  • The leak is rarely acquisition. It's usually activation or retention, the stages teams find least fun to work on.

The idea in depth

In 2007 the investor Dave McClure, later founder of the accelerator 500 Startups, gave a talk called Startup Metrics for Pirates and reduced the sprawl of growth analytics to five letters: AARRR (McClure, 2007). The five stages map a single user's path. Acquisition: they arrive. Activation: their first experience is good enough that they "get it." Retention: they come back. Revenue: somewhere along the way, they pay. Referral: they like it enough to tell someone. McClure's framing was a forcing function, of everything you could measure, these are the five that decide whether a business lives.

flowchart TD
    A(["Acquisition
strangers arrive"]) -->|"~30% activate"| B(["Activation
first 'aha' moment"])
    B -->|"~40% return"| C(["Retention
they come back"])
    C -->|"some pay"| D(["Revenue
they monetise"])
    C -->|"some refer"| E(["Referral
they tell others"])
    E -.->|"new strangers"| A
    classDef leak fill:#f3e8ff,stroke:#9333ea,color:#3b0764;
    class B,C leak;
The five AARRR stages as a funnel. The shaded stages, activation and retention, are where most products lose the users they paid to acquire. The dashed line back to the top is the part a pure funnel underplays. Leaders Loop

The number that matters lives between stages

A funnel is defined by its drop-offs. The headline count at any single stage tells you almost nothing on its own; the conversion between stages tells you where the business is breaking. This is the distinction Eric Ries drew in The Lean Startup between vanity metrics and actionable ones: a total that only ever goes up (registered users, page views) flatters you, while a rate that can move in either direction (activation rate, week-two retention) forces a decision. Ries put it plainly in 2009: "The only metrics that entrepreneurs should invest energy in collecting are those that help them make decisions."

"The only metrics that entrepreneurs should invest energy in collecting are those that help them make decisions.", Eric Ries, 2009

So the move is: stop reporting AARRR as five counts. Report it as conversion rates, acquisition→activation, activation→retention, retention→revenue, retention→referral, each one a percentage you can argue about. The instant you do that, the leak announces itself.

Why the leak is almost never acquisition

Acquisition is the stage teams obsess over because it is the most visible, the most buyable, and the most flattering. It is also, for most struggling products, the wrong stage to fix, and the growth thinking that came after McClure says why.

Sean Ellis, who coined the term "growth hacking" in 2010 and ran early growth at Dropbox, puts the weight on activation: get a new user to the product's "aha moment," where the core value clicks, as fast as possible. In Hacking Growth (Ellis & Brown, 2017) he argues that improving activation has an outsized effect because every later stage inherits from it, you cannot retain, refer or monetise a user who never understood the product. The stage below that carries even more weight: Andrew Chen and the team at Reforge make the case that retention is the foundation of growth. Improve it and you don't just keep more users, you quietly raise the ceiling on every other lever at once, because a retained user has more chances to refer, more chances to pay, and a lower effective acquisition cost. This is triangulated, not one opinion: McClure orders the framework so retention sits at its centre, Ellis builds his method on activation feeding it, and Reforge makes retention the multiplier the others depend on.

The practical upshot: resist the instinct to fix the most visible stage and fix the earliest broken one. A leak high in the funnel poisons every number beneath it, so a poor activation rate is almost always a better use of a quarter than a slightly cheaper acquisition channel. Spend on acquisition only once the stages below it hold water.

flowchart TD
    Q1(["Where does
conversion drop
most sharply?"])
    Q1 -->|"strangers don't sign up"| AC(["Fix Acquisition
channel / message fit"])
    Q1 -->|"sign-ups never 'get it'"| AV(["Fix Activation
onboarding to the aha moment"])
    Q1 -->|"they get it, then vanish"| RE(["Fix Retention
habit, value loop, re-engagement"])
    Q1 -->|"they stay but won't pay"| RV(["Fix Revenue
packaging, pricing, paywall"])
    AV --> RULE(["Fix the EARLIEST
broken stage first,
it poisons all below it"])
    RE --> RULE
Find the biggest leak by conversion drop, then fix the earliest broken stage, not the most visible one. Leaders Loop

The limitation: funnels are linear, growth is often loop-shaped

AARRR has a real limit worth naming: it is shaped like a funnel, and funnels run one way. Pour in at the top, collect at the bottom, with no native account of how the output feeds back to create more input. The products that grow fastest don't run as funnels; they run as loops, where a retained, paying user produces the next user. As Brian Balfour, Casey Winters, Kevin Kwok and Andrew Chen argued in Reforge's "Growth Loops are the New Funnels", the funnel is too micro a view to answer the real question, how does this product grow? Use AARRR for what it's genuinely good at, diagnosis, and reach for loop-thinking when you turn to strategy. We take the loop side apart in the sibling guide on growth loops and flywheels.

A worked example

Picture a small B2B scheduling tool, call it Rota. The team has one growth goal this quarter: more sign-ups. They run paid search, sponsor a podcast, ship a referral pop-up. Sign-ups jump from 2,000 to 3,000 a month and everyone is pleased. Revenue barely moves.

So they instrument the funnel as conversion rates and the story inverts. Acquisition→activation, defined here as "scheduled a first shift", is running at 22 per cent (an illustrative figure, not a benchmark), meaning nearly four in five sign-ups never do the one thing the product exists to do. Of the few who activate, week-four retention is healthy at around 55 per cent (also illustrative). The leak isn't acquisition at all; it's the gap between signing up and the first scheduled shift. The expensive new traffic just poured more users into the same broken first-run experience.

The fix is unglamorous and cheap. Rota rebuilds onboarding around the single activating action: a guided flow that imports a real team and books one shift in the first session, with everything else deferred. Activation climbs from 22 to 40 per cent (illustrative). Because the retained cohort was already sticky, that one change roughly doubles the number of users reaching the part of the funnel that actually generates revenue, without buying a single extra sign-up. One lever, fixed at the right stage, moved everything downstream.

flowchart LR
    S1(["Acquisition"]) --> M1(["sign-ups,
CAC by channel"]) --> L1(["channel & message fit"])
    S2(["Activation"]) --> M2(["% reaching
the aha action"]) --> L2(["onboarding to first value"])
    S3(["Retention"]) --> M3(["week-N cohort
retention curve"]) --> L3(["habit & re-engagement loop"])
    S4(["Revenue"]) --> M4(["paid conversion,
net revenue retention"]) --> L4(["packaging & pricing"])
    S5(["Referral"]) --> M5(["invites,
viral coefficient"]) --> L5(["share moments & incentives"])
Stage → the metric that measures it → the lever you actually pull. Each stage has a different lever; treating them as one undifferentiated "growth" problem is how teams pull the wrong one. Leaders Loop

Frequently asked questions

Isn't AARRR outdated compared with growth loops?

No, they do different jobs. AARRR is a diagnostic: it tells you which stage is leaking. A growth loop is a strategy: it describes how the stages compound once they hold. The funnel shows you the constraint; the loop shows you how to grow once it's fixed. Use both, in that order. More in our guide to growth loops and flywheels.

Which metric matters most?

For most products, retention, because it sits beneath everything else and multiplies it. But "most important in general" is the wrong question. The metric that matters most for you is the one at your biggest conversion drop. That's the whole point of running the diagnostic instead of copying someone else's priorities.

How do I find the biggest leak?

Express each step as a conversion rate, not a count. Read it by cohort rather than in aggregate, lumping every sign-up together averages away the very pattern you need to see. Then find the sharpest drop and fix the earliest broken stage, since a leak high in the funnel poisons every number below it. Fix one stage at a time so you can attribute the result.

What's a PQL?

A Product-Qualified Lead, a user who has hit a usage milestone in the product that predicts they're ready to buy (as opposed to a marketing-qualified lead, who has only shown intent, like downloading a guide). A PQL is essentially an activation signal repurposed for sales: it ties the activation stage of AARRR to revenue by saying "this person has experienced the value, now is the moment to convert them."

We already track all this, so what's new?

Tracking the numbers and using them to choose what to fix next are different activities. Most dashboards show five counts all trending up, and counts hide leaks, a rising sign-up line and a rising churn line can sit side by side and both look like success. The diagnostic only works when you express each step as a rate, read it by cohort, and commit to one fix at a time. Having the data on a screen is not the same as letting it decide your roadmap.

Where to go next

Related in the Toolkit