AARRR Stages

There was a long stretch where every team watched a different number. Marketing arrived at the meeting with monthly visits, product with features shipped, sales with deals closed. All three describe the same people moving through the same product, and there was no way to join them up.

So the meeting stalled in the same place every month. Early on at Pagedeck, Ravi, the growth marketer, would report that visits were up forty percent, and Grace, the CEO, would ask why the number of paying teams hadn't moved. Nobody could say what happened in between. Between a visit and a payment sit a signup, a first use, a return visit, and an invitation to a colleague, and nobody was counting any of those middle steps.

The problem was not a shortage of numbers. It was that the numbers came in one lump. Knowing that ten thousand people signed up does not decide anything. Buy more ads, cut fields from the signup form, or go after the people who leave on day one — the same figure supports all three, and when every option looks equally plausible the loudest voice wins.

In 2007 a startup investor gave a talk proposing that this path be cut into five stages: acquisition, activation, retention, referral, and revenue. String the initials together and you get AARRR, which sounds enough like a pirate that the framework picked up pirate metrics as a nickname. The nickname spread first, which is why the framework is so often mistaken for a fixed list of five metrics. The five stages are not metrics. They are names for places a user passes through, and what gets counted at each place is decided product by product.

The mechanism is plain. Put five gates along the path a user walks and count how many people clear each gate. The share of people past one gate who also clear the next is that segment's conversion rate. Ten thousand visits producing twelve hundred signups makes the first segment twelve percent. Once the path is cut this way, the stalled meeting changes shape. Visits up forty percent with paying teams flat becomes a sentence with an object: four thousand more people entered the first stage and almost none of them cleared the second. Now there is one place to work on.

On the growth map this framework belongs to the funnel branch. A growth model, which writes out in sums and products how users and revenue accumulate, decides which variable to push. The five stages say where that variable lives in the actual product. When the model says to move the share of users who stick from seventy percent to eighty, translating that into "get more people to come back on day three" is the work these five stages do.

One branch sits under this term. It is called stages, and it holds five terms in order. There is only one branch because the framework is a sequence rather than a classification: the five are not parallel categories but gates, each reachable only by clearing the one before.

1. Acquisition — the stage where a new user first reaches the product. Arriving through search, clicking an ad, following a link from a colleague all land here. What gets counted is usually visits plus arrivals broken out by channel, and without that channel split the stage is close to useless. A thousand people from ads and a thousand from search routinely clear the next gate at rates that differ by a factor of two.
2. Activation stage — the stage where someone who arrived first feels what the product is for. It watches what happens after signup rather than the signup itself. At Pagedeck the marker is creating and saving a first document. Pinning down that marker is half the work of this stage. Without one, signups quietly become the activation count, and every number below it inflates to match.
3. Retention stage — the stage where someone who tried the product comes back. It's normally read by grouping everyone who arrived in the same week and checking how many remain after one week, four weeks, twelve weeks. Of the five this one reports last and resists change hardest. Buying more ads lifts the first stage next week; this stage moves only when the product actually becomes worth returning to.
4. Referral — the stage where users who stayed bring other people in. Invite links, sharing, and referral rewards attach here. Products where this stage turns see the cost of the first stage fall, because of the arithmetic: if a hundred users bring an average of 0.4 people each, that many fewer have to be bought.
5. Revenue — the stage where a user actually spends money. Teams converting from free to paid, the amount one team pays, the share that upgrade a plan later all come from here. It sits last but inherits whatever the four stages above it produced, so attempts to lift this stage alone usually end up deferring a problem that lives higher on the path.

There's an old argument about the order. One side says to work the letters as written and start with acquisition; the other says to start with retention. The second argument is simple enough: otherwise you're pouring water into a bucket with a hole in the bottom. If ninety percent of a product's users are gone within a week, doubling the ad budget leaves the number of people still around four months later roughly unchanged. In practice teams follow neither order and instead look for the segment where the conversion rate drops hardest, then work there.

Cutting the path into stages does not by itself make it readable. A row of absolute counts still doesn't say what to do. The values that get used are the conversion rates between stages. Pagedeck's last quarter ran ten thousand visits, twelve hundred signups, five hundred and forty first documents saved, three hundred and twenty returning after a week, ninety colleague invitations, and forty-eight teams converting to paid. As segments that reads twelve percent, forty-five percent, fifty-nine percent, twenty-eight percent, fifty-three percent. The forty-five stands out. More than half of everyone who signed up left without ever creating a document, a bigger loss than the segments on either side of it.

Producing and reading those numbers is the job of growth analytics: recording which screen a user was on and what they clicked, then grouping arrivals by period and following how many are still there weeks later. Without that record the five stages stay a drawing on a whiteboard.

What follows the discovery of a blocked stage splits two ways. Getting more people from one stage to the next out of the same arrivals is conversion optimization. Writing down in advance what change should move which rate by how much, then splitting users into two groups to find out, is a growth experiment. What Pagedeck did about its forty-five percent segment was show an example document instead of a blank one right after signup, and two weeks later that segment read fifty-two percent.

The five stay bundled because watching one alone breaks the rest. Target acquisition by itself and the ads widen until anyone at all is arriving, which collapses the second stage. Target revenue by itself and the fastest answer is to shrink what the free plan allows, which thins the first stage instead. Lined up together, each of these shortcuts shows up immediately in the neighbouring stage. A month where arrivals doubled while the second segment's rate halved is a month where what grew was not users but visits that went nowhere. The counts went up and the product did not, and only the rates say so.

"This quarter we're leaving acquisition alone and watching activation. However much we pour into the top, it's all draining at the second stage."

A declaration that work on arrivals is postponed for a quarter in favour of fixing the experience right after signup. It also means no ad spend, so this sentence usually needs agreement from the marketing side before it holds.

"Signup to first document is forty-five percent. Last quarter the same segment was fifty-two."

Stated as a segment conversion rate rather than a count, and compared against the previous quarter. What follows is a proposal to find out what changed, since something got worse.

"I'd leave the referral stage empty for now. Attaching invitations while retention is broken just recruits people who leave faster."

A judgment about not touching a later stage while an earlier one is blocked — the order of the five applied to an actual schedule.

At Pagedeck this table opens the Monday growth meeting. Ravi puts up five rows with a sixth line of segment conversion rates underneath, and the meeting starts by picking the segment on that bottom line that dropped hardest. Reading the absolute counts first is against the house rules. Even in a month when visits climbed, if the second segment sagged then activation is that month's subject.

Once a stage is chosen, whoever owns it takes over. For activation, Mia, the product manager, and Noa, the product designer, bring one thing to change on the screen right after signup. For acquisition, Ravi brings the next-gate pass rate split by channel. The table itself decides nothing. It decides who does most of the talking this month.

The most common mistake is reading the five stages as a plan to be completed one stage at a time. The initials run in order and the diagram is drawn top to bottom, so it looks that way. In reality all five run at once. No product finishes acquisition and then starts activation, and at this moment there are people entering the first stage and people inviting colleagues from the fourth. What the framework fixes is not the order of the work but the order in which places are inspected. Read as a plan, it produces a quarter spent entirely on the top stage while everyone leaking out of the second keeps leaking — ad spend doubled, paying teams unchanged.

Putting one number per stage on a dashboard and treating that as using the framework is almost as common, because five rows on a screen feel like something under control. Five rows on their own say nothing. Ten thousand visits, twelve hundred signups, five hundred and forty activated supports no decision; twelve percent and forty-five percent show which side is leaking harder. The rows also tend to move together. When arrivals rise, the four rows beneath rise too, so the table always looks better even in months that were worse once converted to segment rates. Teams reading only the counts therefore find problems late. It takes a month where arrivals finally dip for them to notice that the lower stages had been leaking all along. By then the fix is the same one that was available six months earlier, except that the quarter it would have saved is already spent.