Pirate metrics (AARRR): how I use them to find the growth bottleneck

Pirate metrics are five questions about what people actually do

In 2007 the investor Dave McClure gave a short talk with a memorable title, Startup Metrics for Pirates, and a five letter acronym that sounds like a pirate clearing his throat: AARRR.^{1 2} The letters stand for five stages a customer goes through:

1. Acquisition: people arrive, from some channel.
2. Activation: they have a good first experience.
3. Retention: they come back and keep using the thing.
4. Referral: they like it enough to bring someone else.
5. Revenue: they pay.

McClure's own one line definitions are almost that plain. Acquisition is "users come to the site from various channels", activation is "users enjoy 1st visit", retention is "users come back, visit site multiple times", referral is "users like product enough to refer others", and revenue is simply users doing something that earns money.¹

It is worth being clear about what AARRR is. It is not a theory, and nobody proved it. It is a practitioner's checklist, written by an investor for software founders who were drowning in page views and sign ups and could not tell whether their product was working. Its value is that it turns one vague question ("is the business growing?") into five sharp ones, each answered by a count.

The point of the framework, in his slides, is conversion: how many people make it from one step to the next, and what each step is worth.¹ Everything in this article follows from that.

Each box is a group of people and each arrow is the share that moves on. The order is McClure's, and as a later section shows, real customers do not always walk it in that order.

The framework has a close cousin for shops, with stages called awareness, consideration, conversion, retention and advocacy. The article on full funnel growth frameworks works through that version. Here I stay with the original, built for software products.

Worked example: a booking app in Cork

Throughout this article I use an invented business: Trimline, a booking and reminder app for independent dog groomers, run by Aoife in Cork at trimline.example. Nobody real, every number made up, so that every step can be shown.

A groomer signs up for a free 14 day trial. If they like it, they pay €24 a month. After hosting, text message costs and card fees, Aoife keeps €20 of that as margin.

In March she spends €1,500 on marketing: €1,200 on Google Ads and €300 to sponsor the newsletter of a grooming supplies wholesaler. She also writes posts in two Facebook groups for groomers and has a good few helpful articles that people find through search.

Her March, in one line: 4,000 visitors, 240 trials, 60 new paying customers. Is that good? On its own, nobody can say. The rest of this article finds out.

Each stage needs one event you can count

A stage you cannot count is a stage you cannot fix. So before any dashboard, write down one event per stage, in plain words, and keep the definition the same every month.

McClure's slides suggest choosing five to ten conversion steps across the whole journey, with the warning that fewer is better, and giving each one to someone who owns it.¹ In a business of one, the owner is you, which at least settles the meeting schedule.

Good definitions have three properties. They describe something the customer does, not something you did. They can be counted from your own system, ideally your own database, not estimated by a tool. And they are close enough to the money that a change in them means something.

Be careful with analytics defaults. In Google Analytics 4 a session counts as engaged if it lasts longer than 10 seconds, has a key event, or has two or more page views.³ That is a reasonable definition of "did not leave at once", and McClure's own example of an early activation step was almost identical: two or more pages, 10 or more seconds, two or more clicks.¹ But neither is activation in the sense that matters. Ten seconds on a page tells you a person did not leave. It does not tell you they got any value.

Count with consent. Most of these events live in your own product database (a trial started, an appointment booked, an invoice paid), and there you need no tracking script at all. For the website side, the article on event tracking and consent mode covers how to collect it properly.

Worked example: Aoife's five definitions

Aoife writes them on one page:

• Acquisition: a unique visitor to trimline.example in the month, by channel, from her consented analytics.
• Activation: a trial account that books its first real client appointment and sends that client a reminder within seven days of signing up. From her database.
• Retention: a paying account that books at least one appointment through Trimline in the calendar month. From her database.
• Referral: a new trial that arrived through an existing customer's personal link. From her database.
• Revenue: a trial that becomes a paid subscription. From her payment provider.

Notice what is not on the list: page views, time on site, followers. They are not wrong to look at. They just do not answer any of the five questions.

New customers are the product of the rates between stages

Once each stage has a count, the links between them are rates: one count divided by the one before it. Multiply the top count by every rate on the way down and you get the number at the bottom.

$$P = V \times s \times q$$

Here $P$ is the number of new paying customers in a month, $V$ is the number of visitors, $s$ is the share of visitors who start a trial, and $q$ is the share of trials that become paying customers. Each of $s$ and $q$ is a simple ratio of two counts. Foundations: ratios and rates

This is the same chain as in the full funnel article, with fewer links. Because it is a multiplication, a 20% improvement in any one rate gives 20% more customers, and improvements in different rates multiply each other rather than add.

Worked example: Aoife's chain

From her March counts:

$$s = \frac{240}{4{,}000} = 6\%, \qquad q = \frac{60}{240} = 25\%$$

Here $s$ is Aoife's trial rate, 240 trials divided by 4,000 visitors, and $q$ is her trial to paid rate, 60 paying customers divided by 240 trials. Foundations: ratios and rates

Back together: $4{,}000 \times 0.06 \times 0.25 = 60$. Her blended cost per new paying customer is $€1{,}500 / 60 = €25$.

Two links are visible. Activation hides inside $q$, and the next sections pull it out.

Acquisition: judge a channel by the customers it brings

The acquisition question is not "how many people came?". It is "which channels bring people who go all the way?". McClure's slides put it as judging channels by volume, cost and conversion, and then segmenting channels by conversion at the deepest level you can measure, ideally money.¹

In practice that means one table: a row per channel, a column per stage, and a last column for cost per paying customer. The channel with the most visitors is often not the channel with the most customers. For how to tag the traffic so the rows are right, see UTM tracking. For how to choose which channels to test in the first place, see the bullseye framework.

Worked example: Aoife's four channels

Aoife splits March by channel:

Channel	Cost	Visitors	Trials	Activated	Paying
Google Ads	€1,200	1,500	60	15	10
Search (her articles)	€0	1,200	72	27	19
Facebook groups	€0	1,000	70	24	16
Wholesaler newsletter	€300	300	38	18	15
Total	€1,500	4,000	240	84	60

Google Ads brings the most visitors and the fewest customers per visitor: $10 / 1{,}500 = 0.67\%$. The newsletter brings the fewest visitors and the most customers per visitor: $15 / 300 = 5\%$.

Cost per paying customer is the channel cost divided by its paying customers: $€1{,}200 / 10 = €120$ for Google Ads and $€300 / 15 = €20$ for the newsletter. The blended €25 from the last section hides a sixfold difference. Foundations: marginal versus average

The horizontal axis lists the four channels. The vertical axis is how many paying customers each 100 visitors from that channel turned into. The channel that looks biggest in a traffic report is the weakest here, and the smallest channel is the strongest by a distance. Search and the Facebook groups cost Aoife time rather than money, which is real, but not a cash cost per customer.

Aoife does not switch Google Ads off. It still pays: each customer costs €120 and, as the retention section shows, is worth about €229. But she now knows where the next €300 should go first, and she asks the wholesaler about a second slot.

Activation: the first experience that predicts everything after it

Activation is the stage most businesses skip, and in my experience it is often the most valuable one. A sign up is a promise. Activation is the moment the product keeps it: the new user does the thing the product exists for and gets something back.

The best activation event is one that predicts what happens later. You find it by comparing the people who stayed with the people who did not, and asking what the stayers did early that the others did not. In software it is often called the aha moment. In a groomer's day it is less mystical: the first client booked without a phone call.

McClure's own suggestion for improving this stage was plenty of landing page tests and A/B tests.¹ I would start one step earlier, by asking people who did not activate what stopped them. The article on onboarding and habit formation goes deeper into the first days of a software product.

Activation matters because the share of trials that pay is a blend of two very different groups.

$$q = a \times p_a + (1 - a) \times p_n$$

Here $q$ is the share of all trials that become paying customers, $a$ is the share of trials that activate, $p_a$ is the share of activated trials that pay, and $p_n$ is the share of trials that never activated but pay anyway. The two groups cannot overlap, so their shares add up. Foundations: probability

Worked example: Aoife's activation gap

Of her 240 trials, 84 activated, so $a = 84 / 240 = 35\%$. Of those 84, 54 paid. Of the 156 who never activated, six paid.

$$p_a = \frac{54}{84} = 64.3\%, \qquad p_n = \frac{6}{156} = 3.8\%$$

Here $p_a$ is the share of activated trials that paid and $p_n$ the share of never activated trials that paid, each a count of payers divided by the size of the group. Foundations: ratios and rates

Putting them back into the formula: $0.35 \times 0.643 + 0.65 \times 0.038 = 0.225 + 0.025 = 0.25$, the 25% from before.

An activated trial is about 17 times as likely to pay. So Aoife asks the 156 who did not activate one question by email: what got in the way? The answer comes back again and again: "my clients are all in a spreadsheet and I did not have time to type them in."

She builds an import that reads the spreadsheet in one step. Suppose that lifts activation from 35% to 45%, so 108 trials activate instead of 84. If the new activators pay at the same rate as the old ones:

$$108 \times 0.643 + 132 \times 0.038 = 69.4 + 5.1 = 74.5$$

Here $108$ is the number of activated trials ($a \times 240$ with $a = 0.45$), $132$ is the trials that still do not activate, and $p_a = 0.643$ and $p_n = 0.038$ are the pay rates of the two groups. Foundations: probability

So 74.5 paying customers instead of 60, from the same 4,000 visitors and the same €1,500.

One honest caveat. The import makes activation easier, but the groomers it newly activates may be less keen than the ones who typed 200 clients in by hand. Activation predicts paying. That does not prove it causes paying. So Aoife treats the 74.5 as a forecast and checks the real pay rate of the next cohorts before she believes it. The article on statistical testing covers how many trials she needs before the difference is more than noise.

Retention: whether the product keeps its promise

Retention is the stage that tells you whether the product works. Acquisition can be bought. Activation can be coaxed. Nobody keeps using something that does not help them.

For a subscription, the tempting definition is "still paying". I prefer a usage definition, because usage drops first and cancellation follows weeks later. By the time a customer cancels, the decision was made a while ago.

What a point of retention is worth comes from the lifetime value formula, the same one the article on retention and referral works through in years. For a monthly subscription I use months.

$$V = M + M \times \frac{r}{1 + d - r}$$

Here $V$ is the value of a new paying customer in today's money, $M$ is the margin a customer brings in a month, $r$ is the monthly retention rate (the share of paying customers still active a month later), and $d$ is the monthly discount rate, the return you could earn on the money elsewhere. Foundations: lifetime value and Foundations: discounting

A yearly discount rate becomes a monthly one by compounding backwards:

$$d = (1 + d_{\text{year}})^{1/12} - 1$$

Here $d$ is the monthly discount rate and $d_{\text{year}}$ the yearly one. A 10% year is $1.10^{1/12} - 1 = 0.008$, or 0.8% a month, not $10\% / 12$. Foundations: compounding

The formula assumes the same retention rate every month. Real retention curves usually drop hardest in the first months and then flatten. For a careful answer, the article on survival analysis measures the curve as it really is. For a decision about where to spend the next month, the simple version is enough.

Worked example: what a point of retention is worth to Aoife

Over the last six months, 92% of Aoife's paying customers were still booking through Trimline a month later. So $M = €20$, $r = 0.92$ and $d = 0.008$.

$$V = 20 + 20 \times \frac{0.92}{1 + 0.008 - 0.92} = 20 + 20 \times \frac{0.92}{0.088} = 20 + 209.09 = €229.09$$

Here $V$ is the value of one new Trimline customer, $M$ is the €20 of monthly margin, $r$ is the 92% monthly retention and $d$ is the 0.8% monthly discount rate. Foundations: lifetime value

Now suppose she lifts retention by one point, to 93%:

$$V = 20 + 20 \times \frac{0.93}{0.078} = 20 + 238.46 = €258.46$$

The symbols are the same as above, with $r = 0.93$. Foundations: lifetime value

One point of retention adds $€258.46 - €229.09 = €29.37$ to every new customer. On 60 new customers a month, that is $60 \times €29.37 = €1{,}762.20$ of extra value from each month's intake.

The horizontal axis is monthly retention, the vertical axis is in euro. The rising line is what a new customer is worth at each retention rate, and the flat line is the €120 Aoife pays for a customer through Google Ads, her most expensive channel. The curve bends upwards: the step from 88% to 90% is worth €29.17 a customer, the step from 94% to 96% is worth €123.53. In a subscription, small differences in monthly retention become large differences in value, because every month's loss compounds.

Where do Aoife's lost customers go? She asks everyone who stops booking through the app one question. The most common answer: "my regulars still ring me, so I stopped opening it." Her fix is a booking link each groomer can put on their Google Business Profile and Instagram, so clients book themselves. That is a product change, not a marketing one, which is usually the case at this stage.

Referral: ask when the customer is happiest

Referral is where a funnel becomes a loop: a customer brings a new customer with no advertising bill attached. McClure's advice on timing is the part I quote most: ask for referrals only after a happy experience, and his slides suggest waiting until the customer would rate it at least 8 out of 10.¹

The number that measures it is the viral coefficient, usually written $k$: how many new customers each customer brings in, on average.

$$k = i \times c$$

Here $k$ is the viral coefficient, $i$ is the average number of people each customer invites (in practice, distinct new people who open their referral link), and $c$ is the share of those invited people who become customers. Foundations: ratios and rates

Below 1, referral does not make a product grow by itself. It adds a fixed share of extra customers on top of every customer you win another way. The retention and referral article works through the series that turns $k$ into that share, and the Irish rules on asking customers to pass things on. The one rule to keep here: let the customer share their own link, and never write to their contacts yourself.

Worked example: Aoife's referral link

Groomers know other groomers, so Aoife adds a personal referral link to the app: a free month for the groomer who refers, and a free month for the new one once they pay. She shows it not at sign up, when nobody has a reason to recommend anything, but after a groomer's 20th appointment booked through Trimline, when the app has clearly saved them some phone calls.

In the quarter, her 180 new paying customers' links are opened by 54 groomers, so $i = 54 / 180 = 0.3$. Twelve of those 54 become paying customers, so $c = 12 / 54 = 0.222$.

$$k = 0.3 \times 0.222 = 0.067$$

Here $k$ is Aoife's viral coefficient, $i = 0.3$ the groomers reached per customer and $c = 0.222$ the share of them who pay. Foundations: ratios and rates

Using the series from the retention and referral article, every 60 customers from marketing become $60 / (1 - 0.067) = 64.3$ in the end. Four or so extra a month. Useful, and nothing like a rocket.

Revenue: how fast the money comes back

McClure's slide on revenue is the most honest one in the deck: this is the part you have to figure out.¹ The framework tells you to count it. It does not tell you how to price.

For a subscription, three revenue numbers matter: the share of trials that pay (already in $q$), the margin per customer per month, and the payback period, how many months of margin it takes to earn back what you paid to win the customer.

$$n = \frac{CAC}{M}$$

Here $n$ is the payback period in months, $CAC$ is the customer acquisition cost (the marketing spend divided by the paying customers it brought), and $M$ is the monthly margin per customer. Foundations: ratios and rates

Payback matters to a small business for a reason lifetime value hides. Lifetime value says what a customer is worth eventually. Payback says how long your cash is tied up before you see it again, and a business with a long payback can be profitable on paper and short of money in the bank.

Worked example: Aoife's payback by channel

Her margin is €20 a month per customer.

$$n_{\text{ads}} = \frac{€120}{€20} = 6 \text{ months}, \qquad n_{\text{newsletter}} = \frac{€20}{€20} = 1 \text{ month}$$

Here $n_{\text{ads}}$ and $n_{\text{newsletter}}$ are the payback periods for a customer from each channel, the cost per paying customer from the channel table divided by the €20 monthly margin. Foundations: ratios and rates

At 92% retention, how many Google Ads customers are still there after six months to finish paying back? The share still active after $t$ months is $r^t$, so $0.92^6 = 0.61$. About four in ten leave before they have paid back their own advertising, and the ones who stay carry them. Foundations: compounding

That is fine, because the average customer is worth €229.09 against €120. But it tells Aoife something the value alone did not: every euro moved from Google Ads to the newsletter comes back six times faster, and every point of retention shortens the wait for the money.

The order is a convention, not a law

McClure listed the stages as acquisition, activation, retention, referral, revenue, and that is the order on his slides.¹ Customers do not read slides. Aoife's customers pay on day 14, before most of them have referred anyone, so for her the real order is acquisition, activation, revenue, retention, referral. Writing them in that order is just as valid.

It does not matter. The framework is a set of five questions, not a sequence of gates. What matters is that you count each one, and that you look at the stage in the context of the ones before it. A referral programme bolted onto a product with poor retention mostly brings in more people who leave.

The same five questions work for a shop, with different events.

An example: an invented tea shop in Galway, at galwayleaf.example. Acquisition is a visitor. Activation and revenue arrive together, as the first order. Retention is a second order within 90 days, because that is about when a bag of loose tea runs out. Referral is a new customer who used an existing customer's code. The shop's owner would count the same five things Aoife counts, but in a different order, and the rate that usually needs the most work is the step from first order to second. The full funnel article works through a shop in that shape from start to finish.

Run it as a monthly scorecard and fix one bottleneck at a time

AARRR earns its keep when it becomes a habit: the same five numbers, counted the same way, once a month, and one decision at the end.

The decision is which stage to work on next. Not by gut feeling, and not by which rate sounds lowest, because every rate sounds low. I put a euro value on a realistic improvement at each stage and start with the largest, the same method the full funnel article uses. "Realistic" means your own best month, a test you ran, or something plainly broken, never an industry benchmark without a source.

McClure's slides make a related point about the product: spend most of the effort improving what exists rather than adding features, roughly 80% to 20%.¹ The scorecard is how you know which existing thing to improve.

The steps run top to bottom. The decision box is the honest part: when every stage is in good shape, and only then, the next euro belongs at the top.

Worked example: Aoife's April decision

Aoife values three improvements, each against her March baseline of 60 new paying customers worth €229.09 each.

1. Activation from 35% to 45%, through the spreadsheet import: 74.5 paying customers instead of 60, so $14.5 \times €229.09 = €3{,}321.81$ of extra customer value a month. About two weeks of building.
2. Retention from 92% to 93%, through the client booking link: $60 \times €29.37 = €1{,}762.20$ a month. About a week of building, but a year before she can be sure it worked.
3. Referral from $k = 0.067$ to $k = 0.10$, through a better placed prompt: customers from each 60 go from 64.3 to $60 / 0.9 = 66.7$, so $2.38 \times €229.09 = €545.23$ a month. An afternoon.

The horizontal axis lists the three improvements, the vertical axis the extra customer value each one adds per month of new customers. Activation is the clear bottleneck: it is worth twice the retention fix and six times the referral one. The referral prompt is so quick that she does it anyway on a Friday afternoon, but it is not where the month goes.

So April is the import, and nothing else in the product changes until she has enough new trials to judge it. In May she counts again, with the same five definitions. If activation has moved and the pay rate of the new activators holds up, retention is next.

That is the whole framework: five questions, one count each, a rate between each pair, a euro value on each realistic fix, and one change at a time. The pirate theme is a joke. The discipline is not.

Footnotes

1. Dave McClure, Startup Metrics for Pirates: AARRR (long version), slides, SlideShare, https://www.slideshare.net/dmc500hats/startup-metrics-for-pirates-long-version

2. Dave McClure, Startup Metrics for Pirates, Master of 500 Hats blog, September 2007, archived capture of 16 September 2007, Internet Archive, https://web.archive.org/web/20070916121535/http://500hats.typepad.com/500blogs/2007/09/startup-metrics.html

3. Google, [GA4] Engagement rate and bounce rate, Analytics Help, https://support.google.com/analytics/answer/12195621