The AARRR Pirate Metrics Coach skill for Claude

The problem this skill solves is one anyone selling through a site or app will recognise. Traffic comes in, all sorts of things happen, and yet it does not grow. The reflex is almost always the same: buy more visitors. But a funnel that leaks at the bottom does not get better from more volume going in. What it enforces is the other order: map the whole journey first, then flag the biggest leak, only then act. And because every stage gets exactly one KPI that measures a customer action, your team suddenly talks about the same five figures instead of separate dashboards.

It is written for business owners, marketers and product teams: webshops, SaaS, apps and local services, as the file states literally. It is part of the skill library of 100 free skills NL we make available through the AI and automation service, with no account and no sales email afterwards.

One thing you need to know upfront: the skill does not measure anything itself. It needs your figures, or your honest estimates, and that second category stays marked as [ESTIMATE] throughout the whole analysis. That is not a limitation but the whole point: a growth analysis that leans on invented benchmarks is worse than no analysis at all.

Steering happens on vanity metrics. Page views, followers, likes: figures that feel good but do not measure a customer action. A dashboard full of records while revenue stands still. The skill rejects figures like that as a KPI, with a fixed test: does this measure something a customer does, or something we like to see?

The funnel stops at the deal. A classic sales funnel ends where the customer pays. But it is precisely after that where growth becomes cheap or expensive: do customers come back, do they bring others, do they buy again? Anyone not measuring Retention and Referral pays for every customer again through advertising.

Everything gets tackled at once. After a disappointing quarter, an action list sweeps through: new campaign, new onboarding, new pricing plan, all in the same month. If things improve afterwards, nobody knows why. One leak gets flagged and the rest is deliberately put on hold. That feels strict, and it is the only way to know afterwards whether anything actually helped.

Estimates become facts. Someone says in a meeting that conversion is around two per cent, and three weeks later that figure sits in a quarterly plan as a measurement. Every figure you do not supply yourself gets the label [ESTIMATE], and that label stays put through the entire analysis.

Activation gets confused with registration. The subtlest mistake, and exactly the mistake from the example in the file: creating an account is an administrative action, not a success experience. Anyone measuring activation on accounts is blind to the moment the customer genuinely decides whether this product is for them.

In diff form: what gets thrown out of an average growth dashboard, and what the five stages hand back instead.

Then comes the core: a fixed method of six steps. One: map the product and revenue model, and immediately choose the analysis order, AARRR or the retention-first variant RARRA, with the choice explicitly named. Two: define exactly one KPI per stage that measures a customer action, with a fixed measurement point, and reject vanity metrics. Three: fill the funnel with real figures over a fixed period, with every estimated figure getting the label [ESTIMATE].

Four: find the biggest leak, exactly one stage, with the explanation of why that particular stage costs the most. Five: two actions per stage, of which only the one on the leak is fully worked out with what, why and how you measure the effect. Six: plan the re-measurement, in four to six weeks, with the same definitions and the same period length.

The file also prescribes a fixed answer format, with the blocks PRODUCT, FUNNEL, BIGGEST LEAK, DIAGNOSIS, ACTIONS and RE-MEASUREMENT. That format is the reason you can compare analyses across weeks: the re-measurement in week six has exactly the same shape as today's baseline, so any difference you see is a difference in your figures and not in the shape of the answer.

The file also contains the worked example you saw in FIG.01, with the explicit note that all figures are fictional, and a list of six things the skill never does. Those refusals get their own section below, because they matter at least as much as the method itself. The file closes with four source references: Dave McClure's original presentation and three articles that explain the framework and the RARRA variant.

Want to learn to build an instruction file like this yourself? The full structure of a good SKILL.md sits in writing a SKILL.md NL. This skill is a compact example of that: frontmatter with triggers, a numbered method, a fixed output format, an example and a refusals list.

The pirates sit in the acronym: say AARRR out loud and you sound like a pirate. Behind the joke sits a serious point. McClure saw startups steering on figures that impress investors but say nothing about the health of the business. His answer: five stages that together describe a customer's entire life cycle, each with a metric that measures a behaviour rather than a feeling.

RARRA comes from Thomas Petit and Gabor Papp. In 2017 they described the same five stages in a different order, with Retention first. Their argument: for products that run on recurring use, such as apps and subscriptions, acquisition is pointless as long as users do not stay. Every new user who leaves after a week is marketing money evaporating. The skill knows both orders, and it explicitly names which one it chooses and why, which is step one of the method.

The one-leak rule is classic bottleneck logic. A funnel is a chain, and a chain gets stronger at its weakest link, not its strongest. Work on Referral while Activation leaks, and you have satisfied customers inviting others to an experience that loses new users straight away. Hence the fixed order: leak first, then re-measure, only then the next stage. Anyone who likes that principle will find the broader version in the Pareto Analyst skill: finding the small cause that carries the largest share of the effect.

The stages connect to other frameworks in the library. Retention is at its core habit formation, and that is what the Hook Model Builder skill is about: the loop of trigger, action, reward and investment that brings users back. And anyone still shaping their entire revenue model can put the funnel next to the canvas from the Lean Canvas Strategist skill, where the same metrics have their own box.

Even without installing the skill, you can sharpen your own growth question with the framework: draw the five stages, put the most honest figure you have against each stage, and look at where the drop-off is biggest. That exercise alone often already exposes that you have been pushing in the wrong place.

After that, using it is simple: describe your product and give the most honest figure you have per stage. It asks itself what is missing and neatly flags what is estimated. If you get stuck anywhere while installing, every environment is explained step by step in installing Claude skills NL.

There are also situations where you are better off leaving it be, or using it with adjusted expectations. If you sell one-off projects with no repeat character, Retention and Referral structurally say little and you effectively end up with a classic funnel: the skill still works, but the distinguishing part of the framework stays unused. If you have no product and no figures yet at all, do not start at the measurement but at the model: the Lean Canvas comes before the funnel. And do not mistake the skill for an analytics connection: it does not read anything out automatically, you supply the figures and it watches whether they are honestly labelled.

The most important limit is the same as with any measurement framework: the analysis is only as good as the discipline behind it. Anyone who has the leak flagged but skips the re-measurement has a nice diagnosis and no proof at all that the treatment works. That is why it plans the re-measurement in by default, with the same definitions and the same period length as the first time. The one thing it cannot do is carry it out for you.

The whole funnel starts at Acquisition: people need to find you first. If your funnel already leaks there, that is where the real work for this analysis sits: becoming visible to the people looking for you, in Google and in AI answers. The free SEO scan shows in a few seconds where your site stands. And if you want to talk further about what else AI can do for your growth, from separate skills to full automation, we simply do that in a conversation.