Lean Analytics

What is the one metric that matters?

It is the single number you would run the company on if you could see only one, chosen for right now and expected to change. The argument is about attention rather than about measurement: a team that watches 30 numbers watches none of them, and a team that watches one will notice when it moves and will have an opinion about why.

The book is firm that this is a temporary commitment. You pick the number for the question you are currently trying to answer, you set a target before you start, and when the question is answered you pick a different number. That last part is the part teams skip. A one-metric dashboard that has not changed in 18 months is not focus, it is a habit.

Chapter 2, "How to Keep Score", is where the useful tests are. A good metric, the authors say, is comparative, is understandable, and is a ratio or a rate. The same chapter separates leading from lagging metrics and correlated from causal ones, and runs a list of eight vanity metrics to watch out for. Those three tests will clean up most dashboards on their own. A total is almost never a good metric, because a total only goes up and tells you nothing about whether this month was better than last.

What are the five stages, and does the order hold?

Five, in order: empathy, stickiness, virality, revenue, scale.

Empathy is the stage where you are still finding out whether the problem is real, and the metric is how many conversations you have had and how consistent the answers are. Stickiness is whether people come back and keep using the thing, which has to be true before anything else is worth building. Virality is whether they bring other people. Revenue is whether the money works. Scale is whether it keeps working when you add fuel.

The order is the useful claim. It says do not spend on acquisition before stickiness, and do not optimize revenue before you have something people return to. That holds up well, and it maps almost exactly onto the way I file work by stage: stickiness is retention, virality is referral, revenue is this stage.

Where the order needs a note is that virality sits before revenue in the book, and in most companies I have worked in it does not. Referral is the last stage to arrive, because it depends on a product people already keep, and because a referral program built before retention flattens is a discount with a share button. The five stages are a sequence of questions, not a Gantt chart, and virality is the one most often answered out of turn.

The book also sorts companies into six models: ecommerce, software as a service, free mobile app, media site, user-generated content, and two-sided marketplaces. Each gets its own chapter of what to measure. The sorting is still good. It is the reason the book travels better than most analytics writing, which quietly assumes a single business model.

What does the book get right?

Three things, and they are not the things it is famous for.

It makes you name the number before you look. Part III is titled "Lines in the Sand", and the introduction to it puts the case plainly: "Unless you have a line in the sand, you don't know if you're crushing it or being crushed." That is the practice, not the benchmark table that follows it. Writing down the number you would be pleased with, before you pull the report, is the cheapest protection against reading a result backwards. Nearly every disagreement I have sat in about whether a test worked was really a disagreement about what the two sides had expected, and nobody had written it down.

It takes qualitative work seriously as measurement. The empathy stage has metrics, and they are counts of conversations and the consistency of what people said. Most analytics books treat customer conversations as the soft part that happens before the real work. This one treats them as the first stage's data.

Chapter 28 exists. It is called "What to Do When You Don't Have a Baseline", and it is the chapter every early team needs and no analytics book usually writes. The answer is to become your own baseline: measure now, improve against yourself, and stop shopping for a comparison you cannot get.

What has dated?

The benchmark numbers. The lines in the sand were drawn against the cost of attention and the cost of software in 2012. Paid acquisition, churn, email engagement and app install economics have all moved several times since. Read those chapters for which metrics to track per model and ignore the figures entirely. I would not use a 2013 conversion or churn line to judge a company in 2026, and neither would the authors.

The tool assumptions. The examples assume you are instrumenting things yourself, which was true when product analytics was a project. The work of getting the event into the warehouse is smaller now, and the work of agreeing what the event means is exactly as large as it was. The book spends its time on the first.

The implied scarcity of measurement. "One metric" was advice for teams that had to choose because every number cost engineering time. Today the constraint is attention rather than instrumentation, which makes the advice more useful rather than less, but for a different reason than the book gives.

Virality's position in the sequence, for the reason above.

How do you pick your own one metric this week?

1. Write down the question the company is actually arguing about. Not the goal. The argument. "Are people coming back" is a question. "Grow revenue" is not.

2. Pick the number that would settle it, and make it a ratio or a rate rather than a total.

3. Write the line in the sand before you pull it. Two numbers: the one that would make you confident, and the one that would make you change plan.

4. Put it somewhere a person sees without logging in, and put the date next to it.

5. Set a review date. When the question is answered, the metric's job is over.

Who should read it?

A founder between 0 and the first few hundred customers, and any first growth hire in their first month. Read Part I and Part II properly, read Part III for the structure and not the numbers, and read Chapter 28 twice if you have no comparison.

Skip it if you already have a working weekly number everyone trusts. The book's main gift is deciding what to watch, and you have already done that.