Lean Analytics: Choosing the Metric That Matters

Lean Analytics, by Alistair Croll and Benjamin Yoskovitz, argues you should track the one metric that matters most for your current stage and business model, then draw a line in the sand that forces a decision. As that question gets answered, the metric changes and you graduate to the next stage.

Quick Answer: Pick the One Metric That Matters (OMTM) for your stage — Empathy, Stickiness, Virality, Revenue, or Scale — and for your business model. Make it a rate or ratio that changes how you act, set a target line before you test, and move on once you clear it.

Most founders drown in numbers long before they learn anything from them. Lean Analytics is a corrective: a discipline for choosing what to ignore. This guide walks through its core moves — the OMTM, the five stages, the good-versus-vanity test, and the six business-model archetypes — the way an analyst would use them, not the way a dashboard would.

Why One Metric Beats a Full Dashboard Early On

Early on, a single focused metric beats a full dashboard because attention, not data, is the scarce resource. A wall of twenty numbers invites you to admire the ones that look good and quietly ignore the ones that hurt. One metric forces one question and one decision.

This is counterintuitive for a PM-turned-founder. Analytics maturity inside a large company rewards breadth — more instrumentation, more segments, more panels. A startup rewards the opposite: a single, honest answer to "is this working right now?"

Croll and Yoskovitz's point is not that other metrics are useless. You still instrument everything; you still keep the plumbing in place. The OMTM is simply the number you optimize and report against this quarter, the one every experiment is judged by.

A dashboard also hides a subtle trap: it lets you feel busy without being decisive. If nothing on the screen can trigger a specific action, you are collecting reassurance, not evidence. Running a quick vanity vs. actionable metrics audit on your current dashboard usually reveals how many panels are there to soothe rather than to steer.

The analyst's real job at this stage is subtraction. Anyone can add another chart; the discipline is deciding which single number gets to override the others when they disagree. Lean Analytics frames this as being data-driven without being data-blind — instinct still proposes the experiment, but the OMTM decides whether the experiment worked. Reporting metrics belong in the background; the number on the wall should be the one you would bet the next two weeks of work on.

The Lean Analytics Stage-and-Model Map (At a Glance)

Lean Analytics maps your OMTM onto two coordinates: the stage your company is in and the business model you run. The stage tells you which question is urgent; the model tells you which numbers actually express that question. Everything else in the book is detail hung on this frame.

The table below is a conceptual map, not a prescription — the sections that follow unpack each row in turn.

Read this map to locate yourself, then choose a specific number. The stage narrows the question; your business model, covered further down, narrows the metric. The authors are explicit that any benchmark figures attached to these stages are rough, industry-dependent ranges — starting points to test, never fixed targets.

The One Metric That Matters (OMTM), Defined

The One Metric That Matters is the single number you care about most at your current stage — the one that answers your most pressing question and forces a decision. It is deliberately singular and deliberately temporary: when the question is answered, the OMTM changes.

Croll and Yoskovitz give four reasons the OMTM works:

• It answers your most important question. At any moment there is one thing you most need to know; the OMTM is its proxy.
• It forces a line in the sand. Naming the metric compels you to name a target — a concrete definition of success and failure.
• It focuses the whole team. One number everyone shares beats a dozen numbers each person defends.
• It builds a culture of experimentation. When one metric is the scoreboard, anyone can propose a test and see whether it moved.

The word temporary is doing real work here. The metric you obsess over shifts as you cross stages: a retention rate in Stickiness gives way to a spread rate in Virality, then to margin and lifetime value in Revenue. Choosing the one metric that matters by stage is a repeated act, not a one-time setup.

The OMTM is also not the only metric — it is the most important one right now. Guardrails and diagnostics stay on in the background. What changes is where your attention, and your team's, is pointed.

A concrete progression makes the point. A team in Stickiness might fixate on weekly returning users; once that number is healthy and stable, the pressing question stops being "do people stay?" and becomes "do they bring others?" — so the OMTM becomes a referral rate. Later, when growth is real but the bank account is not, the question shifts again to "does each customer pay for itself?" and the OMTM becomes the ratio of lifetime value to acquisition cost. Same company, same discipline, three different numbers — because the most important question kept changing underneath them.

The Five Stages of Lean Analytics: Empathy, Stickiness, Virality, Revenue, Scale

Lean Analytics describes five stages every startup passes through in order — Empathy, Stickiness, Virality, Revenue, and Scale — each with its own urgent question. You are not meant to skip a gate: chasing growth before you have retention just pours users into a leaky bucket.

Each stage below has a different OMTM because each answers a different question. The authors treat these as gates rather than milestones: you can be busy at the wrong gate, and the metrics are how you catch yourself doing it.

Empathy: Have You Found a Real Problem Worth Solving?

Empathy is the discovery stage: you are interviewing people and hunting for a problem intense enough that someone will pay — in time, effort, or money — to make it go away. The signals here are largely qualitative, patterns across conversations rather than conversion rates.

You graduate when you have evidence of a real, widely felt problem and the outline of a solution people actually want. This is the structured problem-discovery work that Edmired is built to support, and it is worth doing thoroughly before you write a line of product code. For the metric side of it, see empathy-stage metrics for customer discovery.

The classic trap here is treating Empathy as a formality to rush through on the way to building. Founders who skip it optimize a product nobody was waiting for, then blame later-stage metrics for a problem that was set in stone at this gate.

Stickiness: Do People Keep Coming Back?

Stickiness asks whether your MVP keeps people engaged and returning. The focus is retention and engagement, not acquisition — the authors are firm that you fix the leaky bucket before you pour more water in.

The OMTM here is typically an engagement or retention rate: are the people who show up finding enough value to come back on their own? Optimizing acquisition while retention is broken simply raises the cost of disappointing more people faster.

You graduate when a meaningful share of users return without being prodded — a curve that flattens rather than decaying to zero. Until it does, Stickiness is where you stay, iterating on the core value, no matter how tempting a growth channel looks.

Virality: Do Users Bring Other Users?

Virality is about growth that compounds — existing users pulling new ones in. Now the OMTM shifts toward a spread rate and how long each cycle takes.

The book separates three kinds of spread: inherent virality baked into using the product, artificial virality driven by incentives and rewards, and word-of-mouth from genuine satisfaction. Each is measured and improved differently, so naming which one you are chasing matters.

The subtle risk at this gate is confusing paid acquisition with virality. Buying users is a Revenue-stage question about unit economics; virality is about whether the product spreads on its own. Conflating the two lets a leaky, unprofitable engine look like organic momentum.

Revenue: Does the Business Model Actually Work?

Revenue asks whether you earn more from a customer than it costs to acquire and serve them. The OMTM moves to unit economics — revenue per customer and the relationship between customer lifetime value and acquisition cost.

This is where problem-solution fit has to become a business. A product people love but cannot be sold profitably fails here, which is why this stage sits close to what product-market fit actually means and why the two ideas are so often conflated.

You graduate when the money coming in from a customer reliably exceeds the money spent to win and keep them, and that gap is repeatable rather than a one-off. The failure mode is declaring victory on a single profitable cohort while the average customer still loses money.

Scale: Can You Grow Beyond the Core Market?

Scale is about turning a working business into a bigger one: new channels, new segments, new geographies, and the organization to support them. Metrics broaden to margins, channel efficiency, and market reach.

You focus here only after the first four gates are genuinely cleared. Scaling a business that has not proven stickiness or economics just makes its problems more expensive.

Good Metrics vs. Vanity Metrics: How to Tell Them Apart

A good metric changes how you behave; a vanity metric only makes you feel good. The authors' test is simple: if this number goes up, do you know what you would do next? If the honest answer is "nothing," it is a vanity metric.

Croll and Yoskovitz list four qualities of a good metric:

• Comparative — you can compare it across time periods, segments, or competitors, so it tells you a direction, not just a value.
• Understandable — people can remember it and talk about it without a footnote.
• A rate or a ratio — inherently actionable, it captures behavior over time and is easy to act on.
• Behavior-changing — the decisive quality: a good metric prompts a specific action.

The pattern is clearest when you line up common vanity metrics against the actionable counterpart that measures the same thing honestly.

Notice the shape: the vanity column is almost always a cumulative total that can only rise, while the actionable column is a ratio that can fall — and therefore forces a response. That single property, whether the number can deliver bad news, is most of what makes a good metric.

Five Pairs of Metric Types Worth Knowing

Beyond good versus vanity, Lean Analytics sorts metrics into five contrasting pairs. Knowing which side of each pair you are looking at changes how much you should trust the number.

• Qualitative vs. quantitative. Qualitative data — interview notes, open-ended answers — captures the why behind behavior; quantitative data captures the what and how much. Early stages lean qualitative; later stages lean quantitative.
• Vanity vs. actionable. Actionable metrics tie to a decision; vanity metrics only flatter. It earns its place in this list because the confusion is so common.
• Exploratory vs. reporting. Exploratory metrics hunt for surprising, unknown insights; reporting metrics keep tabs on normal operations. Early on you want the explorers.
• Leading vs. lagging. A leading metric predicts what is coming, like an active sales pipeline; a lagging metric confirms what already happened, like churn. Leading indicators let you act before the outcome is fixed.
• Correlated vs. causal. Two metrics that move together are correlated; only an experiment proves one causes the other. Correlation helps you predict; causation lets you actually change the result.

The practical upshot is to prefer metrics that are actionable, leading, and — where you can establish it — causal, because those are the ones you can steer by rather than merely watch.

The Six Lean Analytics Business Models and Their Metrics

Lean Analytics groups startups into six business models — e-commerce, SaaS, free mobile app, media site, user-generated content, and two-sided marketplace — because each one lives or dies by different numbers. Your model decides which metric actually expresses "is this working?"

The table below pairs each model with the thing it fundamentally sells and the metrics it lives by. Match your own product to the closest row before you name your OMTM.

The takeaway is that "growth" is not one thing. It means transactions for e-commerce, recurring revenue for SaaS, and liquidity for a marketplace — so a benchmark borrowed from the wrong model is worse than no benchmark at all. Reading the full breakdown of startup metrics by business model is the fastest way to avoid importing a competitor's OMTM that does not fit your economics.

Real products often blur two archetypes — a SaaS tool with a marketplace attached, or a media site monetized through commerce. The book's advice is not to force yourself into a single box but to identify which model drives the decision in front of you right now, and to borrow that model's metrics for this stage. The archetype you optimize against can shift as your revenue mix does.

Segments and Cohorts: Why Averages Hide the Truth

A single average blurs brand-new users with veterans and winners with churned accounts, so Lean Analytics leans hard on segments and cohorts to see what the average conceals. A segment slices users by a shared trait; a cohort groups them by when they started, so you compare like with like as the product changes.

Three techniques do most of the work:

• Segmentation splits users by source, plan, geography, or behavior, revealing where a metric is actually strong or weak instead of reporting one flattened number for everyone.
• Cohort analysis compares, say, January's signups against April's at the same point in their life cycle. Product improvements show up as later cohorts behaving better than earlier ones — and, crucially, a growing top line can no longer hide worsening retention.
• A/B and multivariate testing is how you move from correlation to causation, isolating whether a change actually caused the number to move.

The through-line is honest comparison. A blended average can rise while the underlying product gets worse; cohorts and segments are how you catch that early, which is exactly the kind of self-honesty the whole framework is built to enforce.

Drawing a Line in the Sand Before You Run the Test

Drawing a line in the sand means committing, in advance, to the number that will count as success — before you see the result. The authors treat this as the move that turns a metric into a decision: without a pre-set line, you will rationalize whatever number you get.

The loop is short and repeatable:

1. Pick the OMTM for your current stage and business model.
2. Set the line — the target you will accept as success — and write it down before you run the experiment.
3. Run the test, compare the result to the line, and act: proceed, iterate, or pivot.

Here the book's honesty about benchmarks matters. Lean Analytics does offer rough ranges for some models, but Croll and Yoskovitz are explicit that these are directional guidance drawn from experience — not universal laws. They vary widely by industry, audience, and geography, so treat any published number as a hypothesis to test against your own data, not a target handed down as fact.

The discipline of drawing a line in the sand for metric targets is mostly about sequence: you commit to the threshold before you know the outcome, which is the only reliable defense against quietly moving the goalposts later.

This loop never really stops. Clearing one line raises a new question, which sets a new OMTM and a new line — the Lean Analytics version of build-measure-learn. Each pass either advances you toward the next stage gate or tells you, unambiguously, that the current approach is not working and something has to change. The pre-committed line is what keeps that verdict honest instead of negotiable.

Common Metric Traps Founders Fall Into

The most common Lean Analytics mistakes are optimizing a later stage before nailing an earlier one, tracking totals instead of rates, and moving the line after seeing the result. Each one lets you feel productive while dodging the honest question.

• Skipping the gate — chasing virality or paid acquisition while stickiness is still broken, so growth just accelerates churn.
• Counting vanity totals — reporting cumulative numbers that can only go up and therefore never signal a problem.
• Moving the goalposts — redefining "success" after the experiment, which erases the whole point of a pre-set line.
• Dashboard sprawl — tracking everything and deciding on nothing, because no single number owns the decision.
• Confusing correlation with causation — assuming a metric that rises alongside success actually caused it.
• Copying someone else's benchmark — adopting another company's target number as if it applied to your model and market.

Most of these are failures of restraint, not analysis. The skill Lean Analytics teaches is choosing the one number that will tell you the truth, then having the discipline to act on it.

Key Takeaways

• The OMTM is a focusing device, not your only metric. Track the one number that answers your most pressing question right now; keep the rest as background instrumentation.
• Stage and model together pick the metric. The stage — Empathy, Stickiness, Virality, Revenue, or Scale — sets the question; the business model sets which number expresses it.
• A good metric changes behavior. If a rising number would not change what you do next, it is vanity — prefer comparative, understandable rates and ratios over cumulative totals.
• Move through the five gates in order. Retention before growth, growth before monetization; skipping a stage just makes later problems more expensive.
• Draw the line before you test. Commit to the target that counts as success in advance, so results are unambiguous and goalposts stay put.
• Benchmarks are hypotheses, not laws. The book's ranges are rough, industry-dependent starting points — validate them against your own data rather than adopting them as fact.
• Match your OMTM to your model. Growth means transactions, recurring revenue, or liquidity depending on the business, so a metric borrowed from the wrong archetype misleads.

Frequently Asked Questions

What is the One Metric That Matters in Lean Analytics?

The One Metric That Matters (OMTM) is the single number you care about most at your current stage. It answers your most pressing question, forces a target line, focuses the team, and encourages experimentation. It is temporary — as the question is answered, the OMTM shifts to the next stage's concern.

What are the five stages of Lean Analytics?

The five stages are Empathy, Stickiness, Virality, Revenue, and Scale, passed through in order. Empathy finds a real problem, Stickiness proves retention, Virality drives compounding growth, Revenue proves the economics, and Scale grows beyond the core market. Each stage has its own OMTM and its own question to answer.

What makes a metric a vanity metric?

A vanity metric is one that makes you feel good but does not change how you act — typically a cumulative total like registered users, pageviews, or downloads that can only rise. The test is simple: if the number went up, would you know what to do differently? If not, it is vanity, and you want an actionable rate or ratio instead.

How do I choose the right startup metric for my business model?

Start with your stage to fix the question, then match your business model — e-commerce, SaaS, free mobile app, media site, user-generated content, or two-sided marketplace — to the metric that expresses it. Each model lives by different numbers, so a marketplace's liquidity and a SaaS company's churn are not interchangeable targets.

Does Lean Analytics give benchmark numbers I should hit?

Lean Analytics offers rough benchmark ranges for some models, but the authors are explicit that these are directional guidance drawn from experience, not universal rules. They vary widely by industry, audience, and geography. Use them to form a hypothesis about your own line in the sand, then test that line against your real data rather than treating any figure as a fixed target.