Pick the right metrics.

Key terms

Metric

A number you track over time, with a fixed definition: what’s counted, how it’s calculated, and over what period.

Leading metric

A number that moves early and hints at what will happen later, like how many shoppers save a sofa to their list this week, which hints at next month’s orders.

Lagging metric

A number that tells you what already happened, like last month’s revenue. Important, but slow to react.

Input metric

A number your team can move directly through its work, and that you believe moves a bigger outcome.

Vanity metric

A number that looks good but doesn’t help you decide what to do, like total page views. Eric Ries’s term.

Signal

Something you can see customers do or say that shows whether a goal is being met, like finishing the size check or answering a survey.

Primary metric

The one number you agree, before launch, will tell you whether a change worked.

North Star metric

The single metric a product team uses to capture the value customers get from the product.

Guardrail metric

A number you watch to make sure a change doesn’t cause harm somewhere else, like returns or complaints.

The idea

Most products track hundreds of numbers. Page views, sign-ups, clicks, time on site. It’s tempting to grab whichever one went up and call it success. That’s how teams fool themselves.

A good metric is chosen before you look at the results, and it’s tied to a goal you care about. This lesson gives you three tools for choosing well (HEART, AARRR, and a North Star), the exact formulas for the metrics you’ll meet most, and a short plan you can fill in for any feature. The running example is the made-up online furniture store from lesson 1.

The lesson, in six parts

1. 1What makes a metric useful.

   A metric is useful when it helps you make a decision. Eric Ries put it plainly in 2009: “The only metrics that entrepreneurs should invest energy in collecting are those that help them make decisions.” He called the rest vanity metrics: numbers that “might make you feel good, but they don’t offer clear guidance for what to do.”

   Three pairs of ideas will help you sort the numbers in front of you:

   Pair	What it means	At the made-up furniture store
   Vanity vs. actionable	A vanity number grows on its own and doesn’t say why. An actionable one is tied to a cause you can act on.	Total page views (vanity) vs. the share of sofa shoppers who use the doorway-size check (actionable)
   Lagging vs. leading	A lagging number reports the past. A leading one moves first and predicts what comes later.	Monthly revenue (lagging) vs. sofas saved to shoppers’ lists this week (leading)
   Outcome vs. input	An outcome is the result you want for customers and the business. Inputs are the things your team can change that you believe produce it.	Sofas customers keep, not return (outcome) vs. share of sofa orders where the doorway-size check was used (input)

   Vanity and actionable metrics: Eric Ries, 2009. Leading, lagging, and input metrics are described in Amplitude’s North Star materials.

   Ries’s own fixes are worth copying. Look at numbers per customer, not totals (“Metrics are people, too,” in his words). Compare groups of customers who started at the same time, called cohorts. And where you can, run a split test, where half the shoppers see the change and half don’t. A total like page views can rise just because more people showed up. A rate per customer doesn’t rise just because more people came.

2. 2Turn a goal into metrics with HEART.

   In 2010, Kerry Rodden, Hilary Hutchinson, and Xin Fu, researchers at Google, published a short paper at the CHI conference on measuring user experience at scale. It gave two things: five categories of metrics, called HEART, and a simple process for choosing them, called goals, signals, metrics.

   The five categories are Happiness, Engagement, Adoption, Retention, and Task success. Happiness is about attitudes, like satisfaction, and is often measured with a survey. The other four come from what people actually do in the product.

   Category	Goal (made-up store)	Signal	Metric
   Happiness	Shoppers feel sure about size before buying	Answers to a short survey after checkout	Share who say they felt sure the sofa would fit
   Engagement	Shoppers come back to compare sofas before buying	Saving sofas to a list	Sofas saved per shopper per week
   Adoption	New shoppers try the doorway-size check	First use of the check	Share of this week’s sofa shoppers who used the check for the first time
   Retention	Customers come back for their next piece	A second order	Share of first-time buyers who order again within 6 months
   Task success	Shoppers can finish the size check	Check started vs. finished	Share of size checks finished without an error

   The store and its goals are made up. The categories and the goals, signals, metrics process are from Rodden, Hutchinson, and Fu, 2010.

   Three tips from the paper keep this honest. You don’t need all five: it says “it is not always appropriate to employ metrics from every category.” Choose signals that move only when the experience gets better or worse, not for unrelated reasons. And prefer rates to raw counts, because counts grow with your user base: “ratios, percentages, or averages per user are often more useful.”

3. 3Follow customers through a funnel with AARRR.

   In 2007, Dave McClure presented “Startup Metrics for Pirates”. The pirate part is the acronym: AARRR. It splits the customer’s path into five steps, each with its own number. In plain words:

   • Acquisition: people arrive, from different channels.
   • Activation: they have a good first visit.
   • Retention: they come back.
   • Referral: they like it enough to tell others.
   • Revenue: they pay.

   Step	A metric for the made-up store
   Acquisition	Visitors per week, by channel (search, ads, email)
   Activation	Share of new visitors who save a sofa or use the doorway-size check on their first visit
   Retention	Share of visitors who return within 30 days
   Referral	Share of orders that came from a customer’s shared link
   Revenue	Share of visitors who place an order, and the average amount per order (average order value, explained in part 5)

   The useful move is to compare the steps. If many people arrive but few activate, more ads usually won’t help much: start with the first visit. Look for the step where the biggest share of people drops out, and work there first.

4. 4Choose a North Star metric, and the inputs that move it.

   Teams drift when everyone chases a different number. A North Star metric gives them one to share. Amplitude’s North Star Framework, written up in a playbook co-authored by John Cutler, defines it as “a single critical rate, count, or ratio that represents your product strategy.” It names three qualities. The metric represents the value users get from the product. It’s within the product and marketing teams’ influence. And it’s a leading indicator of revenue.

   A real example from the playbook: in 2005, while Netflix still shipped DVDs, its product team used the percentage of customers who put three or more DVDs in their queue (the list of discs Netflix would mail next) during their first session.

   The second half of the framework matters as much. Under the North Star sit a few inputs, typically three to five, that teams can move in their daily work. Cutler’s test is a good one: “If you can move your North Star directly, it’s probably not a good North Star.”

   A good North Star counts value delivered, not activity. For the store, “pieces delivered and still kept after the 30-day return window, per week” beats “visits”, because a sofa that’s returned didn’t help anyone. Amplitude advises against revenue as a North Star because it lags: it doesn’t give an early signal. It can also be pushed up for a while in ways that hurt customers.

5. 5The common metrics, with the right formulas.

   You’ll meet these everywhere. People often get the formulas slightly wrong, and then two teams argue about numbers that were never the same thing. Here are the standard versions:

   Metric	Formula	Notes
   Conversion rate	Sessions (or users) that completed the goal ÷ all sessions (or users), × 100	Say which: per session or per user. They give different numbers.
   Average order value	Total revenue ÷ number of orders, in the same period	A few very large orders can pull it up.
   Retention rate	(Customers at the end of the period − new customers gained during it) ÷ customers at the start, × 100	Subtract new customers, or you’ll count them as retained.
   Churn rate	Customers lost during the period ÷ customers at the start, × 100	For the same group and period, retention + churn = 100%.
   DAU/MAU	Users active today ÷ users active in the last 30 days	Shows how often people come back: higher means more of your monthly users show up on a given day.
   Net Promoter Score (NPS)	% promoters (9 or 10) − % detractors (0 to 6), from “How likely is it that you would recommend…?” on a 0 to 10 scale	Passives (7 or 8) count only in the total. The result runs from −100 to +100. It’s widely used, but researchers dispute whether it predicts growth better than other loyalty questions (Keiningham and colleagues, 2007), so treat it as one signal, not the goal.
   Customer satisfaction (CSAT)	Responses of 4 or 5 on a 1 to 5 scale ÷ all responses, × 100	Ask right after a specific moment, like delivery.

   Sources: Google Analytics (conversion rate as key event rate, and DAU/MAU), Shopify (average order value), HubSpot (retention and churn), Bain & Company and Fred Reichheld (NPS), Qualtrics (CSAT). Companies vary details, so write your own definitions down.

   One habit saves hours of confusion: for every metric on your team, write down the formula, the time period, and what counts. Is an “active user” someone who opened the app, or someone who did something in it? Decide once, and write it where everyone can see it.

6. 6Put it together: a metric plan for one feature.

   Here’s a way to choose metrics for anything you launch, in order. Write the goal in customer terms. List the signals that would show it in behavior or attitudes. Pick one primary metric that best captures the goal. Add one or two input metrics you can watch daily. Add a guardrail: a number that would catch harm. Then set a target and a date before you launch, so you can’t move the goalposts afterwards.

   Ask

   • What would change for customers if this worked?
   • How exactly is this number calculated?
   • Could it go up while customers are worse off?
   • What will we do if it moves, or if it doesn’t?

   Avoid

   • Picking the number after you see the results
   • Totals that grow with traffic, like page views
   • Ten primary metrics, so nothing is primary
   • A metric nobody on the team can influence

   The template

   Goal: [what changes for customers]. Signal: [what they’d do or say]. Primary metric: [exact formula, period]. Inputs: [one or two numbers the team can move]. Guardrail: [what must not get worse]. Target: [from X to Y by date].

   Filled in, for the made-up furniture store

   Goal: shoppers pick a sofa that fits their room. Signal: they use the doorway-size check before buying, and fewer sofas come back. Primary metric: share of sofa orders returned because they didn’t fit, per month. Inputs: share of sofa orders where the check was used; share of checks finished without an error. Guardrail: sofa conversion rate must not fall. Target: cut fit returns by a quarter within three months (all numbers made up).

   Notice the guardrail in the example. The doorway-size check could cut returns simply by scaring people off sofas altogether. Watching conversion as well tells you whether you solved the problem or just moved it.

With AI: AI tools can now suggest metrics, write the query, and chart the result in minutes. That makes it easier than ever to drown in numbers. The judgment stays with you: which goal matters, which metric reflects it, and whether a number could rise while customers are worse off. AI features also need their own measures, such as how often an answer was correct or helpful, checked on real examples rather than assumed.

FEATURE OR PRODUCT:

GOAL (what changes for customers):

SIGNALS (what they would do or say):
  -
  -

HEART CATEGORIES THAT FIT (Happiness, Engagement, Adoption, Retention, Task success):

PRIMARY METRIC
  Name:
  Formula (numerator / denominator):
  Period:
  What counts, and what doesn't:

INPUT METRICS (the team can move these):
  -
  -

GUARDRAIL (must not get worse):

TARGET: from ____ to ____ by ____

WHAT WE'LL DO IF IT MOVES, AND IF IT DOESN'T:

Ask AI about this lesson

All lessons