Product metrics: how to measure progress that matters to users
More registrations, more clicks and more time in an app can look encouraging. They leave an important question open: are people achieving what they came to do?
Product metrics are measurements that help a team understand use, experience and business results. A key performance indicator, or KPI, is a measure selected because it matters to a particular objective. Every metric is not automatically a KPI.
This guide explains the main measures, how to define them and how to use a small scorecard to decide what to investigate or improve.
Start with the outcome and the decision
An output is something delivered, such as a booking feature. An outcome is a change in what people can achieve, such as arranging an appointment successfully. Scrum.org’s 2024 Evidence-Based Management Guide distinguishes outputs from outcomes and encourages testing assumptions through experiments.
Consider an appointment-booking service. Its team wants people to secure a suitable appointment without calling support. Counting new accounts measures one part of the journey. It cannot show whether people found an available appointment or completed their booking.
Understand the measures before choosing them
Acquisition and activation
Purpose: Separate arrival from first useful value.
Acquisition tracks new users or customers. Specify what makes someone new, the source and the reporting period. A registration count can be inflated by duplicate accounts or people who never use the service.
Activation measures a defined first useful action within a defined window. In the booking example, it could be a new user confirming their first appointment within seven days of registration.
Example calculation: 48 of 80 eligible new users confirm an appointment within seven days. Activation is 48 ÷ 80 × 100 = 60%. Include only registration cohorts that have had the full seven-day opportunity when comparing the completed window.
Engagement and task success
Purpose: Check whether the activity helps.
Define meaningful engagement for the product. Opening a booking page may mean interest, confusion or an accidental visit. More time on the page could mean someone is struggling.
For task success, record a confirmed booking rather than a button click. Define whether the denominator is users, sessions or attempts. Those choices answer different questions.
For this service, a first scorecard could use confirmed bookings divided by valid booking attempts. Define an attempt as a unique booking flow started, set an expiry rule and remove internal test traffic. Track people who cannot find a slot separately from failures at payment or confirmation.
Retention and churn
Purpose: Make the population and time window clear.
Retention asks whether a defined group returns or remains a customer. A cohort is a group sharing a starting event or period. Choose a return action that reflects the product’s purpose. An appointment service may be used much less frequently than a messaging app.
Example: 25 of 100 customers who first booked in January make another confirmed booking in February. That is 25% February repeat-booking retention for that cohort, not a claim about everyone who registered.
Customer churn can be defined as customers lost during a period divided by customers at its start. If a subscription service starts with 200 paying customers and loses 10, that rate is 5%. Specify cancellation and reactivation rules. Churn is not automatically 100% minus a retention figure measured using a different cohort, period or activity.
Revenue and sustainable value
Purpose: Include the cost of providing the service.
Revenue records income under the business’s agreed accounting definition. Average revenue per user needs a clear population; average revenue per paying customer uses a different denominator. Monthly recurring revenue applies to recurring charges, not every payment a product receives.
Revenue growth can coexist with heavy discounts, refunds or costly support. Pair it with an appropriate cost or margin measure and ask the finance owner to agree the definition.
For the booking service, compare confirmed appointments with refunds and support contacts. A rise in bookings followed by a rise in avoidable cancellations deserves investigation before the team declares success.
Recommendation and satisfaction
Purpose: Hear what people say alongside what they do.
Net Promoter Score, or NPS, uses a 0–10 recommendation scale. Responses of 9–10 are promoters, 7–8 are passives and 0–6 are detractors. Calculate the percentage of promoters minus the percentage of detractors.
If 50% are promoters and 20% are detractors, NPS is +30, not 30%. The possible score runs from −100 to +100. Bain’s NPS explanation sets out the method.
NPS measures stated willingness to recommend. It does not explain a failed booking or establish why customers stay. Record response numbers, survey timing and who was asked, then follow up on reasons. A short task-specific satisfaction question may be more useful for a particular change.
Build a small scorecard you can act on
For the appointment example, begin with three views: task success, failed attempts by stage, and support contacts per 100 confirmed bookings. Treat support contacts as a guardrail, a measure that helps you spot an unwanted consequence. Avoid turning a lower contact rate into a target that makes support harder to reach.
Give every measure a definition containing the event, population, numerator, denominator, time window, exclusions, owner and decision it informs. State when a data gap makes the result unreliable. Google’s HEART research offers another way to connect user-experience goals to signals and measures.
Before reporting, check for duplicated events, missing confirmations, timezone differences, changed tracking and test accounts. Compare like-for-like cohorts and appropriate segments, such as mobile and desktop. Protect small groups from identification and avoid collecting personal attributes without a clear need.
Account for AI and delivery performance
If an AI assistant helps people book, add checks for correct appointments, corrections, abandonment, response time and cost per successful task. Include human review and failed attempts when assessing effort. Prompt counts alone cannot show that booking became easier.
Software delivery measures answer a separate question about how changes reach users. DORA’s current guidance uses five: change lead time, deployment frequency, failed deployment recovery time, change fail rate and deployment rework rate. These can inform delivery improvement, but do not establish whether a released feature solved the user’s problem.
Turn the numbers into a next step
Suppose the team simplifies the booking form and completion rises. That is a signal to investigate. Availability, seasonal demand or a change in traffic may also have affected the result. Where appropriate, design a comparison before release and agree what would justify continuing, changing or stopping the experiment.
Choose one decision for your next review. Write its outcome measure and one guardrail, check the definitions with the data owner, and inspect a few real journeys alongside the numbers. Record what you learned and the next check the team will make.
Guidance checked on 21 October 2026. The teaching example and practical exercise were written for Coded Female. Framework terms and established methods are credited below.
You May Also Like
6 ChatGPT prompts to help improve your business’s cyber security
Product Strategy: Dissecting Apple’s Winning Product Strategy
