Stop 04 · Data & Metrics for PMsTopics · 12/12

Defining Success Metrics for a Feature

In short

If you don't decide what success looks like before launch, any result can be spun as a win. Define the metrics first, then build.

Before building anything, deciding in advance what 'this worked' will look like numerically — so you're not tempted to retroactively pick a metric that makes the launch look good.

The purple blob as a judge tastes a dish with a spoon at a cooking contest table.

Think of it like this

The cooking competition

Judges agree on scoring before anyone lights a stove: taste, presentation, time. If they picked rules after tasting, they'd just reward their favourite. Agreeing first keeps everyone honest, including you.

Watch it happen

Metrics for a "Reorder in one tap" feature

One-tap reorder

Primary

Secondary

Guardrail

Repeat orders/wk

Time to order

Feature use %

Wrong-order refunds

1/5Start from the goal: make it easier for regulars to order their usual meal again.

The big ideas

What you need to know

One primary metric

Pick one number that decides success. More than one lets you cherry-pick the winner later.

Secondary metrics explain

Adoption and usage show whether people used the feature and how.

Guardrails protect

Name what must not get worse, like refunds, crashes or page load time.

Set target and timing

Write the expected change and when you'll check it, before launch.

Save yourself the pain

Common mistakes

Oops

Picking "number of people who clicked the new button" as the success metric.

Do this instead

Clicks show adoption. Success is the outcome the feature was meant to change.

Oops

Skipping guardrails because the feature seems harmless.

Do this instead

Many features hurt something nearby. A guardrail catches that before it spreads.

Your mission

Write a metrics plan

~20 min

Quick check

Why add a guardrail metric to a feature launch?

Go deeper

1Pick a feature idea and write its success metric and target threshold before any data exists.
2Separate your primary metric from 2-3 guardrail metrics (things that shouldn't get worse).
3Practice defending your chosen metric against the question 'why not measure X instead?'