Funnels
Ordered steps towards an outcome, with drop-off and time between each. The question: which step loses the most people?
Product analytics: what to measure, and where it usually goes wrong
Product analytics records what people do inside your product as events, so you can see which paths lead to value and where people give up. This is a practical guide: the events worth sending, the five reports that answer most questions, and the traps that make dashboards look busy and say nothing.
Funnel: Sign-up to activeLedgerly · web and iOS · last 30 days
Signed up4,812
Active in week 222%
Median to first invoice1d 4h
Signed up100%
Connected bank61%
First invoice34%
Invited teammate26%
Active week 222%
Biggest leak: Connected bank → First invoice (−44%)
Sign-up isn’t the problem. The first invoice is. Almost half of people who connect a bank never send one.
Web analytics and product analytics answer different questions
Web analytics is about the visit. Product analytics is about the person, across every visit, device and platform. Most products need both, and the useful questions sit where they meet: which campaign brought people who were still active a month later?
| Web analytics | Product analytics | |
|---|---|---|
| Main question | Where did visitors come from, and what did they view? | What do people do in the product, and does it lead to value? |
| Unit | Pageviews and sessions | Events, tied to a person |
| Identity | An anonymous visitor, often per device | A signed-in person across devices and platforms |
| Time span | The visit | Weeks and months: activation, retention, expansion |
| Typical reports | Traffic, sources, top pages, campaigns | Funnels, retention cohorts, journeys, time to value |
Start with a tracking plan, not an SDK
Teams that install an SDK first end up with hundreds of events and no answers. Write down what you want to learn, then send only the events that answer it.
Name the outcomes
Three to five moments that mean the product is working: signed up, set up, reached first value, came back, paid. Everything else supports these.
Name the events
One event per meaningful action, named
object_verbin the past tense:invoice_sent,teammate_invited. Never reuse a name for something else.Add the properties you’ll split by
Plan, platform, source, template. Properties are for slicing, not for personal data: no emails, names or free text.
Identify people
Call
identifywith an opaque user id after sign-in andreseton sign-out. Now one person on a laptop and a phone counts once.Mark releases
Send a release marker on every deploy. Every chart can then answer “did this change after v4.12?”
Three calls do most of the work
Pageviews, screens and sessions are captured automatically. What the SDK can’t know is which actions matter to your business, so those you send yourself.
Keep the list short. Ten well-named events that map to your outcomes beat two hundred that someone has to decode a year later.
// object_verb, past tense, one meaning each
metrickle.track("invoice_sent", {
plan: "team",
source: "template",
});
// After sign-in: an opaque id, never an email
metrickle.identify(user.id, { plan: user.plan });
// On sign-out
metrickle.reset();The five reports that answer most questions
Retention cohorts
People grouped by when they started, and the share still active each day or week after. The question: do they come back, and is it getting better?
Journeys
The paths people actually take from a point. The question: when they don’t take the next step, what do they do instead?
Tasks and time to value
A start, a success and a time window. The question: how many people get there, and how long does it take?
Breakdowns
Any of the above split by plan, platform, source, version or accessibility setting. The question: who is it worse for?
Conversion isn’t the finish line
A sign-up or purchase is where the product starts proving itself. The people who sign up and never reach first value churn, ask for refunds, or never upgrade, and a funnel that stops at sign-up hides all of it.
Carry funnels and tasks past the conversion: onboarding, first real use, inviting someone, coming back in week two. And count conversions net: a purchase refunded or a subscription cancelled within days isn’t a win.
- Activation funnels from sign-up to the first moment of value.
- Net success: refunds, cancellations and problem reports within 7 days take a completion back.
- Retention by cohort to see whether this month’s onboarding change held up.
Where product analytics goes wrong
Tracking everything
Autocapture records every click, which is fine for friction and heatmaps but useless as a source of business events. Name your key events on purpose.
Vanity numbers
“Daily active users” means nothing until “active” is defined as an action that creates value. Opening the app isn’t it.
Averages hide groups
A healthy overall conversion rate can sit on top of a group that’s failing badly: one browser, one plan, or people using large text or a keyboard.
No release context
Without release markers, a drop on a Tuesday is a mystery. With them, it’s a diff.
Numbers without the why
A funnel shows where. Friction signals, a one-question survey and a few replays show why. Without them, teams guess, and ship the guess.
Personal data in events
Emails in user ids and free text in properties turn an analytics store into a privacy problem. Send opaque ids and categories.
Where Metrickle fits, and where it doesn’t
Metrickle covers the core of product analytics: events, identity across devices, funnels, retention cohorts, journeys, tasks, revenue and CSV export of raw events. Its focus is the part most tools leave to you: why people drop off, who it happens to, and whether the last release made it worse, across web, iOS, Android, Flutter and React Native.
- Every chart filters by every dimension: click a row and the rest follows.
- Friction, surveys and replay on the same events, so “people who scored us 2” becomes a funnel filter.
- A verdict after every release on the flows you choose to protect.
Common questions
What is product analytics?
Recording what people do inside a product as events tied to a person, then analysing those events with funnels, retention cohorts, journeys and breakdowns to learn which behaviour leads to value and where people give up.
How is product analytics different from web analytics?
Web analytics measures visits: traffic, sources and pages. Product analytics measures people over time: what they do after arriving, whether they reach value and whether they come back. Many teams use one tool for both.
How many events should I track?
Fewer than you think. Start with the events that mark your three to five outcomes and the steps between them, usually ten to thirty. Add more when a question needs them.
Do I need autocapture?
It’s useful for friction signals and heatmaps, which need every click. For business events, name them on purpose: autocaptured clicks change meaning whenever the interface changes.
Should I use email addresses as user ids?
No. Use an opaque id from your database. Emails change, they’re personal data, and once they’re in an analytics store they’re hard to get out.
How does product analytics relate to UX research?
Product analytics shows what people do at scale. UX research explains why. Metrickle puts both on the same events: tasks and funnels alongside friction, surveys and replay.
See where people get stuck, before and after they convert
Funnels, retention and journeys on web and mobile, with the friction and feedback that explain them. Free to start.
Start free