Product Analytics for Product Managers: Funnels, Retention and Event Tracking
The five questions product analytics answers, how to plan event tracking, read a funnel and a cohort table, and avoid the data traps that fool smart people.
Product analytics turns user behaviour into decisions by answering five questions: where do people drop off (funnels), do they return (retention), who behaves differently (segmentation), is a feature used (adoption), and did a change work (experiments). Plan events before you build, segment every metric, check tracking works, and define each metric in writing.
- The five questions product analytics answers
- It all starts with event tracking
- Funnels: finding where people leave
- Retention: the metric that tells you if you have a product
- A worked example: from a vague worry to a decision
- The tools, honestly
- Data quality traps that fool smart people
- Dashboards that people actually open
- Common mistakes
Product analytics is the practice of turning what users do inside your product into decisions. It is different from marketing analytics, which asks how people arrive, and from business intelligence, which asks how the company is doing. Product analytics asks a more specific question: what are people doing in the product, where do they get stuck, and what makes them stay? If you can answer that with evidence, you make better decisions faster than a team relying on opinions.
The five questions product analytics answers
Funnels show the steps toward a goal and how many people complete each one. Retention shows whether users return, usually by signup cohort. Segmentation splits any metric by group (channel, device, plan, country) to find who is driving it. Adoption measures how many eligible users use a feature, and how often. Experiments compare a change against a control group to see whether it caused an improvement.
It all starts with event tracking
You can only analyse what you recorded. Product analytics tools work on events: a named action a user took, with properties describing it. A clear, consistent naming scheme is the difference between data you can trust and a swamp nobody understands. A simple convention: object_action in lower case, for example lesson_completed, search_submitted, payment_failed.
| Event | Properties worth recording | Why |
|---|---|---|
| signup_completed | channel, platform, referral_code | Compare quality of acquisition channels |
| lesson_started | course_id, lesson_number, device | See where in a course people begin and stop |
| lesson_completed | course_id, lesson_number, score, seconds | Measure real progress, not just opening |
| payment_failed | method, error_code, amount | Find and fix revenue leaks |
| search_submitted | query_length, filters_used, results_count | Understand search success and failure |
Build a tracking plan before you build the feature
A tracking plan is a shared spreadsheet listing each event, its properties, when it fires, and who owns it. Write it while you write the PRD (see how to write a PRD). Then test each event manually before launch, and check the numbers look plausible in the first week. The most common analytics disaster is discovering, after launch, that the key event was never firing on Android.
Funnels: finding where people leave
A funnel counts how many users reach each step of a journey. The insight is almost never the overall conversion; it is which step loses the most people. Here is an illustrative onboarding funnel:
Once you see the cliff, resist jumping to a fix. Segment it: is the drop the same on Android and iOS? For every channel? On slow connections? A drop concentrated in one segment points to a specific cause; a uniform drop points to the step itself being too hard. This is the same logic as the metric-dropped playbook in the metrics interview guide.
Retention: the metric that tells you if you have a product
Acquisition can be bought; retention cannot. A cohort table groups users by when they signed up and shows what share is still active after each period. Read it across each row to see how a cohort ages, and down each column to see whether newer cohorts are better than older ones:
| Signup week | Users | Week 1 | Week 2 | Week 4 | Week 8 |
|---|---|---|---|---|---|
| Week of 5 Jan | 1,200 | 42% | 31% | 24% | 20% |
| Week of 12 Jan | 1,150 | 44% | 33% | 26% | 22% |
| Week of 19 Jan | 1,320 | 41% | 30% | 23% | 19% |
| Week of 26 Jan | 1,280 | 48% | 38% | 31% | 27% |
(Illustrative.) Two things to look for. First, does each row flatten? A curve that levels off means you have a core of users who stay; a curve that keeps falling means you do not yet. Second, is the last row better than the others? Here the week of 26 January is clearly better at every point, so something changed for that cohort, perhaps a release or a campaign, and finding out what is worth more than any dashboard.
A worked example: from a vague worry to a decision
(Illustrative numbers.) Worry: “Activation feels low.” Step 1, define it: activation means completing a first lesson within 7 days of signup. Step 2, measure: 46% overall. Step 3, segment: referral users activate at about 80%, paid-ads users at about 47%. Step 4, hypothesise: ads users arrive with less intent, or the onboarding promises something ads did not. Step 5, test cheaply: send ads users to a version of onboarding that shows the first lesson before asking for a phone number.
Notice the sequence: define, measure, segment, hypothesise, test. Most weak analysis skips the first (nobody agreed what “activation” meant) or the third (an average hid a very different story in each group). In the SQL guide you can run exactly this analysis yourself.
The tools, honestly
| Type | Examples | Good for | Watch out for |
|---|---|---|---|
| Product analytics platforms | Mixpanel, Amplitude, PostHog | Funnels, retention and cohorts without SQL | Cost grows with events; needs a clean tracking plan |
| Web analytics | Google Analytics | Traffic and marketing attribution | Less suited to in-product behaviour |
| SQL and BI tools | Metabase, Looker, Redash plus your database | Custom questions and combining data sources | Needs SQL skills and clean data |
| Spreadsheets | Excel, Google Sheets | Small datasets and quick checks | Does not scale; errors are easy to make |
The tool matters far less than the discipline. A well-planned event schema in a spreadsheet beats a fancy platform fed with messy data. Start with what you can get people to use, and switch when you outgrow it.
Data quality traps that fool smart people
- Broken tracking: an app update silently stops an event firing, and the metric “drops”. Always check the data before the product.
- Double counting: one user on two devices, or an event that fires twice, inflates numbers.
- Bots and internal users: test accounts and staff usage distort small products a lot. Exclude them.
- Changing definitions: if “active user” changes in March, comparing to February is meaningless. Write definitions down.
- Survivorship bias: analysing only users who stayed makes everything look good.
Dashboards that people actually open
Most dashboards are built once and never read. Make yours earn attention: one page per decision (“is onboarding healthy?”), no more than six numbers, each with a target and a date, and an owner who checks it weekly. If a chart has never led to a decision, delete it. A short, trusted dashboard beats a dozen impressive ones nobody trusts.
Common mistakes
Tracking everything and understanding nothing. Reading averages instead of segments. Comparing cohorts of different ages. Trusting a number without checking that the underlying event works. Building dashboards instead of answering questions. And asking analysts for “a report on engagement” without saying what decision it will inform. The better request is always specific: “between the two onboarding designs, which had higher first-lesson completion, and did it differ by channel?”
Continue with SQL for Product Managers, the metrics reference, and the North Star guide.
Frequently asked questions
What is product analytics?
The practice of measuring what users do inside a product to answer questions about funnels, retention, segments, feature adoption and experiments, so that product decisions rest on evidence.
What is the difference between product analytics and web analytics?
Web analytics focuses on traffic and how people arrive on a site. Product analytics focuses on behaviour inside the product: which steps users complete, whether they return, and which features they use.
What should a product manager track first?
A small set of events tied to your core user journey: signup, first key action (activation), repeat use and payment if applicable. Plan them before building, and test they fire correctly.
Do product managers need to know SQL?
Not strictly, but basic SQL lets you answer your own questions quickly instead of waiting on analysts, and helps you spot data problems. See the SQL for Product Managers guide.
Want this sharpened for your specific situation?
I take on a small number of career calls and resume reviews each month.
See how we could work together