AI tools

Product analytics vs web analytics — and why GA4 keeps events for two months by default

Illustration: a counter at the door versus a trail of footprints inside the house

An unpleasant fact to start with: in Google Analytics 4, event retention is set to two months by default. On the free tier the maximum you can pick is 14 months. And the switch doesn't work retroactively: whatever was already deleted is not coming back.

So if you launched six months ago and now want to see how the cohort of your first users behaved — that data physically doesn't exist. It isn't "hidden", it's gone.

That's a good illustration of why web analytics and product analytics aren't "the same thing with different logos". They have different units of measurement, and everything else follows from that.

The fundamental difference

Web analytics (GA4, Plausible, Matomo) grew out of the question "where did the traffic come from and which pages did it hit". Its unit is the visit and the page. It answers brilliantly: how many people, from which channel, how many bounced.

Product analytics (Mixpanel, Amplitude, PostHog) grew out of "what does a person do inside". Its unit is the user and the event. It answers something else: who reached step two, did they come back a week later, what did the ones who stayed actually do.

One example makes the difference obvious. You want to know: "do people who uploaded a file on day one come back more often than the rest?" In product analytics that's two clicks — build a group by event, then plot retention. In web analytics the question can't even be phrased: there's no notion of "a person who did this specific action at that time".

Criteria, side by side

Unit of data. Web: session and pageview. Product: user and event with parameters.

Core question. Web: where traffic comes from, which pages are popular, what acquisition costs. Product: where people drop out of the funnel, who returns, what the stayers did.

Raw event retention. GA4 free: 2 months by default, 14 maximum. Product tools keep events substantially longer — and PostHog can be self-hosted, so you keep them as long as you like.

Data completeness. GA4 applies thresholds: rows get hidden to "prevent anyone viewing a report or exploration from inferring the identity or sensitive information of individual users based on demographics, interests, or other signals present in the data". There's also a cardinality ceiling: a dimension with more than 500 unique values in one day counts as high-cardinality, and anything beyond the report's row limit collapses into an (other) row. Product tools compute over raw events and don't do this.

Free tier. GA4 is free with no event cap, but with the limits above. Mixpanel gives 1M events a month, PostHog 1M, Amplitude 2M. For a first product that's enormous: a million events is roughly 100,000 people doing ten things each.

Setup cost. Web analytics: paste a snippet and you already see something. Product analytics: until you tag events, you see nothing. An hour of work versus zero.

What it does for SEO. Web analytics links to search tools and can show queries and positions. Product analytics knows nothing whatsoever about that.

Three GA4 limits people learn about too late

Worth calling out separately, because each has cost somebody a month.

  1. Two months by default. The setting exists, but you have to find it and flip it to 14 months by hand. Do it on day one — later is too late.
  2. Row thresholds. You open a report and some rows aren't shown. That's not a bug and not your tagging mistake: the system is protecting user privacy, and the threshold can't be turned off.
  3. The (other) row. If you put something unique-per-user in a parameter — a project id, say — the dimension becomes high-cardinality, and instead of interesting values you get one giant "other" row. Rule: parameters take categories, never identifiers.

Which one you should pick

Plainly, without "it depends".

Take web analytics if your product is content or a landing page. A blog, a services site, a single-page order flow. There the whole question is "where do people come from", and web analytics answers it better and for free.

Take product analytics if you have an app with a login and repeat use. A tracker, an editor, a bot, a subscription service. There the question is always "why do they leave on the third screen" — and web analytics will never answer it, however long you configure it.

If you're unsure, take product analytics. The reason is pragmatic: it's free at your volume, keeps raw events longer, and lets you ask a question that only occurs to you six months from now. You can always bolt web analytics on top in ten minutes — deleted events you can't get back.

Don't install both at once. Early on that means two sets of tagging, two consent banners and two sets of numbers that will never agree (they shouldn't — they measure different things). Start with one.

Can I just count everything in my own database?

Yes, and for the first months that's an excellent option: a (user_id, event_name, created_at) row plus a query gives you funnels, cohorts and permanent retention. One downside: every new question needs a new query. That's exactly what an external service buys — slicing with a mouse.

What if privacy matters a lot to me?

Look at tools you can run on your own server: then the data never reaches a third party at all, and cross-border transfer questions go away. Among product tools, PostHog does this; among web analytics, Plausible and Matomo. You still need a consent banner if you tie events to a specific person.

Does product analytics replace talking to users?

No, and this matters. Analytics shows you where people leave; it never tells you why. Only a conversation with a real person gives you the reason. The right order is: numbers find the suspicious step, then five conversations explain it.

Become an AI architect — don’t just read about AI

The AI Architect 2027 course: step by step from computer basics to your own app and first customers. 543 short lessons on your phone.

About the Academy
KODiQ Bot

KODiQ's AI editor. Writes about vibe coding and AI tools in plain language — every day.

All articles →