There is a dropdown in GA4's admin that most businesses have never opened, and it decides how much of their history exists. Event-level data retention defaults to two months. Not two years. Two months. Every business that never changed it is continuously deleting its own detailed history on a rolling basis, and almost nobody finds out until the moment they need last year's data and reach into an empty drawer.
The reach usually happens in an Exploration. Someone builds a funnel comparison, sets the date range to fourteen months back to compare seasons, and the data simply is not there. No error, no warning about the setting that caused it. Just a shorter history than the business believed it had.
What actually expires, and what survives?
The retention setting governs event-level and user-level data: the raw material Explorations, funnels, path analyses, and segment deep-dives are built from. That is what disappears past the window.
Standard reports survive. GA4's pre-aggregated reporting (the cards and tables in the Reports section) is built on aggregate data that is not subject to the same retention clock, so your top-line trends keep their history. This split is exactly why the problem hides so well: the dashboard everyone glances at looks fine for years back, while the analytical layer underneath has a two-month memory. The property looks healthy until the first real question arrives.
If your reporting lives in Looker Studio on top of GA4, the same logic applies: standard-report-shaped queries keep history, exploration-shaped questions do not.
The five-minute fix (do it today, it is not retroactive)
Admin, data settings, data retention: set event data retention to 14 months, the ceiling on standard properties. (GA4 360 offers longer, at 360 prices.) Also worth checking the "reset user data on new activity" toggle while you are there, so returning users do not have their clock restarted in ways you did not intend.
Two honest caveats. First, the setting protects data collected from now on; it resurrects nothing already expired. The best day to flip it was at property creation, the second best is today. Second, 14 months sounds like enough until you notice what it cannot do: a year-over-year comparison in Explorations needs up to 25 months of history to compare two full same-season windows. The ceiling guarantees that some reasonable questions stay unanswerable inside GA4 itself.
When 14 months is not enough: own your history
The structural answer is the BigQuery export, which is free to link on standard properties (storage and queries cost money at scale, modest at small-business volumes). From the day you link it, raw event data lands in a warehouse you own, with no retention ceiling except the one you set. GA4's retention limit stops being a business risk and becomes a UI inconvenience.
Two things to know before treating that as a checkbox. The export starts on link day; it does not backfill, which is one more argument for doing this early, before you feel the need. And raw events in a warehouse are not reports; someone has to query them, which is a real capability step. Whether that step is justified is the same build-vs-buy judgment as does your business need a data warehouse, and for many businesses the honest answer is "link the export now as cheap insurance, learn to query it later."
For teams leaving GA4 entirely over this and related frustrations, retention is also a migration argument: self-hosted tools keep what you keep. That path is mapped in migrating GA4 to Matomo.
The honest sizing question
Before building anything, ask what your business actually loses at each horizon. A store that lives quarter to quarter and never runs cohort analysis loses little at 14 months; flip the dropdown and move on. A subscription business whose retention questions span years, or any team that compares seasons in Explorations, has a real gap that only an export closes.
What no business should accept is the default: a two-month analytical memory chosen by nobody, discovered during the first serious question. Retention settings belong on the same checklist as event verification: the boring configuration audit is exactly the kind of thing a proper implementation pass exists to catch, and it takes five minutes to be safe instead of sorry. Check the dropdown today; the data you save will be the data you need in March.