Reporting & Explore
GA4’s reports hand you a starting point, not an answer. This module is the craft of analysis: which of GA4’s two surfaces to use, how to wield all seven Exploration techniques, how to segment so the truth surfaces from under the average, and how to spot when a number is sampled, thresholded, or quietly corrupted before you stake a decision on it.
What you will learn
Why reporting is a skill, not a screen
GA4’s reports do not hand you insight — they hand you a starting point. The skill is knowing which of GA4’s two surfaces to use, how to segment so patterns surface, and how to spot when a number is sampled, thresholded, or polluted by ‘(other).’ A great analyst with the free tier beats a careless one with 360, because the value was never in the screen. It was in the questions and the segmentation.
Most people’s relationship with GA4 is to open the default reports, read the top line, and leave. That is data consumption, not analysis. The reports tell you what happened in aggregate; the job is to find out why, and the ‘why’ almost never lives in an aggregate number. It lives one segment down.
All data in aggregate is crap.
He is being deliberately provocative, and he is right. A sitewide conversion rate of 2.3% is almost useless on its own. The same number split by device, channel, and returning-vs-new is where the decisions hide: maybe mobile converts at 0.8% while desktop hits 4% — same average, completely different action items. This whole module is about getting one segment below the average.
Two surfaces: Reports and Explore
GA4 gives you two distinct analysis surfaces, and choosing the right one saves hours. Reports are the prebuilt, fast, aggregate views for monitoring — ‘how are we doing’ at a glance. Explore (Explorations) is the freeform workbench for investigation — drag-and-drop dimensions and metrics, funnels, paths, and deep segmentation to answer ‘why.’ Reports are for watching; Explore is for thinking.
The mistake is using one where the other belongs. People try to torture the standard reports into answering a complex question they were never built for, or they jump into a heavy Exploration just to read a daily traffic number. Match the surface to the task: monitoring and trend-watching in Reports, real investigation in Explore.
Prebuilt Life-cycle (Acquisition, Engagement, Monetization, Retention) and User reports. Fast, aggregate, lightly customizable. Best for monitoring trends and answering ‘what happened’ at a glance.
A blank canvas with seven techniques (free-form, funnel, path, segment overlap, cohort, user explorer, user lifetime). Drag dimensions and metrics, layer segments, compare. This is where real analysis happens.
Shows activity right now: events, users, and sources in the last 30 minutes. Useful for confirming a launch is live or a campaign just went out — not for analysis.
The Advertising section holds attribution and conversion-path reports (Module 5 lives here). Separate from the Life-cycle reports because it answers a different question: credit, not behavior.
Standard reports and how to bend them
Standard reports are more flexible than they look. You can add a secondary dimension, apply a comparison, change the date range, and (with edit access) customize the report’s cards and default dimensions through the Library. They will not do deep multi-step analysis — that is Explore’s job — but for monitoring, a well-customized report collection beats a wall of defaults nobody reads.
The single highest-value move in standard reports is the secondary dimension. A traffic report by channel is fine; the same report with a secondary dimension of device or landing page is where a problem becomes visible. The second move is comparisons — pin ‘mobile’ against ‘desktop’ side by side rather than reading them in separate screenshots. Both take one click and change what you can see.
The tell that a GA4 account is unloved is a team that screenshots reports into slides every week. It means the reports don’t answer their real questions, so they rebuild them by hand.
I sit with the team, learn the five questions they actually ask, and customize the report Library — collections, cards, default dimensions — so those five questions are answered the moment they log in.
It converts GA4 from a place they visit under duress into the place they actually look first. Adoption is a design problem, not a training problem.
Explorations: the analyst’s workbench
Explore is where GA4 becomes powerful. It offers seven techniques — free-form (pivot tables and heatmaps), funnel (step-by-step drop-off), path (what users do next), segment overlap (how groups intersect), cohort (retention over time), user explorer (individual journeys), and user lifetime (long-run value). Each answers a different shape of question. Knowing which technique fits which question is the core analyst skill GA4 rewards.
Two things make Explore intimidating at first and trivial once they click. One: it is just dimensions (the rows, the ‘by what’) and metrics (the numbers, the ‘how much’) dragged onto a canvas, plus segments to slice and comparisons to contrast. Two: every technique shares that same grammar — learn it once and all seven open up. Start in free-form; reach for the specialized techniques when the question demands them.
Rows, columns, values, and visualizations (table, bar, line, geo, heatmap). The Swiss-army technique: most questions start here. Drag a dimension to rows, a metric to values, add a segment.
Define ordered steps (viewed product → added to cart → purchased) and see where people fall out. Can be open or closed, and can show elapsed time between steps.
Explore the sequence of events or screens forward (or backward) from a starting point. Surfaces unexpected journeys and dead ends you’d never guess.
Visualize how up to three segments overlap — e.g. mobile users, converters, and returning visitors. Reveals the valuable intersection (or its absence).
Group users by an acquisition week and track a metric (retention, revenue) across following weeks. The honest view of whether you keep the users you get.
Drill into anonymized individual users’ event streams. Powerful for debugging ‘how did this one person actually behave’ and for QA.
Lifetime metrics (LTV, lifetime engaged sessions) by acquisition source. Connects channels to long-run value, not just first-touch conversion.
Segment absolutely everything.
Funnels and paths: seeing the journey
Funnel and path explorations turn a flat conversion rate into a story. A funnel shows exactly which step loses people — the 60% who add to cart but the 12% who reach payment tell you where to spend your next dollar of optimization. A path shows the routes people actually take, including the ones you never designed. Together they replace guessing about the journey with watching it.
The discipline that makes funnels honest is segmentation, again. A single funnel hides that mobile collapses at the shipping step while desktop sails through. Build the funnel, then break it by device, source, and new-vs-returning — the aggregate funnel tells you a step is leaky; the segmented funnel tells you who is leaking and therefore what to fix.
Read that illustrative funnel like an analyst: the biggest single drop is checkout→payment (34% to 18%), so that step earns the next experiment. The lesson is not the specific numbers — yours will differ — it is the habit of letting the largest drop-off, segmented, set your priority instead of opinion.
Segments and comparisons
A segment is a subset of your data — users, sessions, or events meeting conditions — that you apply inside an Exploration to isolate a group. A comparison is the lighter-weight version available right in standard reports. Both exist for one reason: the average lies, and the truth lives in the contrast between groups. Mastering segmentation is most of what separates an analyst from a dashboard-reader.
Segments come in three scopes that mirror the data model: user-scoped (everyone who ever purchased), session-scoped (sessions that included a purchase), and event-scoped (just the purchase events). Choosing the wrong scope quietly changes the answer — a user-scoped ‘purchasers’ segment includes all their non-purchase sessions too. Decide whether your question is about people, visits, or actions, and pick the scope to match.
A segment in isolation is a vanity number. ‘Email visitors convert at 5%’ sounds great until you see everyone else converts at 6%.
So I never look at a segment without its mirror image beside it: this group vs everyone-not-in-this-group. The gap between them is the actual insight; the standalone figure is just trivia.
In Explore that means building the inverse segment too; in reports it means using comparisons. The contrast is the finding.
Everyone misses old bounce rate. But the useful question it gestured at — ‘which pages do people instantly reject’ — is answerable far better in GA4, and almost nobody builds it.
I make a segment of sessions that landed, fired no meaningful event, and left under ten seconds, then break landing pages by it. That surfaces the genuine rage-quit pages, not the long-read articles old bounce rate unfairly punished.
It is the intent behind bounce rate without the flaw — a single rich page no longer looks like a ‘bounce’ just because it didn’t load a second URL.
Sampling, thresholding, and (other)
Three forces quietly distort GA4 numbers, and a real analyst checks for all three. Sampling happens when a complex Exploration over a large date range exceeds an event quota and GA4 estimates from a subset. Thresholding hides rows for small groups to protect privacy (often triggered by Google signals). Cardinality overflow dumps high-uniqueness values into ‘(other).’ None of these are bugs — but none are flagged loudly, so you must look.
The professional habit is to glance at the data-quality icon on every Exploration before trusting it. If you see sampling, narrow the date range, simplify the query, or move to BigQuery for the unsampled raw data. If you see thresholding, consider whether Google signals is worth the hidden rows for this analysis. If ‘(other)’ is large, you have a cardinality problem upstream in your dimensions. Bad data confidently presented is the analyst’s cardinal sin; these checks are how you avoid committing it.
- How do I know if my report is sampled?
- Explore shows a data-quality icon (a green checkmark or a warning) near the top. Hover it to see whether the response is based on 100% of data or a sample, and the percentage.
- Why are some rows missing or showing as a threshold?
- Data thresholding hides rows for small user counts to prevent identifying individuals, frequently when Google signals is enabled. Reducing the date granularity or disabling signals for that analysis can reveal more.
- How do I escape sampling entirely?
- Enable the free BigQuery export and query the raw event data. BigQuery has no GA4-style sampling, which is the main reason serious analysts pair GA4 with it (Module 6).
When a big Exploration trips sampling, most people either accept the estimate or give up. There is a free middle path before you reach for BigQuery.
Sampling triggers on the volume of a single query, so I split one quarter-long Exploration into three monthly ones, each under the threshold and therefore unsampled, then add the results. Same answer, no estimate.
It is a stopgap, not a substitute for the warehouse — but on a deadline, three clean monthly pulls beat one sampled quarterly one every time.
Looker Studio and sharing out
GA4 connects natively to Looker Studio (formerly Data Studio) for free, polished, shareable dashboards. It is the right tool when stakeholders need a recurring branded view they can read without logging into GA4. Just know its limits: the GA4 connector can be slow, inherits sampling and thresholding, and can hit quota on big queries — for heavy or board-level reporting, pipe through BigQuery instead.
The decision rule is simple. For a marketing team’s weekly dashboard, the native Looker Studio connector is fast to build and good enough. For an executive report where a wrong number costs credibility, or a dashboard spanning millions of events, build it on the BigQuery export so it is unsampled, fast, and stable. Choosing the data source deliberately is part of the craft; defaulting to the easy connector for everything is how dashboards quietly lie.
Cohorts, lifetime, and predictive
Three advanced lenses turn GA4 from a rear-view mirror into something close to foresight. Cohort exploration shows whether you retain the users you acquire. User lifetime ties channels to long-run value rather than first conversion. And GA4’s predictive metrics — purchase probability and churn probability — let you build forward-looking audiences once the property has enough data to train the models.
These reframe the questions you can ask. Instead of ‘which channel drove the most conversions last month,’ cohorts and lifetime ask ‘which channel brings users who stay and spend,’ which is a far better basis for budget. Predictive metrics go further still — estimating who is likely to buy or churn next — and they feed directly into the audiences you will build in the next module.
Claim: GA4’s purchase-probability model predicts the likelihood an active user buys within the next 7 days, and requires at least 1,000 returning users who did and 1,000 who did not trigger the relevant event over 7 days to train. Source: Google Analytics Help — Predictive metrics. Context: Low-volume properties may never qualify for predictive metrics, which is a real constraint for small B2B sites.
How analysis goes wrong
Analysis fails in recognizable ways: reading aggregates without segmenting, using Reports for questions that need Explore (or vice versa), trusting sampled or thresholded numbers without checking, picking the wrong segment scope, and presenting a flattering segment with no complement. Every one is a habit problem, not a tooling problem — which is good news, because habits are fixable for free.
A sitewide rate hides the segments where the real story lives. Decisions made on aggregates are decisions made blind.
Forcing standard reports to do deep analysis, or spinning up Explorations to read a daily number, wastes time and breeds frustration.
Acting on an estimated or row-suppressed number as if it were exact leads to confident, wrong calls.
A user-scoped segment answers a different question than a session- or event-scoped one; mixing them up changes the result silently.
A standalone segment number is meaningless without its complement to compare against.
Your reporting checklist
Trustworthy analysis is a habit set, captured here. Tick what is genuinely part of how you work in this property — not what you know you should do.
Prove it. Earn your passcode.
Ten questions, CASE method (Context · Analysis · Strategy · Execution). Pass at 90% to unlock this module’s completion passcode — retake as many times as you like.