Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

Marketing Analytics › Google Analytics 4

GA4 Explorations

A step-by-step tutorial for GA4 Explorations: free-form tables, funnel and path explorations, segments and overlap, user explorations, sampling and sharing.
Tutorialintermediate11 min readUpdated

Before you start

  • The event schema, understood. Explorations are only as good as the events underneath them, so read the events and parameters guide first and have the tracking plan open; a funnel step is a condition on an event name and its parameters, and guessing at either wastes the session.
  • Custom definitions registered for every parameter you intend to use as a dimension or metric; unregistered parameters do not appear in the variables panel. Each registered parameter becomes a custom dimension or custom metric that explorations can use.
  • A question written down, with the KPI it relates to and the decision that depends on the answer. Explorations without a question become screenshots.
  • The retention setting checked, because explorations read event-level data and cannot see past it.
  • Familiarity with the standard reports, so that you do not build an exploration for a question the reports and dimensions guide already answers with a comparison or a secondary dimension.
  • Access to the property with a role that allows creating explorations, and an understanding of who else should be able to see the results.
  1. Frame the question and choose the technique

    Write the question as a sentence with a population, a behaviour and a comparison: "of people who viewed a product this month, what share added to basket, and does it differ by device?" The population becomes a segment or a filter, the behaviour becomes the events in the steps or the metrics in the table, and the comparison becomes a breakdown dimension. Then pick the technique from the question's shape — a share of a sequence is a funnel, a what-next is a path, a how-do-groups-differ is a free-form table with segment comparison, a how-do-groups-intersect is an overlap, a what-did-this-person-do is a user exploration. The decision table after the steps formalises this. Open Explore, start from a blank exploration, set the date range inside retention, and name the exploration with the question so that the list stays readable.

  2. Build a free-form table

    Free form is the general-purpose technique and the one to try first. In the variables panel, import the dimensions and metrics the question needs; in the tab settings, drag a dimension into rows, optionally a second into columns for a pivot, and metrics into values. Add a filter at the bottom of the tab settings to restrict the population — event name equals a value, or a dimension matches a pattern. Nest a second row dimension to see, for example, landing page under channel. Switch the visualisation to a line chart for a trend or a scatter to compare two metrics across rows. Right-clicking a cell offers to build a segment from that selection or to view the users behind it, which is how a table becomes the entry to the other techniques. Check the sampling icon before reading anything.

  3. Build a funnel exploration

    Choose the funnel technique and define the steps, each as a condition: an event name, optionally narrowed by a parameter such as a page path or an item category. Decide whether the funnel is open — people may enter at any step — or closed, where only people who completed the first step count, and the difference matters for anything that is not a strict sequence. Set whether each step must directly follow the previous one or may be indirectly followed with other events in between, and add a time constraint where the business meaning requires one. Add a breakdown dimension such as device category to see where the Funnel behaves differently, switch on elapsed time to see how long each transition takes, and use the next action panel to see what people who abandoned did instead. The trended view shows completion changing over time. Right-click a drop-off to build a segment of the people who left.

  4. Build a path exploration

    Where a funnel tests a sequence you expect, a path shows the sequences that happened. Choose the path technique and pick a starting point — an event or a page — or, more usefully for a question about what led to an outcome, an ending point such as the key event, then read backwards. Set the node type for each column: event names show behaviour, page titles or paths show content. Enable the option to show only unique nodes so that repeated page views collapse into one. Paths branch quickly and the top few nodes at each step are the only readable part; use the breakdown and the filter to narrow the population to one segment at a time rather than reading an unfiltered tree. The output is a set of hypotheses about routes worth encouraging or shortening, which the journey mapping tutorial turns into a map the rest of the business can act on.

  5. Create segments and compare them

    A segment is a subset of users, sessions or events defined by conditions, and it is the tool for every "does this differ for…" question. Open the segment builder from the variables panel, choose the scope carefully — a user segment includes everything those users did, a session segment only the qualifying sessions, an event segment only the matching events — and add conditions, with and/or logic, time windows and exclusions. Sequence conditions, where step one is followed by step two, let a segment express a behaviour rather than a property. Apply the segments to the tab as comparisons and the table or funnel shows them side by side. Segments are retroactive, which is their chief virtue over audiences; the audiences guide covers the forward-looking equivalent, and the segment builder offers to publish a segment as an audience when the analysis has earned it. Comparing the arms of an A/B test is a segment comparison of this kind, and the A/B test analysis tutorial builds one segment per arm from an exposure event.

  6. Measure segment overlap

    When two or three populations might be the same people, the overlap technique settles it. Choose segment overlap, add the segments — the technique accepts a small documented number — and read the diagram and the table beneath it, which lists every intersection with its user count and metrics. Typical questions: how many of the people who used site search also reached checkout, how many mobile users are also returning users, how far a paid audience and an email audience coincide. Right-click an intersection to build a segment from it and carry it into a free-form table or a funnel. Overlap is the fastest way to check whether an audience about to be shared to advertising duplicates one that already exists.

  7. Open a user exploration

    The user explorer lists individual users by their identifier with summary metrics, and clicking one shows a timeline of their events with parameters. It is the technique for debugging rather than for reporting: tracing why a funnel step shows an impossible count, confirming what a particular test session sent, or checking what a user who filed a complaint actually experienced. Reach it from the technique list or by right-clicking a cell in another exploration and choosing to view users, which carries the selection across. The same screen allows a user's data to be deleted, which is the mechanism for honouring a deletion request. Treat it as sensitive, limit who looks at it, and never paste a timeline into a shared document.

  8. Read the sampling and threshold indicators

    Before any result leaves the screen, check the icon at the top of the tab. Explorations are sampled when the query would exceed the documented event quota for the date range, and the icon reports whether the result is based on all data or on a sample, with the share shown on hover. Narrow the date range, simplify the segments or reduce the breakdown to bring the query under the quota; the paid tier offers a more detailed results option. Thresholds are different: when Google signals or demographic data are involved, rows that could identify individuals are withheld, and the indicator says so. A row that is missing is not a zero. Note on the exploration itself whether the figures were sampled or thresholded, because the person reading a screenshot next month will not see the icon.

  9. Share, export and schedule the result

    An exploration is private to its creator until shared. Sharing makes it visible, read-only, to other users of the property, who can duplicate it to make their own edits; it does not grant access to the property itself. Export the table as a file or to a spreadsheet when the numbers need to travel, and export a PDF when the visual does. Google does not provide a native schedule for explorations, so for any view that should arrive weekly, rebuild it as a Looker Studio report on the GA4 connector, which can be scheduled and delivered, and keep the exploration as the analysis behind it. The conversion analytics hub covers turning this kind of analysis into a standing conversion review rather than a one-off. The comparison of GA4 explorations and Looker Studio sets the two side by side, from who reads the result to sampling, sharing and scheduling.

Choosing the technique

Question shapeTechniqueWatch for
How does metric X break down by dimension Y?Free formSampling on long ranges; thresholds when signals are involved
What share of people who did A went on to do B and C?Funnel explorationOpen versus closed; direct versus indirect steps
What did people do before or after an event?Path explorationUnreadable without a filter or segment; unique nodes on
Do two groups behave differently?Free form with segment comparisonScope of the segment (user, session, event)
How many people belong to several groups at once?Segment overlapSmall documented number of segments per overlap
What did one user or test session actually send?User explorerSensitive; debugging only
Which cohort retains better?Cohort explorationReads from first-touch events; retention setting
Which group is worth remarketing to?Segment, then publish as audienceAudiences are not retroactive

The funnel and cohort rows answer different questions, one where a journey loses people and the other whether the people it brought stay, and the comparison of cohort and funnel analysis explains when to read each and how they check one another.

Illustrative example — not a real result

An online course provider's enrolment funnel

An online course provider wants to know where prospective students drop out between a course page and enrolment, and whether it differs on mobile. A closed funnel is built with four indirect steps — view_item, add_to_cart, begin_checkout, purchase — broken down by device category over the last six weeks, with elapsed time shown. Of 48,200 users who viewed a course page, 9,160 added it to the basket, 4,310 began checkout and 2,590 enrolled: a completion rate of 5.4% overall. On desktop the checkout-to-purchase step keeps 71% of people; on mobile it keeps 49%, and the elapsed time on that step is almost three times longer. A segment of mobile users who abandoned at checkout, carried into a path exploration ending on begin_checkout, shows a large share returning to the course page and the pricing page, which points at the mobile checkout hiding the instalment option. The segment is published as an audience for a reminder campaign while the checkout is fixed.

The same free-form technique is the main tool on the diagnosis page for a sudden traffic spike, which uses it to separate automated visits from genuine demand. The site search analysis tutorial applies the free-form, segment and path techniques to the terms visitors type into a site's search box.

Verify it worked

  • Check the exploration's name and date range. The name should state the question, and the range should sit inside the retention setting.
  • Check the sampling icon. Note on the exploration whether the figures rest on all data or on a sample.
  • Compare one free-form total with the matching standard report. Any difference should be explained by sampling, thresholds or the date range.
  • Read down a closed funnel. No step should show more people than the one before, and the completion rate should be the last step divided by the first.
  • Open the path exploration. It should end on your chosen outcome, show unique nodes and be filtered to one segment, so the top branches are readable.

If it goes wrong

  • Results are empty or cut off for older dates. The date range reaches beyond event data retention. Move the range inside it, because a longer setting does not recover expired data.
  • The numbers differ from the standard report. Check the sampling icon, the date range and any thresholds before blaming either figure.
  • A parameter is missing from the variables panel. Its custom definition is not registered. Register it, then import it.
  • A segment's conversion rate looks too high. The scope is probably set to users when sessions were meant. Rebuild the segment with the right scope.
  • A funnel counts people at later steps who skipped earlier ones. The funnel is open. Close it, or read it as open.

Next steps

Common mistakes

  • Starting without a question. An exploration opened to "see what the data says" produces a table nobody acts on.
  • Ignoring the sampling icon. Figures from a sampled exploration quoted as exact, then contradicted by the standard report.
  • A date range beyond retention. Empty or truncated results blamed on the tag.
  • Wrong segment scope. A user segment where a session segment was meant, so every session of a converting user looks like a converting session and the conversion rate (CVR) is wrong.
  • Open funnels read as closed. People entering at step three counted as if they had passed steps one and two.
  • Unfiltered path explorations. A tree of every page on the site, read as if the top branch were the typical journey.
  • Treating the user explorer as a report. Individual timelines pasted into decks.
  • Rebuilding the same exploration weekly. A recurring view belongs in a scheduled dashboard, with the exploration kept for the analysis.

The GA4 hub places explorations in the wider model; the reports and dimensions guide covers the standard reports that should be exhausted before an exploration is built.

Frequently asked questions

Why does an exploration show different numbers from the standard report?

Standard reports are built from pre-aggregated tables and explorations query event-level data, so they can differ when sampling applies to the exploration, when the exploration's date range reaches beyond the data retention period, or when thresholds hide rows in one and not the other. Metric definitions can also differ subtly between scopes. Check the sampling indicator and the date range before assuming either number is wrong.

How far back can an exploration look?

Only as far as the property's event data retention setting allows, because explorations read event-level data and that is what the setting governs. A property on the shorter option will return empty results for older dates even though the standard reports still show them. Set retention to the longer option early; the change does not recover data that has already expired.

Can I schedule an exploration to be emailed?

Not from within GA4; explorations can be shared with other users of the property and exported by hand, but Google does not provide a native scheduled delivery for them. For a recurring version of the same view, rebuild it in Looker Studio on the GA4 connector or pull it through the Data API, both of which can be scheduled. Keep the exploration for the analysis and the dashboard for the distribution.

Terms used on this page

Where this page is used

Part of these learning paths

Also related

Sources

  1. [GA4] Get started with Explorations — Google Analytics Help
  2. Google Analytics Help — Google Analytics Help

Related