GA4 funnel and path explorations: when they lie to you

Short answer: an exploration and a standard report can disagree about the same funnel without either being broken, because they are built from different tables under different rules. Explorations query raw event and user-level data, so they sample above 10 million events, they stop at your data retention window, and they get rows withheld by thresholding — while standard reports read pre-aggregated tables that ignore the retention setting entirely (Analytics Help, checked 13 August 2026). Before you rebuild anything, check the data quality icon, the retention setting and the open/closed toggle. Three settings explain most of the discrepancies people escalate.

Sampling starts at 10 million events and announces itself in an icon you've never clicked

The threshold is a query limit, not a property limit, so a busy month can sample while the quarter beside it doesn't.

Google's rule: sampling "may occur when the number of events used to create a report, exploration, or request exceeds the quota limit for your property", and "the quota limit for event level queries is 10 million events for standard Google Analytics properties and up to 1 billion events for Google Analytics 360 properties" (Analytics Help, checked 13 August 2026). Analytics then uses a portion of the data and scales up to give what it calls directionally accurate results. When that happens, "it is indicated in the data quality icon with the percentage of data used to create the results".

That icon is the whole warning. No banner, no colour change on the numbers, nothing in an exported PDF that stops a reader treating 9,214 conversions as a count rather than an estimate. On a 360 property, the default is tighter than the ceiling — 100 million events per query, with a "more detailed results" option to reach the higher limit.

One documented trap catches people who assume standard reports are immune: "filtering large datasets by country may result in data sampling in standard Google Analytics reports and Explore, even when using expanded data requests" (Analytics Help, checked 13 August 2026). A country filter is the most common thing a UK marketer adds to a global property, which is an unfortunate coincidence.

Row limits produce a second class of disagreement that looks identical from the outside. The (other) row "appears in a report, exploration, or Data API response when the number of rows in a table exceeds the table's row limit", with Analytics keeping the most common dimension values and condensing the rest; any dimension with more than 500 values "should be considered a high-cardinality dimension" (Analytics Help, checked 13 August 2026). Your totals stay right. Your breakdown quietly stops being a breakdown.

Explorations can't see further back than your retention setting, and standard reports can

This is the difference that produces the most alarming version of the discrepancy: a report showing a healthy year and an exploration showing a cliff.

Google's data retention control governs "user-level and event-level data associated with cookies, user-identifiers, such as User-ID, and advertising identifiers", with two months or 14 months available on standard properties. Then the sentence that explains the cliff: "the data retention setting does not affect standard aggregated reports (including primary and secondary dimensions) in your Google Analytics property, even if you create comparisons in the reports. The data retention setting only affects explorations and funnel reports" (Analytics Help, checked 13 August 2026).

Two months is the value a property carries unless somebody changed it. Build a six-month funnel on a property still on the default and the first four months contribute nothing, with no error and no shaded region on the chart. Google spells out the same behaviour for reports opened in Explore: "if you create a report with a date range outside the user and event level data retention settings, and then open it in explorations, data prior to that range won't be included" (Analytics Help, checked 13 August 2026).

The fix takes about fifteen seconds in Admin, and it isn't retroactive. Set retention to 14 months today and you get 14 months of exploration data starting from what you still have, not from what you deleted last spring.

Thresholding takes rows out and leaves no hole where they were

Suppression is applied at query time to protect individuals, which means the same segment can appear one week and vanish the next.

Google applies data thresholds "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", and is direct about your options: "Data thresholds are system defined. You can't adjust them" (Analytics Help, checked 13 August 2026). The trigger is a low user count in the date range, and demographic dimensions or audiences built on them are the usual cause. Google's own advice is to widen the date range, because a narrow one with low user counts is exactly the shape that gets withheld.

For anyone segmenting hard, this is the failure mode to internalise: a segment that returns zero and a segment that was suppressed look the same on screen. The message appears in the data quality indicator at the top, not next to the row that went missing.

Google names one escape and one catch with it. Exporting to BigQuery avoids thresholding, and "Analytics doesn't export data from Google signals to BigQuery", so "you may see different event counts per user between Analytics and BigQuery" (Analytics Help, checked 13 August 2026). You trade a suppression problem for a reconciliation problem. That's usually a good trade, and it is not a free one.

Open and closed funnels answer different questions, and the toggle is one click

The toggle changes who is allowed into step one, which changes every conversion rate below it.

Google's definitions are short enough to quote whole: "In an open funnel, users can enter the funnel in any step. In a closed funnel, users must enter the funnel in the first step" (Analytics Help, checked 13 August 2026). Beyond that, "users are only counted in the steps they complete in the specified sequence", and if someone misses a step "they fall out of the funnel and aren't counted in any subsequent steps".

The behaviour that catches people out sits in the back-navigation case. If a user starts mid-funnel and then goes back to the first step before continuing, a closed funnel counts them in step one, "because closed funnels look for the first instance of step 1 in the user event stream". An open funnel does not: it "inserts the user in the funnel with their first qualifying action", and if the user later navigates back to step 1, "the action is ignored" (Analytics Help, checked 13 August 2026). Same visitor, same events, two different step-one totals depending on a setting most people never open.

One more rule worth knowing before you present a funnel to anyone: if a single user completes the steps several times in the date range, "only the first sequence of steps will be reported". A funnel exploration counts people entering a journey, not the number of times the journey happened, which is why it will never reconcile with an event count.

Path explorations truncate on a similar principle. The graph "shows the top 5 nodes in a step" by default, expandable to 20, with everything beyond that grouped into an "Others" node, and paths are calculated "using the first instance of the dimension value you select as the starting or ending point" (Analytics Help, checked 13 August 2026). A long tail of pages doesn't appear as a long tail. It appears as absence.

What to check before you escalate a discrepancy

Open the data quality icon on the exploration first. It carries the sampling percentage and the thresholding message, and it answers two of the four causes in one click.

Then read Admin for the retention setting. If it says two months and your exploration covers longer, you already have your answer and no amount of rebuilding the funnel will change it.

Then open the funnel's settings and confirm whether it's open or closed, and whether the report you're comparing it to counts events rather than users entering a sequence. If all of that checks out and the numbers still differ, the remaining suspect is who GA4 thinks your users are — the reporting identity, and whether User-ID is in play, which changes user counts without changing a single event. Untangling that chain, from the consent signal through to what the bidding platform receives, is what we do.

Sources

1. 1.Analytics Help — [GA4] About data sampling · Checked 2026-08-13
2. 2.Analytics Help — Data differences between reports and explorations · Checked 2026-08-13
3. 3.Analytics Help — [GA4] Data retention · Checked 2026-08-13
4. 4.Analytics Help — [GA4] About data thresholds · Checked 2026-08-13
5. 5.Analytics Help — [GA4] Funnel exploration · Checked 2026-08-13
6. 6.Analytics Help — [GA4] Path exploration · Checked 2026-08-13
7. 7.Analytics Help — [GA4] About the (other) row · Checked 2026-08-13