Put the question and the counting unit together
A funnel chart looks simple because it hides several decisions. Who can enter? What counts as completing a step? How long can someone take? Does the chart count people, visits, attempts or events? Until those rules are written down, a percentage is difficult to interpret and almost impossible to compare fairly with another report. Begin with one question and a sentence defining the population that can answer it.
For a hypothetical shop, consider: among observed users who begin checkout during a chosen week, how many complete a recorded purchase within the agreed observation window? This is a user funnel. A question about checkout attempts would require a different counting method. One person can try twice, reload a page several times and place more than one order. Those facts make users, events and orders different units.
Keep the denominator visible
Suppose a hypothetical closed funnel contains 1,000 eligible users at checkout start, 600 at the payment step and 450 at recorded purchase. The checkout-to-purchase completion rate is 450 divided by 1,000, or 45%. Payment-to-purchase completion is 450 divided by 600, or 75%. The missing 150 between payment and purchase represent 25% of payment-step users and 15% of checkout starters. Both percentages are valid when their denominators are stated.
Write counts beside rates and use the same unit in the numerator and denominator. Do not divide purchase events by checkout users and call the result a user completion rate. When reporting several segments, calculate an overall rate from the combined eligible counts rather than averaging the segment percentages. Define the empty-denominator case explicitly; no eligible users means there is no observed rate for that segment.
Make the funnel rules match the customer journey
Google's GA4 funnel documentation distinguishes closed funnels, which require entry at the first step, from open funnels, which allow entry at a later step. It also distinguishes directly following steps from indirectly following steps that permit intervening actions. These settings change which sequences qualify. Record them beside the report instead of treating the interface defaults as a definition of your customer's journey.
A hypothetical customer might read delivery information between starting checkout and entering payment details. Requiring those two events to be adjacent would exclude that valid sequence. An express checkout might also bypass a step required by the standard checkout. Decide whether to represent those journeys separately or use a shared milestone that both genuinely reach. Then inspect a few known sequences to confirm that the configured rules include and exclude the intended behaviour.
Documentation: Google Analytics: funnel exploration
Give each cohort enough time to finish
A person who starts just before your observation ends has less time to complete than someone who started several days earlier. If the analysis gives entrants seven days to finish, a cohort that began yesterday is not yet complete. Choose an entry period and an observation window deliberately, and mark incomplete cohorts. Otherwise a chart can manufacture a recent decline simply by cutting off the remaining time for people to act.
GA4 funnel steps can have a time limit between them, and its documented user-funnel behaviour reports only the first qualifying sequence when a user completes the funnel repeatedly in the date range. Check whether that behaviour suits the question. For repeat purchases, repeated applications or separate attempts, define a method that represents those repetitions instead of assuming a user funnel is an attempt counter.
Use paths to explain what happens around a step
A funnel tests progress through a predefined sequence. A path report helps investigate the actions around a selected point. Google's path documentation notes that paths can span sessions, and that Event count and Total users measure different things. Keep the selected metric visible. A branch with many repeated events is not necessarily a branch used by many different people; unique-user branch totals can also overlap.
For a custom page-type analysis, decide whether Product → Product → Cart represents two product-view occurrences or one continuous product-browsing stage followed by Cart. Both can be useful. If you collapse consecutive repeats, label the resulting units as stages or occurrences of a page-type block. Preserve sequences with no observed next stage in your denominator. Silently dropping terminal records makes continuation look stronger than it was.
Documentation: Google Analytics: path exploration
Check the query's unit before trusting its chart
When a warehouse query powers the analysis, write down what one row represents before and after each transformation. Google's GA4 export schema includes repeated event parameters and item records. Expanding an item array can create several rows from one purchase event. Counting those rows as purchases would change the unit without changing the chart's label. Reconcile a small set of known journeys before applying the query to a large period.
Define how records are ordered, how identity is grouped and how ties or missing identifiers are handled. The export documentation notes that multiple events can share a timestamp and provides batch-order fields. A sequence query needs a deliberate ordering rule, not an assumption that a timestamp is always unique. Keep ambiguous records visible in a validation count and explain whether their exclusion materially changes the result.
Documentation: Google Analytics: BigQuery export schema
Treat abandonment as an observation, then investigate
A missing later event establishes that the analysis did not observe a qualifying continuation. It does not establish why. The person may have left, chosen another route, returned outside the window or completed an action whose measurement failed. Separate a reproducible tracking defect from a plausible customer-experience problem. In a payment journey, reconcile a permitted sample with operational outcomes before calling the whole difference failed payments or lost revenue.
Break down the affected population using a specific hypothesis. If a payment redirect is suspected, compare that route with an appropriate alternative while checking their volumes and definitions. If a mobile issue is suspected, reproduce the relevant flow on the affected device conditions. A segment difference helps choose what to investigate; it does not by itself prove that the segment characteristic caused the drop-off.
Finish with a finding someone can test
A useful finding names the period, eligible population, affected transition, counts and rate. It states what was verified, what remains uncertain and the next test. For example, a hypothetical finding might report a larger observed loss after one payment route, alongside a reproducible missing return-page event. The recommended action would be to repair and verify that event, then re-evaluate the funnel before estimating a business impact.
Keep the current definition when comparing after a change, or clearly explain the revised one. Record releases, campaign mix changes and measurement interruptions that affect interpretation. If the next decision requires a causal answer, plan an appropriate experiment or stronger evaluation rather than treating a before-and-after chart as proof. The funnel is useful when it narrows uncertainty enough to choose the next action with a defensible reason.