Short answer
For a GA4 funnel analysis, first write the 3 to 5 steps of the path and check in DebugView that every step has an event. Then build a funnel exploration in Explore, choose open or closed, split it with breakdowns and segments, and turn the biggest drop between steps into one hypothesis.
Picture this: on a services website, the number of quote-form enquiries has been falling for two months. In the meeting everyone has a guess; one person blames the ads, one the length of the form, one the season. Behind none of the guesses is an answer to “at which step of the path do people drop out?” A GA4 funnel analysis answers that question. According to Google’s help page, funnel exploration visualises the steps users take to complete a task and lets you see quickly how well they are succeeding or failing at each step.
In this article we go in order, from designing the funnel on paper to reading the table and reaching one hypothesis. For the general logic of the reports, see what GA4 is and how to read its reports; for the checkout steps of an online shop, see how to reduce cart abandonment. Here is the funnel technique that works on any site. For a short definition of the concept, see the conversion funnel glossary entry.
1. Write the path on paper first
Before you open the screen, work backwards from the goal action and write the steps of the path. Each step should be a single action a visitor takes that can be measured. Google says a funnel can have up to 10 steps; 3 to 5 steps are enough for a first pass (this is our suggestion). The example in this article is a services website, and it is an illustration from start to finish: viewing the service page, clicking the form, starting the form, successful submission.
If you have a written measurement plan, the steps are already in it. If not, this list becomes its first page; in Detmore’s data and CRO work the order is the same: first write down what an event means, then look at the funnel data.
2. Check that every step has an event
According to Google’s help page, the conditions for a funnel step can be based on events users trigger or on dimension values they share; steps cannot be based on metrics. Our inference: if a step’s event never arrives, that step cannot be measured in the funnel, and the result will be empty or misleading.
Do this before you build:
- Turn on debug mode for your own device with Tag Assistant. Google’s DebugView page suggests tagassistant.google.com or preview mode for this.
- In GA4, open DebugView under Admin, Data display; it shows events in real time as they are triggered.
- Walk the path yourself from start to finish and tick off that the relevant event appears in the stream at every step.
One warning: Google writes that events are not visible in debug mode if consent mode is implemented and the user has not given consent for Analytics cookies. To fill in missing events, go back to the GA4 setup checklist. One more detail: a button click and a successful submission are not the same thing; tie the last step to the event that shows the submission actually happened (our suggestion).
3. Build the funnel
The steps below use the interface names from Google’s English help page.
- Open Explore; under Start a new exploration, choose the Funnel exploration template.
- Click the edit pencil. Give each step a distinctive name and add the step’s condition. To add more clauses to a condition, you can use AND or OR.
- Add the following steps with Add step. For each new step, choose whether the previous step is indirectly or directly followed by it: with indirect steps other actions can happen in between, while a directly followed step must occur immediately after the previous one.
- If you wish, use Within to require that a step happens within a set time frame.
- Look at the Summary card on the right; according to Google, it shows how your conditions match your data.
- Choose the visualisation: a standard (stepped) funnel or a trended (line chart) funnel.
The date range also affects the result. According to Google’s getting-started page, an exploration saved with a preset range (for example, the last 28 days) shows the last 28 days as of the day you reopen it. If it is saved with a custom range, the range stays fixed. If you will compare the funnel before and after a campaign, use a custom range (this is our suggestion).
4. Open or closed: decide who enters the funnel
The “Make open funnel” toggle determines how users can enter the funnel. Google’s definition: in an open funnel users can enter at any step; in a closed funnel users must enter at the first step. In both cases users are counted only in the steps they complete in the specified sequence, a user who misses a step is not counted in any later step, and a user can enter the funnel only once.
We have summarised Google’s second example in the figure below. A, B and C are the funnel steps; five users complete them in different orders.
The numbers in the figure come from Google’s example. In the closed funnel, users 1, 3 and 5 are counted at step A. In the open funnel, user 5 is not counted at step A because they entered at step B. Google explains it this way: an open funnel places the user in the funnel with their first qualifying action and looks at what follows. Going back to step 1 after entering mid-funnel is ignored in an open funnel, while in a closed funnel the user is counted at the first step. Our inference: in an open funnel a later step can show more users than the first step (in the figure, 4 users at step C and 2 at step A). So choose a closed funnel if the entrance to the path is fixed and an open one if people can also join the path in the middle, and do not compare the two funnels as if they were interchangeable.
5. Split it with breakdowns and segments
A single total hides where the problem is. Google’s help page offers two ways to split the data.
- Breakdown. When you select a dimension, each step is split by that dimension’s values; for example, with “Device category” you see desktop and mobile. The important detail: users are attributed only to the first instance of the breakdown value that applies to them. A user who enters the funnel on mobile and completes every step on desktop appears in all steps, but only under the mobile breakdown. In a trended funnel, users are attributed to the first date on which they were active in each step.
- Segment comparisons. You can apply up to 4 segments to focus on specific sets of users; Google says you can also create a segment by right-clicking in the data table.
Our order: device breakdown first, then a segment by traffic source or by new and returning users (this is our suggestion).
6. Elapsed time and next action
Two settings give clues about the cause of a drop.
Show elapsed time. When you turn it on, the average time elapsed between steps appears. According to Google, for each step this is the sum of the time elapsed between the previous and the current step, divided by the number of users who reached the current step. If the same event occurs several times, only the first is counted. Our reading: if the average time between two steps is long, users may be hesitating at that step; it is a clue, not proof.
Next action. When you select a dimension, the “Top 5 Next Actions” card shows the most common actions after each step. The “(no next action)” entry shows that a significant portion of users who completed the step did not go on to the next one. But Google writes that when page or screen dimensions are selected this does not always mean a drop-off in the funnel, because the dimension value may have stayed the same.
7. Read the drop between steps
In the funnel table, look at the number of users at each step and write down two figures: the share that continued from the previous step and the number of people lost at that transition. The table below is an example; the numbers are round and invented, and do not come from any real site. In the services-website example, let 1,000 users enter the path.
| Step | Users | Continued from previous step | People lost |
|---|---|---|---|
| 1. Viewing the service page | 1,000 | start | start |
| 2. Clicking the form | 400 | 40% | 600 |
| 3. Starting the form | 300 | 75% | 100 |
| 4. Successful submission | 60 | 20% | 240 |
From start to finish 60 of 1,000 remain, which is 6%; the people lost, 600 + 100 + 240 = 940, equal the difference between 1,000 and 60. The two views point to two different transitions: the largest loss in people is at the first transition (600), and the lowest continuation rate is at the last (20%). Our reading: the 300 people who started the form showed intent, so the last transition is often the more telling clue; some of those lost at the first transition only glanced at the page.
8. Why the numbers differ from standard reports
Funnel totals may not match the numbers in standard reports exactly. Google writes that the differences between reports and explorations are expected, and lists the reasons. The table below collects the relevant ones.
| Reason | What Google’s help page says |
|---|---|
| Supported fields | Some dimensions and metrics in reports are not supported in explorations; those fields are dropped from a report opened in explorations. |
| Comparisons and segments | Comparisons in reports are converted into segments in explorations; unsupported fields are not included in the segment. |
| Date range | Date ranges in explorations are limited to the property’s data retention settings. |
| Low user count | Data may be withheld in a report or exploration when the user count is low (data thresholds). |
| Behavioural modelling | If it is enabled for consent mode, minor differences can appear between standard reports and explorations. |
| Processing time | Queries for the past 48 hours may give slightly different results. |
Two more details. The getting-started page says an exploration may be based on sampled data if more than 10 million events are part of a particular exploration query. It also says that the icon in the right corner changes from green to yellow when an exploration is subject to sampling or data thresholds. The data retention page says the retention setting does not affect standard aggregated reports and only affects explorations and funnel reports; the options for user-level data are 2 months and 14 months. According to the getting-started page the default is 2 months. We explained how to change the setting in the setup checklist.
9. Turn the biggest drop into one hypothesis
This part is entirely our own method; it is not in Google’s pages. Pick the transition that stands out in the table, then write a one-sentence hypothesis: “Visitors who start the form leave without submitting because of [reason]; if we [change], the submission rate will rise.” An example hypothesis: “Some people who start the form give up because of the number of required fields; if we reduce the fields, the submission rate will rise.” For evidence of the cause, look at behaviour data such as recordings and heatmaps; the funnel shows where, not why.
Once the hypothesis is written, the work moves on to testing. For the next steps, see how to increase conversion rate and how to run an A/B test; the A/B test sample size calculator shows the number of visitors you need. From the measurement plan to the tests, we carry out this work in our data analysis and conversion service.
Common mistakes
- Building the funnel on screen by trial and error without writing the steps first.
- Building a funnel without seeing the events in DebugView, then reading an empty step as “nobody gets there”.
- Comparing open and closed funnel results as if they were the same thing.
- Reading the device breakdown as “the last device used”; the breakdown shows the value the user had when they first entered the funnel.
- Treating a “(no next action)” entry as abandonment in every case.
- Asking about last season while the retention period is short.
- Trying to match the funnel exactly to a standard report.
- Producing a fix the moment you find the drop; prove the cause first.
Common questions
Should I choose an open or a closed funnel?
Google’s definition: in an open funnel users can enter at any step, and in a closed funnel they must enter at the first step. Our suggestion is to start with a closed funnel if the path has a single entrance (for example, the service page). If people also join the path in the middle through direct links, we suggest trying an open funnel as well.
Why does the funnel not match the number in standard reports?
Google writes that differences between reports and explorations can arise from supported fields, segments and comparisons, the date range, data thresholds, behavioural modelling and processing time, and that these are expected differences. The table above has all of them.
Can I look at last year’s funnel data?
The date range in explorations is limited by the property’s data retention setting. According to the getting-started page the default is 2 months; the options for user-level data are 2 and 14 months. If you increase the period, the new period applies to data that was already collected and has not yet been deleted.
One thing you can do today
Write the 3 to 5 steps of the path on a sheet of paper, open DebugView, walk each step yourself and tick off each step’s event. When all of them show up, open the Funnel exploration in Explore, watch the last 28 days as a closed funnel with a device breakdown, and write next to each transition how many people were lost.
Sources
This article is general information, not legal, financial or technical advice. Platforms, their rules and the law change over time, so some details may have changed since 14 November 2026 or contain small errors. Before you make a decision for your business, check the current official sources or write to us.