How to Track Funnel Events When the Action Is Not a Page
How to track funnel events: pages where a step is a page, events where it is an action, one identity for visitor and person, and a walkthrough to verify it.
To track funnel events, use a page view where a step is a page, fire an event where a step is an action, and make sure the anonymous visitor and the identified person are recognised as one. Then walk the whole funnel yourself before you trust a single number.
You need a small number of reliable signals, not comprehensive coverage. Comprehensive coverage is how instrumentation projects turn into quarters. This post is about the minimum that makes a funnel readable, and how to know it is working.
How to track funnel events: pages, actions and the gap between them
Page views cover part of a funnel and miss the rest. The reason is that the most important moments are frequently actions rather than destinations. A form submitted, a setup completed, a first item created and a plan selected are all things a person does. None of them is guaranteed to be a page they arrive at.
So the rule is short. Where the step is a page, use the page. Where the step is an action, fire an event. A funnel built purely on page views will be missing exactly the steps that matter most, and it will look complete while doing it. That last part is what makes it dangerous: the chart still draws, the lines still slope downwards, and nothing warns you that the middle of the story is absent.
| The step is | Measure it with | Why |
|---|---|---|
| A page the person reaches | The page itself | Reaching it is the state you care about |
| A form submitted | An event | The action matters, not the screen that follows |
| A setup completed | An event | Completion is something done, not somewhere visited |
| A first item created | An event | It happens inside the product and is not a destination |
| A plan selected | An event | A choice, whatever page it appears on |
This is the practical consequence of defining steps as states. If you wrote your steps as in how to define funnel steps, you already know which states are pages and which are things a person did. The instrumentation follows from that list and not the other way round.
Start from the steps, not from the tool
It is tempting to switch on whatever the tracker records by default and build the funnel from that. Resist it. A tracker records what is easy for it to see, and what is easy to see is rarely the same as what you need.
Instead, go down your five or six steps one at a time and write beside each the single signal that tells you a person reached it. Some will be a page. Some will be an event. If you cannot name a signal for a step, you have found a gap before it cost you anything, which is the cheapest time to find one.
Keep it to five or six steps. More steps feel more precise and are in practice worse, because each one adds noise, splits your volume into groups too small to read and adds maintenance. If a step turns out to be complex inside, you can break that one down later, once you know it is the problem. Only then does it earn more events.
Keep the names consistent from the first event
Whatever you decide to track, the way you name it needs to be agreed before the first event is written. That is a topic of its own, and it is covered in how to name tracking events. The short version is that inconsistent names produce separate series that nobody can reconcile later. Decide the shape once, write it down, and come back here to wire up the signals.
Join the visitor to the person
For a funnel that continues past signup, one more thing has to be true. The person must be identifiable at the moment they identify themselves. The anonymous visitor and the known user must be recognised as the same individual.
If they are not, the funnel breaks exactly where it becomes interesting. You will see a cliff at signup that is really your instrumentation losing the thread, and you may spend weeks trying to fix a step that was never losing anyone. This is the same identity join that governs attribution, and it deserves the same kind of check: do it by hand, once, following a real person from first visit through to a completed action.
Forms are usually where the handoff happens. A form that carries the history through to the person record keeps the thread. A form that drops it leaves you with two halves of a story and no way to put them together.
Verify the whole thing by walking it
Before you trust a single number, go through the funnel from start to finish as a genuine user would. Use a device that is not yours, a browser with no history, and arrive from a real link. Then confirm that your own journey appears in the data, with every step in the right order.
It takes about twenty minutes, and it catches the errors that would otherwise cost you a month of confident decisions.
| What the walk can reveal | What it would have done to your decisions |
|---|---|
| An event that fires twice | A step would look healthier than it is |
| A step that never fires on mobile | Everyone on mobile would appear to leave there |
| A redirect that loses everything | The journey would be cut in two with no explanation |
| Tracking that never ran because consent was declined | Part of the audience would be missing from every count |
Repeat the walk on mobile, and check two things in particular: that no event fires twice, and that no step is silently missing. The second is the harder to spot, since a missing step does not announce itself. It just leaves a funnel that looks shorter than it should.
Know what consent leaves out
Consent handling needs to be in place, and you should know roughly what share of visitors it excludes. If a person declines, tracking may not run for them at all. That does not make the funnel wrong, but it does mean the counts describe the people who agreed, not everyone who came.
The point is to know the size of the gap, not to remove it. A funnel whose blind spot is understood can still guide decisions. A funnel whose blind spot is a surprise cannot be trusted when something looks odd.
What good signals let you answer
A useful test of your instrumentation is to ask what it lets you say afterwards. With the right signals you can state, for every step, how many people reached it, in the order your funnel defines, and you can follow one real person through all of them. You can say which steps are pages and which are actions without looking anything up. And you can say what the tracking cannot see.
If any of those answers is a guess, the work is not finished. It is better to have five or six signals you can vouch for than a long list that nobody has checked, because a broken funnel still draws a chart and a chart you cannot vouch for is worse than none.
Then wait before you change anything
Once the signals are verified, look at the funnel for a month before changing anything. You need a baseline and a sense of ordinary variation, or you will react to a normal Tuesday. A change made before you know the baseline can never be evaluated, and the first month is the only clean read you will get of what the funnel does when nobody is interfering with it.
During that month, record the starting conversion rate at every step, and note how much the numbers move from one week to the next. When you later see a change, you will be able to tell whether it is larger than the usual wobble. Sanity check the drop-off numbers against something you know independently, too. After that, how to find where a funnel loses people shows how to read them.
A checklist before you rely on the data
- Steps that are actions fire events, and steps that are pages use pages.
- The naming convention is written down and followed.
- The anonymous visitor and the identified user are joined as one person.
- Consent is handled, and I know roughly what it excludes.
- I walked the whole funnel from a device and network that are not mine.
- My journey appeared in the data in the correct order, on desktop and on mobile.
- No event fires twice, and no step is silently missing.
- I have a month of data before I make any change.
Frequently asked questions
Can I build a funnel from page views alone?
Only if every step is a page. Most funnels include actions, such as a form submitted or a setup completed, and a page-only funnel will silently miss them while still looking complete.
How many events do I need?
Few. One reliable signal per step, across five or six steps, is the target. More coverage adds maintenance and noise before it adds insight.
Why walk the funnel myself if the tracker is installed?
Because a tracker can be installed and still fail quietly. Events fire twice, steps do not fire on mobile and redirects lose the thread. Only a real walk shows what the data does with a real journey.
Why wait a month before changing anything?
You need a baseline and a feel for ordinary variation. Without them a change cannot be evaluated, and you may react to normal fluctuation as though it were a result.
The full method, with the checklists, is in The Funnel Playbook, a free 30-page playbook.