Two analysts use the same ecommerce events and produce different purchase funnels. One says shoppers disappear before checkout; the other shows people arriving directly at the final step. Before investigating the theme or rewriting checkout, compare the rules that decide who can enter each funnel and which paths qualify.
EcomToolkit recommends treating a funnel configuration as part of the metric definition. This guide uses original hypothetical shopper paths to explain open and closed funnels in GA4. It focuses on analytical admission and sequence rules, rather than offering a universal checkout conversion benchmark or assuming every missing step represents customer abandonment.
Table of Contents
- Choose the question before choosing the toggle
- Work through a small set of shopper paths
- Keep sequence rules visible
- Separate users from orders and event counts
- Audit a suspicious drop before proposing a redesign
- Compare devices without rewriting the population
- Publish a reproducible funnel note
- The EcomToolkit view
Choose the question before choosing the toggle
A closed funnel asks how users progress after entering at the first defined step. An open funnel allows entry at a later step. Google’s funnel exploration documentation also explains that qualification follows the specified sequence; opening a funnel does not make skipped intermediate steps disappear from its logic.
For a product-discovery question, starting with a product view may be appropriate. For a checkout-entry question, begin with checkout initiation. Those definitions describe different populations. A returning shopper following a cart link may legitimately belong in a checkout analysis without having a new product-view event in the same observed path.
Write the question in a complete sentence beside the chart. For example: among observed users who begin this defined checkout sequence, how many reach its completion step under the selected time rules? That wording is more useful than a chart simply called conversion, because it exposes both the population and the conditions.
Work through a small set of shopper paths
Consider four defined steps: product view, add to cart, begin checkout, and purchase. In this hypothetical fixture, each row is a different identified test user with one journey. All events fall inside the reporting period and any configured time limits, and there are no extra filters.
| Test shopper | Observed path | Closed funnel result | Open funnel result |
|---|---|---|---|
| A | View, cart, checkout, purchase | Qualifies through purchase | Qualifies through purchase |
| B | Checkout, purchase | Does not enter | Enters at checkout, reaches purchase |
| C | View, checkout, purchase | Counts at view, then drops | Counts at view, then drops |
| D | Purchase only | Does not enter | Enters at purchase |
| E | View, cart | Counts through cart | Counts through cart |
In the closed version, three users enter at product view and one completes the whole sequence. In the open version, later entry adds qualifying users at checkout and purchase. Shopper C still misses the cart step after entering at view; open entry is not permission to skip any step at any time.
This is a deliberately simple fixture, not a reproduction of a merchant’s report. Its purpose is to give analysts a small expected result they can calculate manually. Once the configuration matches that expectation, add more complex paths one at a time so failures remain understandable.
Do not divide all open-funnel purchases by first-step users and call that the completion rate from product view. Later entrants do not belong to that first-step population. Use the report’s defined progression measures or construct an explicitly qualified cohort before calculating a new rate.
Keep sequence rules visible
Google distinguishes steps that directly follow the previous step from steps that indirectly follow it. Direct steps require immediate succession, while indirect steps permit intervening actions. Optional time limits add another qualification condition. These choices can change counts without any change in shopper behavior.
For an ecommerce path, a delivery-information interaction between cart and checkout may be perfectly normal. If the analysis requires immediate succession, that intervening action can matter. The analyst should know whether the question concerns an exact event sequence or broader progress through a journey with other actions between milestones.
Maintain a configuration record containing the property, date range, step predicates, direct or indirect relationships, time limits, segments, and filters. A screenshot of the chart alone is insufficient when another person needs to reproduce the result next month or explain a sudden change after an edit.
Review event naming during releases. If a theme starts emitting an additional event between two previously adjacent milestones, a direct-sequence funnel can change while the visible commerce flow remains the same. That is a measurement contract issue requiring investigation before anyone claims that checkout has deteriorated.
Separate users from orders and event counts
An order report, an event-count report, and a user-based funnel need not agree. A shopper can place more than one order or repeat an action. The Google funnel documentation explains that a user enters a funnel once in the date range and that the first qualifying sequence is reported.
Therefore, a funnel is not an order ledger. Use it to understand the defined progression of observed users. Use authoritative order data for operational order totals, and reconcile purchase events separately when investigating collection quality. A mismatch is a starting point for analysis rather than immediate evidence of lost revenue.
Extend the test fixture with a user who completes two purchases and another who revisits the same step. Record the expected user-level result before looking at the interface. This prevents a repeated event from being mistaken for an extra shopper simply because the raw export contains another row.
For a broader review of business denominators, the guide to checkout abandonment metrics explains why carts, sessions, users, and orders describe different quantities. The open-versus-closed decision is one additional layer within that measurement discipline.
Audit a suspicious drop before proposing a redesign
Start with the exact transition that changed. Compare the step definitions and collection release history, then inspect controlled paths through that transition. Look for missing events, renamed parameters, new filters, or a time constraint that no longer matches a legitimate checkout route.
A product with an accelerated buying path may bypass a step used by the standard cart journey. Decide whether the funnel is intended to measure only the standard route or all eligible buying routes. If the latter, revise the definition intentionally and document the break in comparability with older reports.
| Observation | Configuration question | Evidence needed next |
|---|---|---|
| Purchases exist outside the funnel | Must users enter at step one? | Qualified paths and entry setting |
| Drop appears after a tracking release | Did predicates or event order change? | Release diff and test events |
| One checkout route disappears | Does that route emit every required step? | Route-specific fixture |
| Long consideration paths fall out | Is a time limit excluding them? | Timing distribution and rule |
| Raw events exceed funnel users | Are repeat actions being compared with users? | Metric definitions and identifiers |
Avoid changing several settings until the chart looks plausible. That approach can produce a comforting number without explaining the original discrepancy. Change one hypothesis at a time, save the configuration, and retain the expected result for the corresponding fixture.
Compare devices without rewriting the population
A device comparison is useful only when both sides use the same question and collection assumptions. A mobile route that hands the shopper into another context can have different observable coverage from a desktop journey. Examine the tracking boundary before interpreting a count difference as a design problem.
Identity continuity also matters. If steps cannot be linked to the same observed user, an actual journey may not appear as a continuous analytical path. Conversely, an identity model that joins activity changes the meaning of the population. Document the available observation rather than claiming to reconstruct every shopper perfectly.
Keep consent and collection coverage distinct from qualification. A funnel can be correctly configured for the data it receives while still observing only part of the business. Report that scope in plain language. Do not extrapolate an observed completion percentage to every customer without a justified method.
Publish a reproducible funnel note
A useful reporting note contains the business question, named entry population, ordered steps, sequence and timing rules, reporting period, and known collection limitations. Add the configuration version and any material difference from the prior reporting period. That gives the trading team context before it acts on a percentage.
When a configuration changes, preserve the previous version long enough to compare both on the same fixed period. Show how much of the difference comes from the definition rather than from new customer behavior. If the old and new series answer different questions, avoid joining them into one uninterrupted trend line.
The next action should match the evidence. A failed controlled event path belongs with instrumentation ownership. A reproducible shopper obstacle belongs with the product team. A misunderstood denominator belongs in the reporting definition. Assigning the right action is the point of the funnel, not merely producing a polished visualization.
The EcomToolkit view
A funnel tells a story selected by its entry and sequence rules. Those rules can be appropriate and still answer the wrong business question. Start with a small path fixture, make the configuration visible, and interpret losses only after understanding which users were eligible to progress.
If your purchase totals and funnel reports tell conflicting stories, request an EcomToolkit audit to review event coverage, route definitions, and reporting assumptions together.
