# Funnel exploration in GA4 broken down by experiment variant, sanity check my setup

Asked by pete_h on 2025-03-26. Tags: ga4, funnels, experiments.

We are running a checkout experiment in Croct. The results panel gives me the
topline conversion per variant, which is fine, but I want to see where exactly
variant B loses people between basket and payment. The obvious tool is a GA4
funnel exploration.

Two questions. First, can the variant data Croct exports be used as a breakdown
dimension in a funnel exploration, or is it limited to standard reports? Second,
should I expect the funnel's conversion numbers to match what Croct reports, or
will there be some drift between the two

I have read the analytics documentation but it does not cover funnels
specifically.

## 2 answers

### Answer from skowalski (2025-03-26)

Yes to the first question. The exported variant assignments arrive in GA4 as
event data, and once registered they work as a breakdown dimension in funnel
explorations the same way as any other dimension. So you can rebuild your
basket-to-payment funnel and break every step down by variant. That is exactly
the kind of analysis the export is meant for.

On the second question: expect small discrepancies, and do not chase them.
GA4 drops events beyond 100k per user per day and explorations can be
sampled, so the populations will never line up perfectly with Croct's numbers.
Croct's own results are unsampled, so treat those as the source of truth for
the win/lose decision and use the GA4 funnel for the diagnostic question of
where the drop happens.

#### Reply from pete_h (2025-03-27)

That distinction is helpful. Decision from Croct, diagnosis from GA4.
Rebuilt the funnel this morning and variant B loses people at the
delivery options step. Thanks.

### Answer from Ingrid Bakker (2025-03-27)

Adding one thing before you dig deeper in GA4: Croct's results panel has
drill-downs of its own. The experiment results are Bayesian and unsampled,
and you can break them down by device, audience, time window, and custom
events. If you track an event at each funnel step and hook them up as goals,
you can often answer the "where does B lose people" question without leaving
Croct. There is a thread on
[tracking a goalCompleted event on a button click](/answers/usecroct-goal-completed-button-click)
that covers the wiring.