Extra orders alone do not tell you whether a shop change helps the business. If more of them are returned, refunds and handling can consume the additional revenue. Your evaluation therefore needs the retained items and remaining contribution, with enough time for possible returns to become visible.
More orders can produce a lower contribution
The following simplified example is hypothetical. Each group has 10,000 comparably defined visits, and every order contains one item. Each retained item contributes €25 after the variable costs included in the model. A fully returned item contributes none of that after refund and incurs an additional €8 in return and handling costs. Other costs are not modelled here.
| Measure | A | B |
|---|---|---|
| Orders | 100 | 110 |
| Full returns | 20 | 33 |
| Retained items | 80 | 77 |
| Simplified contribution | 80 × €25 − 20 × €8 = €1,840 | 77 × €25 − 33 × €8 = €1,661 |
The ten additional orders in B come with three fewer retained items and 13 extra returns here. The simplified contribution is therefore €179 lower per 10,000 visits. This result depends on the stated assumptions.
Allow equal time for outcomes
Yesterday’s order initially looks like a retained purchase even though it could still be returned. Comparing it with older orders that have already been returned would give the newer group an advantage simply because of when you looked. Both groups therefore need an appropriately complete observation window.
The start of that window depends on the process: with long delivery times, counting from purchase differs from counting from delivery. Agreeing its start and end before evaluation lets you keep unresolved cases visible instead of prematurely treating them as permanently retained.
Partial returns need separate treatment because a multi-item order is not simply a complete success or complete loss. You therefore need to settle whether the calculation uses items, orders or customers before comparing results. Keeping that level consistent is what makes the comparison meaningful.
Connect the report to the actual transaction
Your shop system needs a traceable relationship between refund and original order. Google documents a GA4 refund event linked to the transaction, with optional item details. Check whether your implementation records it correctly.
A return reason provides a clue rather than an automatic cause. “Does not fit” might mean measurements, cut or misunderstood compatibility. Use cleaned support questions and product data to narrow the next investigation. Personal messages do not belong in a general analytics export.
Choose the follow-up measure before changing the page
For a new size guide, confirmed purchases, retained items and documented sizing problems can together show how selection changes. To relate those measures to each other, they need the same comparison groups and an observation window defined in advance. Unresolved return cases also remain unresolved in the report.
If order and return data cannot yet be reconciled, an impact analysis can first establish the data basis and evaluation question. That work comes before deciding to retain a change permanently as part of our CRO work.
