# Attribute filters on collection pages: +10.1% CR but -10.3% RPV

- Outcome: Flat / inconclusive
- Page type: Collection page (Ecommerce)
- Elements: Filters, sorting & site search
- Test run by: [Clean Commit](https://liftarchive.com/practitioners/clean-commit/)
- Commentary by: LiftArchive (our analysis, not theirs)
- Original source: https://cleancommit.io/ab-tests/dynamic-filtering-system/
- Page: https://liftarchive.com/test/cc-dynamic-filtering-system/

Adding a filter panel for style, colorway, text layout, text arch and city lifted conversion rate by 10.1% but cut revenue per visitor by 10.3%, so we treat the result as inconclusive. Clean Commit ran this test for a Shopify store and published it in its A/B test library. The conversion gain had an 84.6% probability of being real, short of significance, and smaller orders offset the extra buyers.

## Result

**+10.1% CR but -10.3% RPV**

- Conversion rate: +10.1%, with an 84.6% probability of beating the original. Not significant.
- Add-to-cart rate: +12.8%, and orders +6.0%. Bounce rate fell 11.5%, so shoppers engaged more with filtered pages.
- Average order value: -16.1%, and -38.2% on desktop's 1,234 visitors. Negative.
- Revenue per visitor: -10.3%. Negative, which makes the net effect on sales unclear.

## Control

Collection pages relied on fixed collections, with no way to filter by product attributes.

## Variant

A filter panel appeared on the left on desktop and as a collapsible drawer at the top on mobile. Shoppers could pick several values per filter, incompatible combinations were hidden, and choices were kept when shoppers came back from a product page.

## Why it likely came out this way (our reading)

- Filters let shoppers jump straight to the exact configuration they wanted, which helps explain more add-to-carts.
- Shoppers who found one exact match may have bought a single item instead of browsing into extra items.
- The large desktop order value drop rests on only 1,234 visitors, so part of the revenue loss may be noise.

## Caveats

Total visitors and test length were not shared, and the conversion lift did not reach significance. The same change showed a loss in a separate run (the 'attribute filters' test), so results did not repeat.

## How to run this test

- Hypothesis: If we let shoppers filter collections by the product attributes they care about, then conversion rate will rise because they find the exact item they want faster.
- Primary metric: Revenue per visitor
- Guardrail: Conversion rate
- Guardrail: Average order value
- Guardrail: Items per order
- Who should test it: Stores with large catalogs of configurable or near-identical products, where browsing fixed collections is slow. Watch order value closely, and skip it if your catalog is small enough to scan in one screen.

---

Credit: Clean Commit ran this test and owns the result. Summary and commentary by LiftArchive (https://liftarchive.com/test/cc-dynamic-filtering-system/).
Usage: Free to use, quote and build on, with credit and a link: name the practitioner or organization that ran the test and link to its LiftArchive page. The results belong to the practitioners who ran and shared them; the summaries and commentary are LiftArchive's.