A filter helps your customers when it excludes products that will not suit their task. Someone choosing a laptop bag needs to narrow the selection by compatible device size, whereas an internal purchasing group tells them little. The filters your shop needs therefore follow from the selection questions in each category.

Write selection questions before naming attributes

You can begin with one category and gather questions from sales advice, support and product information. “Will my device fit?”, “Can I use this outdoors?” and “Does it work with my existing mounting?” reveal what the selection needs to achieve before you name the technical attributes.

Turning that question into a filter requires suitable data. If the answer is buried in free text, it first needs to become a consistent attribute. Its meaning also needs to be clear: “15 inches” might describe an approved device size or merely come from the product description. Only the former directly answers the suitability question.

A worksheet can place the question beside its field, allowed values, unit and missing information. This reveals whether filtering is already possible or catalogue work comes first.

Define how multiple values combine

Baymard’s filtering research discusses category-specific attributes and understandable multiple selection. Your design still needs to specify the combination rules.

Selecting “Blue” and “Green” usually means either colour. Selecting “Waterproof” and “Fits a 15-inch device” may require both conditions. Define each filter’s meaning and check it against suitable products.

Separate filtering from sorting too. “Under €100” excludes products. “Lowest price first” orders the remaining results. Sorting cannot replace an essential suitability criterion.

Missing data does not mean “not suitable”

A product without maintained laptop dimensions may or may not fit. If it disappears after filtering, the interface may imply more certainty than the catalogue supports.

Agree how to handle missing values. This may require data cleanup before launch. An extra “Information unavailable” option only helps if customers can use it meaningfully; it does not replace maintaining important properties.

Result counts must reflect the current selection. Show which conditions are active and how to remove each one. On mobile, this feedback needs to remain part of the flow after the filter panel closes.

Check real combinations

Deliberately chosen catalogue items let you check whether the rule produces the intended selection. A suitable item, a known mismatch and a missing value reveal different limits of the rule. Repeating the selection through navigation and search should not change its meaning.

Then observe someone performing a realistic product-selection task. They should know their requirements without having to learn your internal attribute names. Record the labels or ranges that cause difficulty.

One category and its unresolved selection questions give a shop diagnosis a concrete basis. The CRO investigation can then distinguish missing data from a selection that is difficult to understand. Whether search engines should crawl filter pages is a separate decision, covered in reconciling Shopify filters with robots.txt.