Size and fit friction in the fashion shopping journey

Investigate size selection, fit guidance and variant availability before proposing a product-page conversion experiment.

Your question

Is the problem missing size advice, a broken selector or an unavailable size?

Put the thinking to work
CRO Agent ↗
THE SHORT ANSWER

Size and fit uncertainty can be a useful investigation area when shoppers hesitate on a fashion product page. Separate missing information, interaction problems and unavailable variants before choosing a change.

Distinguish different problems

A shopper unable to select an in-stock size faces a different problem from one who does not understand the fit. A popular size being sold out is different again. Do not label all three situations as a size-guide problem.

Review the interface state, variant data and relevant behaviour observations together. Note the specific device and product context so the proposed action addresses the observed issue.

Build a repeatable research record

For each observation, record the product URL, SKU, market, language, device, browser, date and selected size. Capture the expected action and what actually happened. Check an available size, an unavailable size and a size with limited stock. Reproduce a broken interaction before proposing an experiment; a confirmed defect needs a fix and verification.

Check whether measurements describe the garment or the body, whether centimetres and inches are labelled, and whether guidance belongs to this product or category. Test opening, closing and reading the guide with keyboard navigation and mobile zoom. Keep a missing fact, a usability defect and a research hypothesis as separate findings with separate owners.

Use accurate guidance

Fit advice should reflect the actual item and the information the brand can support. A size chart needs a clear unit and a meaningful explanation of what is being measured. Avoid treating generic model imagery as a substitute for missing product information.

When using behaviour data, follow the account’s approved access and masking setup. Look for a repeated pattern in the relevant segment rather than selecting a single dramatic recording as proof.

Define the change and guardrails

A proposal might improve access to verified size information, clarify a selection state or make stock status easier to understand. State why that change is expected to help the identified group.

Evaluate completed purchases alongside appropriate downstream context. A higher add-to-cart rate does not settle whether shoppers chose the right size or kept the order after delivery. The review should fit the product and return cycle.

Agree on the return review before launch

Define the primary purchase outcome, eligible population, assignment unit, sample plan and stopping rule before exposing visitors. Size-guide opens and selector clicks are diagnostic signals, not the final success criterion. Keep the population definition identical across variants; analysing only visitors who opened the new guide can introduce selection bias.

For the downstream review, link delivered order lines to the experiment assignment where authorised and technically reliable. State how exchanges, cancellations and unlinked orders are handled. Compare physical return rates at the same age after delivery, using returned units over delivered units for each group. This conditional rate is a quality guardrail, not a complete causal estimate: the test may also change who purchases and what they buy. Review retained purchases per assigned visitor when the data and method support it. If the return window is still immature, label that outcome pending.

Source: Microsoft Research: pre-experiment checks ↗

ILLUSTRATIVE EXAMPLE

A focused mobile hypothesis

Illustration: repeated mobile observations suggest that shoppers miss an existing, verified size guide. The proposed variant places a clearly labelled link beside the size selector; the control keeps the current placement. Eligible mobile visitors are assigned consistently before interaction. The team specifies completed purchases per assigned visitor as the primary metric, guide opens as a diagnostic and a later matched return review as a quality check. The analyst agrees the sample and decision rules before launch; the product owner checks accessibility and rollback. No improvement is claimed from this proposal alone.

Use this at work

  1. Separate usability, information and stock issues.
  2. Verify the item’s measurements and fit statements.
  3. Define the affected devices and products.
  4. Include a suitable downstream quality measure.
  5. Record the URL, variant, device, timestamp and reproducible finding.
  6. Separate a verified defect from a testable hypothesis.
  7. Agree assignment, primary metric, sample plan and decision rules.
  8. Plan the matched return review and name the owner of pending outcomes.

Reference material

Platform guidance checked 3 October 2026. Examples and working checklists are Faccelerate editorial illustrations.

Questions about this approach?

Can we promise fewer returns from a clearer size guide?

Only if the relevant evidence supports that outcome. The proposal is a hypothesis until it has been evaluated.