On this page
Return rate alone tells you almost nothing about why a product got sent back. A jacket returned for shipping damage looks identical in your dashboard to a jacket returned because it runs two sizes small — until you build a metric that separates the two. A product-level fit score does exactly that, and it is quickly becoming the standard way fast-growing apparel brands triage their catalog.
What a Product Fit Score Actually Measures
A fit score is a composite metric, not a single data point. Where return rate just counts units sent back, a fit score isolates the portion of returns and exchanges attributable to sizing, then blends that with corroborating signals so a single noisy month doesn't distort the picture.
The core inputs are:
- Sizing-coded return rate — returns tagged "too small," "too big," or "doesn't fit" as a share of units sold
- Exchange-to-return ratio — a high ratio of size exchanges relative to refunds suggests the product is desirable but mis-sized, a very different problem than a product nobody wants
- Size-purchase skew — whether sales cluster abnormally at the extreme ends of the size curve (a signal the mid-sizes are mis-labeled)
- Review-text sentiment — mentions of "runs small," "runs large," "true to size" mined from review copy
- Size-guide engagement — how often shoppers open the size chart or fit-recommendation widget before buying that SKU
None of these signals is reliable alone. A product can have a normal return rate but terrible review sentiment if shoppers who kept an ill-fitting item simply didn't bother returning it. Combining signals is what makes the score predictive rather than descriptive.
Why Raw Return Rate Misleads Merchandising Teams
Return rate is a lagging, blunt indicator. Two SKUs can share an identical 22% return rate for entirely different reasons — one from quality defects, one from a size chart that's off by a full inch at the waist. Treating them the same wastes fix effort on the wrong lever.
Raw return rate is also slow. By the time enough units have shipped and come back to move the return-rate needle, a brand has often already reordered the same cut for the next season. A composite score built from earlier signals — size-guide engagement spikes, review sentiment drift, exchange-request patterns — can flag a problem while there's still time to adjust the pattern before the next production run. This is the same logic used in the metrics that predict a high-return product, where leading indicators consistently outperform trailing ones for catching sizing problems early.
Retail-analytics research backs this up: composite, multi-signal scoring systems consistently outperform single-metric monitoring for catching product-level quality and fit problems before they scale, a pattern documented across retail operations research from firms like McKinsey.
Building the Score: A Practical Formula
You don't need a data science team to start. A workable v1 fit score can be built in a spreadsheet or a lightweight internal tool with four weighted components.
| Component | What it captures | Suggested weight |
|---|---|---|
| Sizing-coded return rate | Direct evidence of fit failure | 40% |
| Exchange-to-return ratio | Desirability despite fit issue | 20% |
| Size-purchase skew | Distribution anomaly vs. category norm | 20% |
| Review sentiment (fit mentions) | Qualitative confirmation | 20% |
Normalize each component to a 0-100 scale against your catalog average, weight and sum them, and you get a single fit score per SKU. Products scoring well below the category median are your triage list — the SKUs where a size chart correction, updated fit photography, or garment-spec adjustment will move the needle fastest.
Setting Thresholds That Trigger Action
A score alone doesn't fix anything — it needs a threshold that triggers a workflow. Most merchants find three tiers useful:
- Green (above 70) — no action, monitor quarterly
- Yellow (40-70) — flag for size-chart review and updated fit copy
- Red (below 40) — pull from active promotion, escalate to product/design for a spec review
Teams that automate this triage — rather than reviewing scores manually once a quarter — catch problems roughly a full selling season earlier, because the score updates continuously as new orders and reviews come in.
From Score to Fix: Closing the Loop
A fit score is only useful if it feeds a repeatable process. The natural first stop for any SKU that drops into the yellow or red tier is a size-chart and fit-guide audit, following the same triage approach in how to spot a sizing problem SKU. From there, the fix usually falls into one of three buckets: update the published size chart, add or adjust fit-recommendation guidance on the PDP, or route the issue to the design/production team for a spec revision on the next run.
The highest-leverage version of this loop treats returns data as a design input, not just a customer-service cost center — a shift explored in depth in how to turn returns data into better-fitting products. Brands that close this loop systematically tend to see fit-related return rates drop within one to two seasons of consistent score-driven fixes, rather than the multi-year drift that happens when sizing issues are only addressed reactively.
Data You Need Before You Start
Building a reliable fit score requires a few data pipes to already exist:
- Order-level data joined to size and return reason
- Review text with product ID, ideally with fit-sentiment tagging
- Exchange records distinguished from straight refunds
- Size-guide or fit-tool interaction events, if you run one
If any of these are missing, start with what you have. A score built from return-reason data and review sentiment alone is still far more useful than return rate on its own — you can add exchange ratio and engagement signals once those pipes exist.
Common Pitfalls When Scoring Fit
The most common mistake is scoring at the wrong grain. Scoring at the parent-product level instead of the SKU/size level hides the real problem, since a jacket that runs true to size in Medium but small in Large will show a misleadingly moderate score when averaged across all sizes. Score at the size-variant level wherever your data allows it.
The second mistake is ignoring sample size. A SKU with eight total orders and three sizing returns will show an alarming 37.5% sizing-return rate that means almost nothing statistically. Set a minimum order threshold — most merchants use 30-50 units sold — before trusting a score enough to act on it.
The third mistake is letting the score go stale. Fit issues shift with fabric-batch changes, factory changes, and even seasonal fit trends in customer expectations, so a score calculated once at launch and never refreshed will drift out of sync with reality within a few months.
Frequently Asked Questions
What's the difference between a fit score and a return rate?
Return rate counts all returns regardless of cause, while a fit score isolates the sizing-related share and combines it with corroborating signals like exchange ratio, size-purchase skew, and review sentiment. This makes the fit score predictive of future sizing issues rather than just a historical tally of what already went wrong.
How often should a product fit score be recalculated?
Most merchants recalculate weekly or after every batch of new orders and reviews, since fit issues can emerge quickly after a fabric or factory change. A monthly or quarterly cadence works for lower-velocity catalogs, but anything slower than that risks missing a problem before it affects a full production run.
Can a fit score work without size-guide engagement data?
Yes. A usable first version can be built from return-reason codes and review-text sentiment alone, which are the two most commonly available data sources. Exchange-to-return ratio and size-guide engagement can be layered in later as those data pipes come online, improving accuracy over time.
How many orders does a SKU need before its fit score is trustworthy?
Most merchants set a minimum threshold of 30-50 units sold before treating a fit score as reliable, since smaller sample sizes produce misleadingly extreme percentages. Below that threshold, treat the score as a watch-list flag rather than a trigger for action.