On this page
Stop Guessing About Fit, Start Testing
Most brands change sizing content the way they change anything else — on instinct, then assume it helped. But fit changes are uniquely measurable, and uniquely worth measuring, because they affect both conversion and returns. A/B testing turns "we think the new size guide is better" into "the new size guide cut fit returns by X% and lifted conversion by Y%." This guide covers how to run sizing experiments that actually prove their case.
What's Worth Testing
The highest-value sizing experiments include:
- Size guide vs size recommendation — does a recommendation outperform a static chart, the question behind Size Chart vs Size Recommendation?
- Placement — sizing help at the size selector versus in a tab.
- Fit content — adding model height, fit notes, or "runs small" signals.
- Imagery — fit-focused photography on multiple bodies.
- Recommendation UX — how the tool is introduced and how its output is shown.
Start with the changes most likely to move the number — usually placement and the chart-versus-recommendation question.
The Metrics That Matter
Sizing tests are different from typical CRO tests because the headline metric — conversion — only tells half the story. A change that lifts conversion but raises returns can be a net loss. Track, per variant:
- Conversion rate and add-to-cart.
- Fit-related return rate — the critical second metric.
- Bracketing rate — multiple sizes per order.
- Net revenue after returns — the metric that ties it together.
Optimizing conversion alone is how brands accidentally make returns worse, which is why the full metric set from How to Calculate and Track Your Return Rate belongs in every sizing test.
The Return-Lag Problem
Here's what makes sizing tests trickier than button-color tests: returns lag the sale by days or weeks. A test that looks like a conversion win on day three may show higher returns by week four. You must run sizing experiments long enough to let the return window close, or you'll draw conclusions from incomplete data. Build the return lag into your test duration explicitly — ending too early is the most common sizing-test mistake.
Designing a Valid Test
Sound experiment design still applies:
- One variable at a time where possible, so you know what caused the change.
- Adequate sample size — fit effects can be smaller than UX effects and need enough traffic to detect.
- Clean split — randomized and consistent, with no contamination between variants.
- Pre-registered success criteria — decide what "win" means before you start, the same discipline as the pilot in How to Choose a Size Recommendation Solution.
Segment Your Results
A sizing change rarely affects all products equally. Segment results by category and even by SKU — a recommendation might dramatically help fit-complex categories like dresses and denim while doing little for basics. Segmentation tells you not just whether a change worked but where, so you can deploy it surgically to the products that benefit most, like the high-return categories in How to Reduce Returns for a Denim Brand.
Watch for Conversion-Return Trade-offs
The most important pattern to watch for is the trade-off: a variant that lifts conversion while raising returns, or vice versa. Sometimes resolving fit uncertainty slightly lowers conversion (a shopper learns the item won't fit and doesn't buy) while sharply cutting returns — a clear net win that a conversion-only view would reject. Always judge sizing tests on net revenue after returns, not conversion alone, so you reward the changes that actually improve the business.
Build a Testing Cadence
Treat sizing as an ongoing optimization surface, not a one-time fix. Maintain a backlog of fit hypotheses, run tests continuously, ship winners, and re-test as your catalog and audience evolve. Over time, a disciplined sizing-testing program compounds into a structurally lower return rate and a product page tuned for fit confidence, the goal of How to Improve Conversion on Fashion Product Pages.
Common Sizing-Test Pitfalls to Avoid
Sizing experiments fail in predictable ways. Knowing the pitfalls saves you from drawing wrong conclusions:
- Ending too early. The number one mistake — calling a test on conversion data before the return window closes, missing a returns spike that arrives weeks later.
- Judging on conversion alone. A variant that lifts add-to-cart but raises returns can look like a win and be a loss. Always include net revenue after returns.
- Underpowered tests. Fit effects can be subtler than UX effects, so a test without enough traffic produces noise mistaken for signal.
- Contaminated splits. If the same shopper sees both variants across devices or visits, your data is muddied.
- Over-segmenting after the fact. Slicing results until you find a "significant" subgroup invents wins that won't replicate.
Avoiding these comes down to discipline: pre-register your hypothesis and success criteria, run long enough and large enough, and judge on the full metric set. A rigorous sizing test that shows a smaller-but-real effect is worth far more than an exciting result you can't trust. Treat each test as a decision you'll bet budget on, and hold it to that standard before you scale the change across your catalog.
Where YourSizer Fits
YourSizer is built to be measured — deploy it as a test variant, track fit returns and net revenue against your control, and let the data prove the impact. Its analytics also surface the fit signals that generate your next round of test hypotheses. See how it works for brands, the analytics view, or pricing.
Sources
Frequently Asked Questions
How do I A/B test a sizing change?
Run a clean, randomized split with one variable at a time and pre-registered success criteria, then track not just conversion but fit-related return rate, bracketing, and net revenue after returns. Run the test long enough for the return window to close, since returns lag the sale.
Why can't I judge a sizing test on conversion alone?
Because a change can lift conversion while raising returns, making it a net loss — or slightly lower conversion while sharply cutting returns, making it a net win. Sizing affects both sides of the transaction, so you must judge tests on net revenue after returns, not conversion alone.
How long should a sizing A/B test run?
Long enough for the return window to close, which is days to weeks after purchase. Ending a sizing test as soon as conversion data looks good is a common mistake, because higher returns from a variant often only show up weeks later. Build the return lag into your test duration.