What a Size Guide Actually Does to Conversion — and to Returns

A size guide test usually wins on conversion rate within a week. The returns effect lands three to four weeks later, in a system that does not know the test existed. If you only read the first number, you can ship a variant that raises orders and lowers profit.

The Test That Wins and Loses at the Same Time

Two effects, two timelines, one decision made before the second one arrives.

By the time the returns data exists, the test has been archived, the variant is live for all traffic, and nobody is looking for a cohort that no longer exists as a cohort.

This is not a hypothetical failure mode. It is the default outcome of running a size-guide test on a standard testing setup.

What a Size Guide Changes in the Buyer's Head

The intervention works on hesitation at the add-to-cart moment, not after purchase.

A shopper who cannot determine their size has three options: leave, guess, or hedge. A size guide converts some of the leavers into buyers — which is the conversion lift — and converts some of the guessers into confident buyers, which is the returns benefit.

It can also convert leavers into hedgers, and that is where the trap sits. A shopper who orders two sizes intending to return one has converted. They have also booked a return.

AdScale's analysis across more than 7,000 apparel stores found that single-item apparel orders come back refunded 6.6% of the time, rising to 10.2% on orders containing four or more items — a pattern consistent with shoppers ordering options rather than making a single confident choice.

So the same intervention can move conversion up and net contribution down, depending on which of those three behaviours it shifts most.

For the content side of what a good size guide should contain, our post on pre-purchase education and conversion rate covers it. This post is about the test.

Why You Cannot Read This Test on Conversion Rate

Three structural reasons, and the third is the one that breaks most setups.

The effect is delayed. Conversion resolves in days. Returns resolve over weeks, and the window varies by your return policy. A fourteen-day test cannot observe a thirty-day return cycle.

The data lives somewhere else. Conversion sits in your testing platform and GA4. Returns sit in Shopify, your returns app, or your 3PL. Different systems, different reporting cadences, usually different owners.

Nothing carries the variant ID across the boundary. This is the specific break. Your testing tool assigns a visitor to variant A or B and records that assignment in its own cookie and its own database. The order lands in Shopify with no variant field. The return is processed against the order, still with no variant field.

So when the returns data finally exists, there is no way to ask which variant did this returning customer see. The join was never built, and it cannot be reconstructed afterwards.

That is why the conversion number gets read alone — not because anyone decided to ignore returns, but because the architecture made the second half unanswerable.

Designing the Test So Both Halves Are Readable

Four changes, all made before launch.

1. Write the variant assignment into the order. At checkout, pass the variant ID into a Shopify order attribute, note field, or line-item property. This is the critical step. Without a variant ID on the order record, nothing downstream can be attributed.

2. Define the measurement window before you start. Test runtime plus your full return window plus processing time. If your policy is 15 days and processing takes a week, your returns window closes roughly 22 days after the last order in the test. Commit to that date before launch.

3. Hold the test open, or hold the decision. Two options. Either run the test long enough that the earliest cohort has passed its return window, or call the conversion result and explicitly defer the shipping decision until returns data arrives. The second is usually more practical. What does not work is shipping on the conversion result and intending to check returns later.

4. Tag the returns, not just count them. A return with no reason code tells you the order came back. A return tagged size or fit tells you whether your size guide worked. Most returns apps support reason codes; most brands do not enforce them.

The Win Condition: Net Contribution Per Visitor

Not conversion rate. Not average order value. Not even revenue per visitor, because revenue per visitor does not deduct what the returns cost.

Net contribution per visitor =

The second line is the one that usually gets left out. A return costs you the margin you booked plus the reverse logistics — outbound shipping already spent, return shipping, inspection, restocking, and any unit that cannot be resold at full price.

Run both variants through this and the ranking can invert.

A worked illustration

Figures below are illustrative, chosen to show the mechanism rather than drawn from a specific engagement.

Per 1,000 visitors, with a 40% gross margin:

The variant wins on conversion by 15% and still nets more delivered orders — but only just, and that is before the cost of processing 2.5 additional returns. Add reverse logistics and the gap narrows further or reverses entirely depending on your return cost.

The point is not that the variant loses. It is that a 15% conversion lift translated into a 3% gain in net orders, and no reading of the conversion number alone would have told you that.

This is the same reasoning we apply to RTO as a measurement problem and to prepaid conversion — three different levers, one shared failure mode: optimising a placed order rather than a kept one.

When the Size Guide Is Not the Fix

The honest section, because sometimes the intervention is wrong.

The chart may be inaccurate. A better-presented wrong chart raises returns. If your size chart was built from a spec sheet rather than measured garments, improving its visibility propagates the error faster. Measure a sample of actual stock before testing presentation.

It may be a photography problem. Fit perception comes substantially from imagery — model height and size, how the garment drapes, whether it is pinned at the back. A size chart cannot correct an expectation set by a photograph.

It may be a fabric description problem. Stretch, weight and structure determine how a stated measurement feels. "38 inch chest" means different things in rigid cotton and in jersey.

The sizing may genuinely be inconsistent. If garments vary across production batches, no chart is accurate for all of them. That is a sourcing and QC problem presenting as a CRO problem.

Baymard Institute's testing found that 83% of desktop and 87% of mobile apparel sites did not provide sufficient sizing information — so there is usually room to improve. But "insufficient" and "wrong" need different responses, and only one of them is a size guide.

What We See in Practice

Observations from running these tests, offered as observations rather than measured findings:

The conversion effect is usually real and usually smaller than the first read suggests. Early-stopped tests overstate uplift, and size-guide tests are particularly prone to early stopping because the effect appears quickly.

Mobile and desktop behave differently. A size guide hidden behind a modal link performs differently on a phone than on a desktop, and the gap is frequently larger than the overall effect being measured. Split by device before concluding anything.

Reason codes are almost never in place. The most common blocker to running this test properly is not the variant ID join — it is that returns are recorded without a reason, so even a well-instrumented test cannot distinguish a fit return from a changed-mind return.

Vendor claims should be read with their interest in view. Sizing technology vendors report 18–30% reductions in fit-related returns, with higher figures for single-brand recommendation engines. Those are vendor-published figures from companies selling sizing tools. They may well be accurate; they are not independent.

Worth noting that one widely cited statistic — around 70% of apparel returns being size or fit related — comes from a Coresight Research survey that was sponsored by a sizing technology company. Independent estimates put the fit share lower, in the range of roughly 45% to 70% depending on the study and method.

The Short Version

Instrument the variant ID onto the order before you launch. Define a measurement window that covers your return cycle. Judge the test on net contribution per visitor, not conversion rate. And check your chart is accurate before you improve how it is presented.

A size-guide test is one of the few interventions where the conversion winner and the commercial winner can be different variants. Design for both or you will not know which one you shipped.

If your product pages and checkout have never been examined with this lens, our post on signs your PDP or checkout needs a CRO audit covers what else tends to surface. And for the broader version of this argument — distinguishing what a test proved from what it merely observed — see how to read a CRO agency's monthly report.

Running fit-sensitive products and want tests designed to measure both halves? Talk to FunnelFreaks about a CRO Programme.