Conversion Engine Metrics for Shopify

The board slide says 3.8%. The sales deck says 35%. The budget request says 10x. By the time the numbers reach the weekly trading meeting, all three have somehow become “conversion rate.”

They are not interchangeable. The 3.8% figure is an observed conversion rate among Antla try-on users. The 35% figure is an average relative conversion lift associated with try-on use. The 10x figure is return on investment for the merchant. Each has a different numerator, denominator, and business question.

A finance-aware operator labels those boundaries before presenting the result. The broader conversion engine definition explains the system being measured. This article deals with the less glamorous but essential job of keeping its numbers honest.

Conversion rate, lift, ROI, engagement, and return reduction belong in separate columns before they belong in a presentation. Editorial image in Classic Antla disposable-camera style.

Put the question beside the metric

A metric is useful only when it answers the decision in front of the team. Merchandising may need evidence that a try-on experience resolves uncertainty around drape or silhouette. Finance needs incremental contribution after costs. Retention needs revenue recovered from shoppers who first left without buying.

Use this table before choosing the largest available number:

The wording in the final column is not decorative caution. It preserves the meaning of the evidence. Removing “try-on users,” “average,” “at least,” or “up to” changes the claim.

Shopify’s conversion-rate guide makes the same underlying point about benchmarks: industry, traffic source, price, device mix, geography, and purchase cycle all affect conversion. A Shopify fashion team should compare like with like before treating any external benchmark as a verdict.

Quote 3.8% when the audience wants an absolute rate

The 3.8% number answers a cohort question: among people who completed a try-on, what percentage converted under the measurement definition?

Write the formula in the report:

try-on-user conversion rate =
  qualifying try-on users who purchased
  / qualifying users who completed a try-on

Then add the window, order rule, and cohort filters. A same-session purchase rate will differ from a seven-day rate. Counting orders for any product will differ from counting only the product tried. Counting generations can also inflate the denominator because one shopper may generate several previews.

Do not apply 3.8% to all store sessions. It is not a forecast that a store-wide conversion rate will become 3.8% after installation. It describes shoppers who used the experience across more than 500,000 Antla try-ons.

That distinction is especially important for larger merchants. The Shopify Plus conversion-engine guide shows how to evaluate a high-traffic cohort without claiming that 3.8% beats every Plus store. Channel mix and product economics still matter.

Quote 35% when the question is relative improvement

A lift compares two rates. If try-on users convert at 2.7% and an appropriate comparison group converts at 2.0%, the relative lift is 35%:

relative conversion lift =
  (2.7% - 2.0%) / 2.0%
  = 35%

That is a 0.7 percentage-point increase and a 35% relative increase. Both statements describe the same example, but they are not the same unit. A slide labeled “conversion increased 35 percentage points” would turn 2.0% into 37.0%, which is a very different business and probably a shorter meeting.

Use the 35% average when a stakeholder asks whether try-on users outperform the defined comparison. Show the underlying rates beside it. Also state whether the comparison is observational or experimental.

Try-on users are self-selecting. They may begin with more interest than non-users, so an observed cohort lift does not prove the entire difference was caused by the experience. A randomized rollout or stable holdout gives a stronger causal estimate. If that is unavailable, segment by eligible PDP traffic, device, source, product category, and period, then describe the result as association rather than incrementality.

Keep the tool category clear as well. Conversion engines, CRO tools, and personalization affect different parts of the customer decision, so their scorecards should not be collapsed into one blended lift.

Quote 10x when finance asks what the investment returned

ROI belongs in a budget conversation, not in a conversion-rate chart. Its denominator is cost:

ROI multiple =
  attributable incremental value
  / total conversion-engine cost

For a conservative Shopify fashion model, incremental value should begin with contribution from additional orders, not gross revenue. Deduct discounts, cost of goods, payment costs, media or message costs, cancellations, and expected returns. Add return-related savings only when the store can measure them without counting the same benefit twice.

The cost side should include the app, implementation work, internal operating time, and any incremental channel spend. Antla’s merchant results support a minimum 10x ROI claim, but the store should retain the calculation behind its own number. “10x ROI” does not mean ten times the conversion rate, ten times revenue, or ten times more shoppers.

The Shopify ecommerce optimization guide treats conversion as a journey spanning discovery, product evaluation, checkout, and retention. That broader view is useful for finance because one tool may affect order volume, customer confidence, and return costs at different points. The model still needs one counted benefit for each outcome.

Keep engagement and returns in supporting roles

Engagement helps diagnose whether shoppers are using the virtual try-on on-site layer to evaluate a garment. Antla data shows try-on users spend about two to three times longer onsite. That can indicate useful consideration around cut, length, color, or drape, but longer sessions do not automatically create profit.

Report the engagement unit explicitly. “Time on site,” “engaged sessions,” and “pages per session” are different measures. Compare the same measure, on the same eligible products, across the same period.

Returns sit at the other end of the journey. Antla customers have seen reductions of up to 30% when try-on narrows the expectation gap. Quote that as a relative reduction, not as a promise to subtract 30 percentage points from every merchant’s return rate.

For example, moving from a 20% baseline return rate to 14% is a 30% relative reduction:

(20% - 14%) / 20% = 30%

Report return results by product category, return reason, and order cohort. A preview may help with appearance expectations around silhouette or styling. It cannot correct a mislabeled size chart, damaged stock, or late delivery.

Give recovered revenue its own ledger

Recovered revenue comes from eligible shoppers who completed a high-intent action, left without buying, received a follow-up treatment, and purchased within the recovery window. It belongs to the second loop of the engine. The Shopify virtual try-on conversion guide covers the on-site and recovery path from the try-on perspective.

Do not fold recovered orders into try-on-user conversion and then present the same orders again as campaign revenue. Create separate rows for:

1. In-session revenue: qualifying orders placed before the original session ended.
2. Post-session observed revenue: later orders from eligible non-buyers.
3. Platform-attributed recovered revenue: post-session orders credited under a channel’s attribution rules.
4. Incremental recovered revenue: treatment revenue above what a comparable holdout generated.
5. Recovered contribution: incremental recovered revenue after discounts, returns, channel costs, and cost of goods.

Klaviyo’s email and SMS benchmark report is useful for checking whether flow delivery, clicks, and conversion look unusual for the industry. It cannot establish incrementality for one store. High-intent shoppers sometimes return without a message, which is why the holdout belongs in the finance readout.

The Antla conversion engine connects the on-page try-on signal with the unresolved product decision that remains after a shopper leaves. Keep one stable shopper assignment across email, SMS, and ads so three channel reports do not claim the same recovered order.

Questions finance and ecommerce teams ask

When should I quote 3.8% versus 35% lift?

Quote 3.8% when answering the absolute cohort question: what share of Antla try-on users converted? Quote 35% when answering the comparison question: how much higher was conversion among try-on users than in the defined comparison cohort, on average? Present the underlying rates, time window, and cohort definition with either figure. Never substitute either number for the store-wide conversion rate.

What is 10x ROI measuring?

The 10x figure measures value returned relative to the cost of operating the conversion engine. A merchant-level model should use incremental contribution and defensible return savings in the numerator, then include software, implementation, operating, and incremental channel costs in the denominator. It is an investment multiple, not conversion lift or revenue growth.

How do I report recovered revenue separately?

Begin with shoppers who were eligible for recovery because they completed a high-intent action and left without purchasing. Separate in-session orders from later orders, deduplicate outcomes across channels, and show observed, attributed, and incremental recovered revenue in different columns. Use a stable holdout to estimate what would have happened without treatment, then deduct discounts, returns, and variable costs to report recovered contribution.

Send one number with one definition

Before the next trading or board meeting, add five fields beside every headline metric: cohort, numerator, denominator, window, and comparison method. If a number cannot survive those labels, it is not ready for the slide.

About the author: Aaron founded Antla and has seen one number get used in three slides to mean three different things. He wrote this so that happens less.

Choose one eligible product cohort, define the commercial outcome, and keep conversion, lift, ROI, engagement, returns, and recovery in separate rows. When you are ready to measure the full loop on your store, add Antla to Shopify.