New research guide: The Confidence Economy, how AI is changing fashion commerce.

Read it →

Holdout lift report

[ 07 ] · Holdout lift report

The number it actually moved.

Every vendor reports influenced revenue. We report the counterfactual.

Start 30-day free trial

Experience Elara

No credit card · Live in under an hour

HoldoutElara

AOV

Items / session

Conversion

Sample report · your numbers come from your own holdout

[ 01 ]

in one sentence

What is a holdout lift report?

A holdout lift report measures what Elara changed by keeping a random slice of your traffic — typically 5 to 20% — from ever seeing the stylist, then comparing both groups on the same days: conversion, average order value, items per session and revenue per visitor. The difference is incremental lift, not attributed revenue, and it ships in the dashboard with significance noted, so the number you take to your CFO is one that survives the question “compared to what?”.

Also called · incrementality test · control-group measurement · A/B holdout

In the glossary →

[ 02 ]

the gap

Why does it matter?

Every vendor reports “influenced revenue”. Nobody reports the counterfactual.

Attribution without a control is a story.

Your CFO asks one question: compared to what?

[ 03 ]

how it works

How does the holdout lift report work?

01
Choose the slice
5–20% holdout, set in the dashboard; random assignment per session.
02
Run it
Both groups measured on the same days and the same traffic; nothing else changes.
03
Read the delta
Conversion, AOV, items per session, revenue per visitor, with confidence intervals.
04
Decide
Keep the holdout running, shrink it, or switch it off once you are convinced; export CSV or a shareable PDF.

SAMPLE REPORT

Conversion   4×    vs holdout
AOV          +25%  vs holdout
Items        2.4×  vs holdout
 
SAMPLE REPORT · YOUR NUMBERS
COME FROM YOUR OWN HOLDOUT

Same days, same traffic. The delta is the lift.

[ 04 ]

the numbers

What does it do to the numbers?

+0%

Average order value

0×

Conversion, assisted sessions

5–20%

Holdout share

Industry standards for AI-assisted shopping sessions. Your holdout report shows your own.

Online fashion conversion rates typically sit in the 1–3% range, according to the industry benchmarks in our stats post — the baseline most fashion stores start from. Against a baseline that low, small changes look large and attributed figures look larger. A holdout keeps the comparison honest: the same days, the same traffic, one group without Elara — so whatever moves is the stylist’s doing, and nothing else.

Source · industry benchmarks, via our 2026 fashion ecommerce stats →

[ 05 ]

versus

How is it different from “influenced revenue” dashboards and last-click attribution?

ElaraInfluenced revenueLast-click attribution

Control group that never sees the tool

Measures conversion, AOV and items, not clicks

Significance shown

Same days, same traffic

Exportable for finance

Compare ·

Elara vs Rebuy →

Elara vs Nosto →

[ 06 ]

use cases

Which use cases does it power?

[ Use case ]

Complete the look

From any product page, Elara builds the outfit around the piece they are already looking at.

[ Use case ]

Reduce returns

Full-outfit try-on before checkout means fewer surprises after it.

[ Use case ]

Repeat purchases

Taste learned from skips and saves follows the shopper back.

[ 07 ]

who it is for

Who is it for?

Heads of Ecommerce who have to defend a line item, and finance teams who have stopped believing “influenced revenue”.

RANDOM PER SESSION

CSV · PDF

ON EVERY PLAN

[ 08 ]

pairs with

save · silk

[ 05 ]

Style Graph

A map of how every piece relates to every other, tuned by each shopper’s skips and saves.

[ 06 ]

Widget Studio

Accent, radius, opener, placement. It looks like your brand because you made it.

Hesitation detected

Still deciding? I can build a full outfit around this piece.

[ 01 ]

Proactive engagement

Elara notices a shopper hesitating and speaks first, with a reason to stay.

[ ? ]

questions

Questions a Head of Ecommerce asks first.

Anything else, book twenty minutes and ask the founder directly.

Ask a question

Yes, once you are convinced — and back on at any time. Most brands keep a small holdout running so every month’s number stays measured rather than assumed.

Sessions are assigned at random, both groups run on the same days and the same traffic, and the report shows confidence on every delta. It is the same method as an A/B test, with Elara as the only variable.

Conversion rate, average order value, items per session and revenue per visitor for the Elara group versus the holdout, plus engagement with the stylist for the Elara group.

10% is the default. Smaller stores can run 20% to reach significance sooner; larger stores can run 5% and lose almost nothing.

Yes. CSV for your own analysis and a shareable PDF for finance, both from the dashboard.

Yes. The holdout lift report is on Starter and above, because a stylist that cannot prove its number should not be paid for.

[ 09 ]

go deeper

Go deeper

How to measure whether your AI shopping assistant works

A practical guide to measuring an AI assistant honestly.

Incremental lift

Glossary: the difference a tool makes over a control group.

Fashion ecommerce metrics that actually matter in 2026

The metrics a fashion brand dashboard should lead with.

Give your store a stylist that actually sells.

Thirty days free. Live in under an hour. Measured against a holdout from day one.

Slide to start your trial

No credit card · Cancel any time