For ecommerce

Turn product questions into orders — and prove it.

Avarto answers from your catalogue, runs a proactive side-by-side comparison when a shopper is deciding, and rescues carts. Every nudge is an experiment with a holdout, so you see the revenue it actually earned — not a bare bar chart.

What the assistant does on a store

Answer product questions from your catalogue

Sizing, materials, stock, delivery — grounded in your own product data, with citations, instead of guessing.

Proactively compare products

When a shopper is weighing two options, the assistant offers a side-by-side comparison rather than waiting to be asked.

Rescue carts and nudge at the right moment

Offer a comparison, a size guide, or a callback — then measure whether the nudge actually earned revenue.

See which nudges moved revenue

Each experiment runs against a randomized holdout. The bar is the 95% confidence interval — never a bare point estimate. Green helped, red hurt, slate isn’t yet significant, and a sample too small to trust is labelled “not enough data” instead of a confident wrong number.

Conversion lift vs holdout, by experiment

Assertive "buy now" toneAssertive "buy now" toneFree-shipping threshold nu...Free-shipping threshold nudgeProactive product comparisonProactive product comparisonSize-guide prompt (mobile)Size-guide prompt (mobile)Cart-recovery offer (late l...Cart-recovery offer (late launch)+30%+30%+20%+20%+10%+10%0%0%−10%−10%−20%−20%Conversion lift vs holdoutholdout (0)−8% ✓+5% ns+10% ✓+14% ✓not enough data

The bar is the 95% confidence interval; the dot is the point estimate; the line at zero is the holdout baseline. A “✓” marks a statistically significant result.

Illustrative — based on the demo store, not a real customer result.

When the uplift became real

Cumulative incremental revenue measured against the holdout, day by day, with 95% confidence intervals. It’s honest about uncertainty: noisy and crossing zero early, then it clears significance once the winning variants separate from the holdout.

Cumulative incremental revenue vs holdout

$150.0k$150.0k$100.0k$100.0k$50.0k$50.0k$0$0-$50.0k-$50.0k-$100.0k-$100.0kCumulative incremental revenue vs holdoutDay 1Day 1Day 5Day 5Day 9Day 9Day 13Day 13Day 17Day 17Day 21Day 21Day 25Day 25holdout baseline ($0)significance reached
Cumulative uplift
95% CI band

The shaded band is the 95% confidence interval; the line is the point estimate. Below the baseline the variant is losing money. The marker shows the day the lower bound first cleared zero.

Illustrative — based on the demo store, not a real customer result.

The campaign loop: a ratchet, not a one-off test

Optimisation isn’t a single A/B test — it’s a programme that compounds. Inside each shaded experiment band the line wobbles as variants win and lose; when the experiment ends you apply the winners as normal site behaviour and the line settles and holds. Run a second experiment and the cycle repeats — even dipping while a losing idea is live — but its learnings push the metric to a new high, above where the first round left it. The holdout line shows what conversions would have done with no experiments at all; the gap is the illustrative lift Avarto adds.

Conversions vs holdout, across two campaigns

130130125125120120115115110110105105100100959590908585Conversion index (holdout = 100)W1W1W4W4W7W7W10W10W13W13W16W16W19W19W22W22W25W25W28W28Hold-outExperiment 1Learnings appliedExperiment 2Learnings applied
With Avarto
Holdout (no experiments)

The shaded area is conversions with Avarto; the dashed line is the holdout (no experiments), held flat at the baseline index of 100. The dashed-edge bands mark each live experiment period (“begin ┄ end”) — where the line wobbles; between them the winning learnings are applied and the line holds, then climbs.

Illustrative — based on the demo store, not a real customer result.

  1. 1

    Hold-out

    A randomized slice of visitors never sees an experiment. That flat line is your true baseline — the counterfactual every later gain is measured against.

  2. 2

    Experiment period

    Inside the shaded band the line wobbles week to week — some variants win, some lose. That noise is the programme working, not failing: you're learning what actually moves the metric.

  3. 3

    Learnings applied

    When the experiment ends, the proven winners become permanent site behaviour. The wobble stops and the line settles — and the lift holds instead of snapping back to baseline.

  4. 4

    Repeat — and climb

    The next experiment period wobbles again (sometimes it even dips while a losing idea is live), but once its learnings are applied the line steps up to a new high — above where the last round left it.

A holdout that proves causation

A holdout group never sees the assistant, so you measure causal sales uplift — the revenue Avarto actually added, not the sales it happened to sit near.

Avatar-optional, one script tag

Run text-only or add an animated avatar with voice. Install directly or via Google Tag Manager with no other site code change.