For education & courses

More enrollments, and the proof it was Avarto.

Pre-purchase questions kill course sales. Avarto answers syllabus, format, and refund questions, then nudges checkout.

What the assistant does on a education & courses site

Answers syllabus, format & prerequisite questions

Cited refund & access policies

Course-page → checkout nudge

See which experiments moved enrollments

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.

Enrollments lift vs holdout, by experiment

Hard-sell enrollment pushHard-sell enrollment pushCourse → checkout nudgeCourse → checkout nudgeCited refund & access policyCited refund & access policySyllabus & prerequisite answ...Syllabus & prerequisite answersCohort spots-left, when rea...Cohort spots-left, when real (late launch)+30%+30%+20%+20%+10%+10%0%0%−10%−10%−20%−20%Conversion lift vs holdoutholdout (0)−5% ✓+5% ns+9% ✓+16% ✓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.

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 enrollments would have done with no experiments at all; the gap is the illustrative lift Avarto adds.

Enrollments vs holdout, across two campaigns

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

The shaded area is enrollments 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.

What moved

+16%
enrollments
−18%
unanswered questions
+12%
page → checkout

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

Grounded answers, measured causally

Answers are grounded in your own content, and a holdout group never sees the assistant — so you measure the enrollments it actually added, not the ones 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.