Numbers you can defend
Every widget vendor claims lift. Almost none run the experiment that would prove it. Yokaify ships with a randomized holdout group built in — so the lift on your Results page is causal, not correlational.
assisted
gets the mascot
holdout
never sees it
See it working
A built-in experiment, always running
A randomized slice of your visitors never receives assists. Comparing their conversions against everyone else is what turns “we influenced this revenue” into “we created this revenue” — the difference between marketing and measurement.
Every number can be audited
Every assist is logged with its trigger, page, action, and outcome. Attribution is tiered — direct, assisted, influenced — instead of one inflated number, so you can defend the dashboard in front of your CFO.
Did the mascot cause the lift?
A few visitors never see the mascot — your silent control group.
Lift, on the record
The Results page shows assisted and holdout cohorts side by side, so the number your dashboard reports is the number an experiment produced.
What you get
Built-in holdout
A share of visitors never receives assists. The conversion difference between groups is the assist effect — the same methodology as a proper A/B test, running permanently.
Tiered attribution
Not every touched conversion is a caused conversion. Revenue attribution is tiered by strength of evidence, so you always know how confident each dollar is.
Results, not activity
The Results page leads with lift, conversions, and attributed revenue — not “messages sent” and other activity theater.
Visitor journeys
Per-visitor timelines show pages, assists, conversations, and outcomes — the qualitative story behind the quantitative lift.
Offer economics
Every offer tracks shown → redeemed → revenue, so discounting proves it pays for its own margin.
Your data, exportable
conversion.recorded and other events stream to your systems via signed webhooks, so your BI can mirror every number we show you.
How it works
- 1
Visitors are randomized
Each visitor is assigned to the assist group or the holdout on arrival — stable across their sessions.
- 2
Both groups are measured
Conversions and revenue are tracked identically for both groups; only the holdout never receives proactive assists.
- 3
Lift is computed honestly
The gap between groups — with the volume behind it — is your causal lift. Small samples are shown as “not enough data yet,” never dressed up.
- 4
Revenue is attributed by tier
Conversions link back to assists and conversations with graded confidence, so attributed revenue separates “caused” from “was present.”
The holdout costs us money — those visitors get no assists, and some of them leave. We run it anyway, because a lift number without a control group is marketing, not measurement.
Common questions
Why should I trust your lift number?
Because it’s built to be distrusted: a permanent randomized control group, tiered attribution confidence, and honest “not enough data” states. You can also stream raw events to your own warehouse and check our math.
What is tiered attribution exactly?
Conversions are linked to mascot activity with different strengths of evidence — from directly assisted purchases down to weaker associations — and revenue is reported by tier rather than lumped into one inflated number.
Can I turn the holdout off?
The holdout is what makes your numbers causal — it’s central to the product’s honesty. Removing it would make the Results page just another activity dashboard.
Can I get the raw data?
Webhooks stream events (including conversion.recorded) in real time today; CSV exports for conversations and leads are coming in Q3 2026.
See it on your own store
One script tag, live in minutes. Measured with a real holdout, so you'll know if it's working — not hope.