Measurement
How we measure
Kinect reports revenue the way a CFO would accept it: an A/B split against your own baseline, attribution that requires real engagement, and cohort claims labeled as cohort claims. This page is the full methodology — including the claims we refuse to make.
Why publish this? Because this category has a lift-inflation problem. Tools advertise guaranteed conversion ranges before they've seen your store, and count every order that happened near their widget as revenue they drove. The brands we work with have usually been burned by exactly that. So we put the methodology in the open and invite you to hold us to it.
The method
- 01
Every store launches behind an A/B split
Visitors are bucketed deterministically at the edge: the control group never sees Kinect, the test group does. Conversion and AOV are compared between cohorts on the same store, same traffic, same period. This is the only structure that answers “what would have happened without Kinect?”
- 02
Significance before declarations
We only call a result a result when it's statistically meaningful. Early numbers wobble; we say so instead of screenshotting the best week. If Kinect doesn't win the split, we want to know as much as you do.
- 03
Attribution requires engagement
A Kinect-attributed order means the shopper sent a message or clicked a recommended product before buying. “The widget was on the page” doesn't count. This is the strictest common-sense definition we could hold ourselves to.
- 04
Cohort claims are labeled as cohort claims
When we say engaged shoppers convert at 6–21%, that's a description of a high-intent cohort, and we label it that way. It's evidence the conversations are working — it is not, and we never present it as, store-wide causal lift.
What the numbers we publish mean
Every number on this site comes with its definition attached. Here they all are in one place.
| The number | What it actually is |
|---|---|
| 3–6% more revenue | Measured across live Kinect stores against each store's own baseline, substantiated by Kinect-attributed order share. Not a guarantee, and not a cherry-picked best store. |
| 6–21% of engaged shoppers purchase | Engaged-cohort behavior on live stores — multiples of a typical low-single-digit site average. Engaged shoppers are higher-intent to begin with, which is exactly why we don't sell this as causal lift. |
| 24% more time on page | An increase in time on page when Kinect is present and selling, relative to comparable sessions without it. |
| Live in 1 day | From kickoff to live on your storefront, with the Kinect team doing the onboarding: integration setup, catalog ingestion, voice tuning, and a review with you before launch. |
What we refuse to claim
- No guaranteed lift percentages. If a vendor guarantees a conversion range before seeing your store, ask how it's measured — usually the guarantee is doing the marketing, and the attribution method is doing the inflating.
- No store-wide causal claims without a holdout. Correlation dressed up as causation is the category's favorite trick; we don't play it.
- No counting every order on a widget-loaded page as “AI-driven revenue.” Presence isn't influence.
- No named-customer numbers without the customer's sign-off.
Five questions to ask any AI vendor — including us
If you're evaluating AI shopping assistants, put these to every vendor in the room. We'll answer all five on the first call.
- How is the control group constructed, and can I see the bucketing logic?
- What exactly counts as an “AI-influenced” order?
- Is the headline number causal (from a holdout) or correlational (from a cohort)?
- At what point do you declare statistical significance, and what happens to the claim if it isn't reached?
- Can my analytics team re-derive your numbers from raw data?
Frequently asked questions
Does Kinect guarantee a conversion lift?
No, and that's deliberate. A vendor who guarantees a lift percentage before seeing your store is telling you about their marketing, not your revenue. Kinect launches every store behind an A/B split and reports results against your own baseline. Across live stores that has meant 3–6% more revenue, but your number is measured on your store, not promised in advance.
What counts as Kinect-attributed revenue?
An order counts toward Kinect only when the shopper actually engaged — sent a message or clicked a product Kinect recommended — before purchasing. Orders from shoppers who merely loaded a page where the widget was present are not counted. This is a stricter standard than most tools in the category use.
Why do you report engaged-shopper conversion instead of store-wide lift?
Because they answer different questions and conflating them inflates numbers. Shoppers who engage with Kinect convert at 6–21% — multiples of a typical site average — but engaged shoppers are higher-intent to begin with, so we present that as cohort behavior, not causal lift. Causal claims come from the A/B split, where the control group answers the counterfactual.
Can we audit the numbers ourselves?
Yes. The cohort definitions are in the open, your dashboard shows exactly which conversations and orders Kinect claims, and the underlying conversation and order data is yours. If your analytics team wants to re-derive the numbers, we'll walk them through the queries.
See the results this methodology produces: customer case studies or how Kinect compares.
Measure it on your store
Every Kinect engagement starts with an A/B split on your own traffic. Book a demo and we'll walk you through the methodology against your store.