A measured before, dated
Battery of 12Twelve questions asked identically of every store, brand name shown and domain withheld, so the assistant has to recognise the store the way a shopper would.
Evidence
No customer results yet.
Burnish is new. Rather than invent a result, here is the bar one has to clear.
Nothing to list
Nothing is published here yet.
No store has been measured long enough, with its owner’s agreement to publish, for a case study to exist. An instrument that manufactures its own evidence has stopped being an instrument, and that is the only thing this product is.
The obvious shortcut
Why not anonymise a good result?
Because an unnamed, uncheckable result is indistinguishable from a fabricated one — and we would know which it was while you could not.
Instead
What you can check today.
The trail the product keeps while it works, which is what a case study would be built from.
- Every write keeps a before Indefinite
- One-click rollback, where a strategy exists 37 strategies
- Approval before anything consequential 29 of 41
- Full autopilot needs your explicit consent Per store
The requirement
Five things a case study has to contain.
Written now, while there is nothing on the page to bend them around. Anything short of all five is a testimonial with a chart on it.
The work that actually ran
41 fix typesWhich fixes: twelve run automatically under confidence floors, twenty-nine wait for approval, and six dependency checks run before any write.
A measured after, same instrument
21-day windowSame battery, same rolling window, sample size printed beside the result. Under ten samples the product draws no band at all.
Revenue described as what it is
CorrelationCorrelation-based, not a randomised controlled trial. Orders are canonical; the cohort view runs sixty days against a thirty-day baseline.
The merchant’s agreement
NamedNamed store, named numbers, published because the owner read it and said yes. No composite, no initials, no “a leading skincare brand”.
That asymmetry is the whole problem Burnish removes from AI answers about your store. Importing it into our own marketing would be a strange way to begin.
Per-Action lift shows nothing until at least seven days and three measurement points sit behind it. Where a change cannot be undone, the app says so instead of pretending otherwise.
The first one
The first case study will be a merchant’s.
If you run Burnish and the numbers move, we would like to publish it with your name on it and your figures intact.