One script, no replatform
It runs beside what you already have. If you turn it off, your store is exactly as it was, because your own order is the fallback.
Decision API
The runtime reads the current visit and orders your catalogue against what it can infer. It runs alongside your existing store and search, holds back a share of traffic as a control, and falls back to your own order if anything is slow.
The visit starts
a page view, a referrer, nothing else
We infer
what this session appears to be deciding
We rank
your catalogue against that, with a reason per product
Or we stand aside
your own order, if anything is slow
A share never sees it
held back, so the result stays checkable
No cookie, no login, no identity. The session is the input and it ends when the session does.
Where you are
One shelf, before anything is ranked
216
Every other page here describes work that touches nothing on your site. This one puts a script tag on your storefront, which is a different kind of decision and deserves a different level of scrutiny.
It is also the wrong place to start. Ranking a catalogue against what a visitor seems to need cannot help if the catalogue does not record the facts that visitor is deciding on. Running the engine over an unrepaired shelf produces a confident ordering of products that are still indistinguishable.
So the audit comes first, the repair comes second, and this comes when you have a shelf worth ranking and a reason to believe ordering it live would change something.
A personalisation engine pointed at a catalogue that cannot answer questions is an expensive way to reorder the same confusion.
Why this one comes last
Same engine, same session, two catalogues. The difference is not the model.
If two products differ on a field nobody recorded, no ranker can separate them. It can only change which of the indistinguishable pair appears first, which is not a decision anyone was waiting for.
A better order over the same missing facts.
Once the deciding attributes are recorded, the session's implied requirement maps onto fields that exist, and the order changes for a reason you can read on every product.
This is why the audit is sold first, and this last.
What it looks like
Base, persona, price and brand, per product, so your merchandiser can see why anything moved.
Every ranked product carries this. Nothing is a black box to your merchandiser.
What happens
A single script. It does not replace your search, your storefront or your platform, and it does not need a login, a cookie or an identity to work.
What someone browses, searches and filters in this visit. Every ranked product carries a per-factor breakdown, so nothing it does is a black box to your merchandiser.
A held-back control group, stable per session. That is the only way any number anyone reports about it means anything, including ours.
What you get
It runs beside what you already have. If you turn it off, your store is exactly as it was, because your own order is the fallback.
Base, persona, affinity and price, per product. Your merchandiser can see why anything moved, and override it.
Not an option you enable when you want to measure. A share of traffic always sees your original order, so the comparison is always available.
If the engine is slow or unreachable, the storefront renders your own order. Degrading quietly is the default path, not the exception.
Before you commit
Questions
The runtime comes after a pilot, and the pilot comes after a shelf worth ranking. That begins with one category export.
Request a Decision AuditRuntime pricing is agreed per pilot, never before one.