Cold case · demo background
Production surface — everything below is what the live dashboard would show. This fold is for the presenter.
Talk track.
They said this case is 33% of profit — empty facings are the lever, and
we measure empty time, not dollars. Cold vault leads the pitch: it is
their biggest dollar story, and what we do here we can do anywhere in
the store. Joey prefers this planogram view — out of stock by product
and brand. The Locations toggle is what the pilot ships on day one; the
planogram view needs their planogram file, and it assumes the planogram
is mapped right: we detect that an expected facing is empty, we do not
check planogram compliance. Maverik counts empty holes and targets 10 or
fewer per store including beer; a store-level hole count is not on this
page yet. The distribution panel shows
how this store’s configured dwell bands behave. Walked away is Joey’s
phrase for the 5–60s band. The same people sit in the
Presence heat.
Walk rule.
Walk now = confirmed empty during selling hours (headline + queue).
Soon = running low during selling hours (warn queue, not in the headline).
No walk = scheduled cleanout, overnight empty, blocked, or no claim.
Claims.
Numbers are illustrative, on the shelf-facing OOS prototype. Facing
fullness + consecutive-sample gate — not a can count, not a census.
Blocked means we could not see; it never counts as empty or stocked.
Walked away and Converted are configured dwell proxies, not proof that a
person did or did not take a product. This illustrative view uses a 2s
detection floor, 5s presence floor, 60s conversion threshold, and 120s
long-dwell threshold: under 2s is Flicker and ignored, 2–5s is Passing,
5–60s is Walked away, 60s+ is Converted, and 120s+ is the Long subset.
Staff are not filtered. Cold-vault
pickups may be faster, so Maverik’s thresholds need labeled pilot data.
Product-in-hand VLM verification is not built. “Visited while a
facing was empty” is walked-away visits (5–60s) whose start overlaps a
confirmed-empty span. It is a subset of Walked away, not all tracked
visits. It says a facing was empty during the visit. It does not say
the person wanted that product; the customer reads the impact. The
window is midnight to now; it is not limited to selling hours.
Median time to restock counts confirmed-empty → refilled spans. 33% is
their profit mix, not a RavenEye dollar. Est. lost sales is
assumption-gated: empty minutes in selling hours (ours) × units per
hour × price for that facing (theirs, from the planogram file), every
input visible, labeled estimated. The tile stays blank until their
velocity and price feed exists: a sample dollar on the page is the
number that gets screenshotted. Never visits × price: a visit is not a
unit. Planogram view only; the Locations view has no SKU, so no dollar.
Not built; needs their planogram file.
Build readiness.
Artifact is a POC. Facing state is a prototype. Planogram product
identity and the est. lost sales tile are not built — both need their
planogram file. Empty minutes,
dwell bands, visited-while-empty, restock latency, and the hourly
overlay are derivable from production visits and confirmed-empty spans.
4 facings empty.
A shopper in front of Water is unknown, not out of stock. Prime has no full-seed yet, so it reports coverage and stays out of the queue.
Visited while a facing was empty counts walked-away visits whose start overlaps a confirmed-empty span. It says a facing was empty at the time, not that the person wanted it.