From detection to execution, and to cash

Detection was not the problem. An in-aisle robot already walked the floor every night and identified two things reliably: product missing from the shelf, and price wrongly labelled. What arrived the next morning was an undifferentiated list. 100% of alerts repeated every night for as long as the underlying problem stayed there, so the list never shortened and reading it stopped being worth anyone's time. None of them carried the impact in money, so a missing high-margin line and a mislabelled slow mover weighed exactly the same. And 0% of the actions taken against them was recorded, so nobody could measure what had been done, defend a decision, or show a result to a manager or an auditor. The chain was paying for detection it had already installed and capturing almost none of the margin that detection exposed. Detection remains the robotics vendor's; what was built here is everything that happens after the alert.

• Every alert converted into a task carrying an owner, a rank and a value in dollars — which is what stops it being an infinite list and starts it being work.
• A value in dollars attached to each alert from the store's own prices and margins, which is what makes ranking possible at all — and what is absent in the stores where those files have never been loaded.
• A replenisher queue on tablet, ordered from the task that recovers most margin downwards, closed in a single tap.
• A record of who acted, when, and what was done — the artefact that can be shown to a manager, a client or an auditor, which did not exist before.
• Twelve rounds of improvement run with real users, with the unfavourable numbers kept in rather than filtered out.

317 of 318

assigned tasks resolved

3.1x

value of a selected task vs. a random alert

9.4%

of exposed value captured from 3% of alerts

• 3% of alerts captured 9.4% of the value: the average executed task was 3.1x more valuable than a random alert. Measured on one store over 18 days of operation.
• 99.7% execution — 317 of 318 assigned tasks resolved.
• 37% utilisation: 10 of 27 available days worked. The ceiling in the pilot was set by the routine, not by the product.
• Per-store monthly ladder: US$ 1,967 today (10 of 27 days) → US$ 8,976 with the same team every day (+294%) → US$ 12,987 with one more person on half shift (+560%) → US$ 21,566 with a dedicated shift (+996%). Each rung is an operating decision, not a development.
• First-year cash of US$ 86,872 per store, accumulated as a ramp and not as a flat run rate: US$ 3,662/month at ≈M2, US$ 6,074/month at ≈M6, US$ 11,379/month at ≈M11–M12 — a run rate of US$ 136,548/year/store by month 12.
• US$ 995,940/year/store of exposed margin sitting in alerts nobody was prioritising, of which 2.4% was being captured.
• Across the 42-store chain: US$ 41.8M/year of exposed margin — the ceiling, not the promise — US$ 3.6M of year-one cash and a US$ 5.7M year-two run rate. Published as a range of US$ 2.6M to US$ 3.6M, because the pilot's second store showed 56% less exposed margin than the first.