Cold chains fail quietly. A freezer drifts two degrees on a Sunday night; nobody notices until Monday's write-off. The loss is real but invisible, which is exactly why it persists.
The problem
In hyperlocal fulfillment, perishables move through many small nodes, each a potential point of temperature failure. Losses showed up only at reconciliation — too late to save the product and too diffuse to assign a cause.
The approach
The fix was to make the invisible visible in real time: instrument the cold points with IoT sensors, stream readings centrally, and alert on excursions while there's still time to intervene. The design bias was toward actionable signal over dashboards nobody watches — an alert goes to the person who can open the door or move the stock, not to a report.
- Temperature sensors at the nodes that matter, not everywhere.
- Central streaming with excursion thresholds tuned to product tolerance.
- Alerts routed to the nearest person who can act, with escalation.
Outcome
Catching excursions early converted a recurring, unattributable write-off into a managed exception — over $1.5M+ in losses avoided, and a repeatable pattern for any temperature-sensitive node in the network.
Why it connects
This is the "sensors" node of the career arc — the same instinct as the allocation engine (observe state, act before failure), applied to a physical rather than a coordination problem.