A single walk-in failure at one of 200 stores is an $8,000 to $15,000 event. Lost product, an emergency truck roll, a same-day service charge two or three times the planned rate, and a food-safety incident waiting to be inspected. That's one site. Multiply by the failure rate across a fleet and refrigeration becomes the line item that quietly defines your year.
Reactive maintenance catches the failure after the loss. Preventive maintenance fixes equipment on a calendar whether it needs it or not. Predictive maintenance for multi-site refrigeration is different. It uses sensor data from the equipment that's actually drifting, not the schedule, and it flags the unit headed for failure before product is at risk.
This guide walks through how that works, what the IoT stack looks like, what a usable alerting layer does at scale, and what a 90-day pilot should produce on a 200-site fleet.
Three modes operators run today.
Reactive. A walk-in alarms. The team finds out from a store manager, a customer complaint, or a thawed product. The truck rolls. Product gets transferred or thrown out. Cost shows up next month.
Preventive. Every cooler gets a quarterly or semiannual PM visit. Some of the visits catch real issues. Many don't. The fleet still has failures between visits because failures don't follow the calendar.
Predictive. Sensor data from each cooler streams to a central platform. Anomaly detection flags units whose performance is drifting outside spec. Temperature climbing. Compressor amp draw rising. Defrost cycles running long. The flag goes out before the unit fails. The fix is scheduled, not emergency. Product never gets close to the line.
The savings from preventive vs. reactive are real but capped. The savings from predictive compound across the fleet because the platform catches drift across thousands of units a human can't watch.
Compressor failures rarely arrive without warning. The warning shows up as drift in four signals. Each is small in isolation. Together they're a forecast.
Compressor amp draw. A healthy compressor pulls a consistent current. As bearings wear, refrigerant charge drifts, or condenser coils foul, the unit pulls more amps to do the same work. A 15 percent rise above the unit's baseline, sustained over a week, is a service ticket waiting to happen.
Evaporator temperature climb. If the evaporator setpoint is 35 degrees and the actual reading creeps from 35 to 36 to 38 over ten days, something is changing. Could be a refrigerant charge issue, a failing fan motor, a fouled coil. The trend matters more than any single reading.
Door-open frequency. A walk-in door propped open for an hour explains a lot of bad data. A 200-site fleet has hundreds of these events per week. Knowing which sites have them lets you separate operator behavior from equipment failure.
Defrost cycle anomalies. Defrost cycles that run too long, too often, or at the wrong time signal control board issues, sensor failures, or refrigerant problems. Catching these early avoids the harder failures they precede.
A platform that watches all four and correlates them is doing predictive maintenance. A platform that watches one and triggers on a hard threshold is doing alarms.
The hardware layer is simpler than vendors make it sound.
Sensors. Wireless temperature and humidity sensors inside the cooler. Compressor monitors on the electrical panel. Door sensors on each walk-in. NIST-traceable accuracy is table stakes. Anything else is a recurring calibration problem.
Gateway. A single gateway per site collects data from sensors and pushes it upstream. A real one runs on cellular or LoRaWAN backhaul, not store WiFi. Store WiFi goes down, gets rebooted, gets put on a guest network. The data layer can't depend on it.
Cloud platform. Where the data lives, where the analytics run, and where the dashboard surfaces what matters. The platform layer is what separates a sensor company from an intelligent monitoring platform.
Workflow. Alerts route to the right person at the right time, with enough context that the recipient can act. SMS, email, push, escalation paths if the first responder doesn't acknowledge.
The whole stack should install in under a day per site for a typical multi-unit retailer. If a vendor's deployment plan starts with "we'll need a controls engineer for two weeks," that's the wrong vendor for a 200-site rollout.
At fleet scale, a dashboard either earns its place in the daily standup or it doesn't get opened. Five things separate the two.
Site-level rollup. A facilities director needs the fleet view in one screen. Sites trending off baseline, sites with active alerts, sites stable. Drill-down comes after.
Anomaly flagging that doesn't cry wolf. If half the alerts are false positives, the team starts ignoring all of them. Good alerting suppresses door-open events as anomalies, distinguishes daily defrost from drift, and doesn't fire on every two-degree fluctuation.
Ticketing integration. When the platform flags an issue worth fixing, it should open a work order automatically and route it to the right vendor. A flag that requires a human to translate it into a ticket is half a tool.
Drill-down to single-unit data. When the regional manager wants to see the actual amp draw curve on cooler 4 at site 87, that data needs to be one click away. Not a CSV export, not an API call.
Audit trail. Who changed which setpoint when. Who acknowledged which alert. Who closed the work order. Required for HACCP, useful for performance management, essential for compliance reviews.
Multi-site refrigeration in QSR and franchise environments comes with three complications a single-brand retailer doesn't have.
Mixed equipment vendors. Acquisitions, rollouts over time, and franchisee-by-franchisee purchasing produce a fleet with three or four cooler vendors, two compressor brands, and a half-dozen control board generations. The platform has to ingest data from all of them without a custom integration per equipment type.
Franchisee variability. Service-call habits, vendor relationships, and willingness to act on alerts differ across franchisees. The platform's alerting and reporting need to give corporate visibility without forcing a single workflow on every operator.
Cost allocation. Who pays for the platform. Who pays for the service calls. Who keeps the energy savings. The answer varies by franchise agreement, and the platform's billing and reporting should accommodate that, not force a renegotiation.
The platforms that work in QSR are flexible enough to absorb that variability. The ones that don't end up running in pockets of the fleet, not the whole fleet.
Reasonable benchmarks for a 200-site pilot, based on real deployments.
These are pilot outcomes, not vendor deck numbers. If a vendor's projected savings are dramatically higher, ask for the math.
Predictive maintenance for multi-site refrigeration is a software problem, not a hardware swap. The cooler still cools. The compressor still runs. What changes is whether someone catches the drift before the failure.
GlacierGrid runs a 90-day free pilot for qualified multi-unit operators with 50 or more sites. Real equipment, real data, real anomalies caught. No long-term contract. Learn more about GlacierGrid Cooling Intelligence or start a free pilot.