Why 60% of tower truck rolls clear nothing — and what to do about it

Ask any TowerCo operations director where they’d find the most savings tomorrow, and a sizeable share will name the same thing: no-fault-found dispatches. The truck rolls, the tech drives, the site clears in twenty minutes with nothing replaced, the ticket closes with disposition “could not duplicate” or “transient resolved.” Operator surveys consistently put this above 60% of all tower truck rolls.

The cost is straightforward to count: a half-day of a senior tech, fuel, vehicle wear, the lost opportunity of the work that tech is not doing somewhere else. The cost that doesn’t show up on the spreadsheet is the alarm-fatigue cost: operators learn to dismiss certain alarm types because they so reliably resolve themselves, which means they also dismiss the rare instance when that alarm type really mattered.

Three structural failures account for most of the no-fault-found dispatch population. First: single-reading alarms. A sensor reports a value, the value crosses a threshold, the alarm fires. No second reading is required to confirm. If the sensor was momentarily faulted, if the reading was a digitization artifact, if the value oscillated around the threshold for thirty seconds and then settled, the truck still rolls. Most vendor RMS systems do not require sustained drift before firing.

A truck dispatched on a single-reading alarm is a 4-hour trip to look at a sensor that wasn’t quite stuck.

Second: no correlation with adjacent telemetry. A genset transfer alarm at 02:17 should be evaluated against: did the utility feed actually drop, did the ATS exercise, did the BTS aggregator stay up, did any adjacent site on the same utility loop report the same event. If the utility didn’t drop and the ATS didn’t exercise, the transfer alarm is a sensor noise event, not a real transfer. Most vendor RMS systems do not have access to the cross-source telemetry needed to make this call.

Third: vendor RMS limitations. Most vendor RMS systems are optimized to flag anything that might be a problem, on the principle that the cost of missing a real fault is higher than the cost of a no-fault dispatch. From a vendor liability perspective this makes sense. From a TowerCo operator perspective it produces an alarm queue dominated by transient noise.

The pattern that distinguishes a real fault from a transient is almost always sustained drift across multiple consecutive readings. A fuel pressure that crosses threshold for one sample and comes back is sensor noise. A fuel pressure that holds below threshold across 12 consecutive 30-minute samples is a real fault. The first should not roll a truck. The second should.

Correlation with adjacent telemetry adds a second filter. A genset transfer alarm with no corresponding utility loss is almost certainly a sensor noise event. A fuel pressure drop on a genset whose runtime since last refuel is below 30% of the tank is almost certainly not a real low-fuel state. The dispatches that should happen become easier to see when the noise is filtered.

There’s a compounding problem. When operators learn that 6 out of 10 dispatches will clear nothing, the L1 desk starts batching, deferring, and downgrading alarms in ways that are individually defensible but cumulatively dangerous. The rare alarm that was a real precursor to a major incident gets sat on for three hours because the operator has been burned six times this week by single-reading alarms that resolved themselves.

The fix is not “train operators harder to take every alarm seriously.” Operators are doing the right thing relative to the alarm quality they’re being handed. The fix is to raise alarm quality — fewer, more meaningful alarms — so that when the queue says something is wrong, it’s worth a truck.

A high-quality alarm queue has three properties: every alarm represents sustained drift, every alarm is cross-correlated with adjacent telemetry that either supports or refutes the fault hypothesis, and every alarm carries a recommended next action calibrated to the confidence level. A P3 (“fold into next maintenance loop, no truck needed”) is a different alarm class from a P1 (“dispatch within 2 hours, tenant-affecting if not addressed”). Properly tiered alarms also produce better evidence — when the L1 closes a P3 by deferring it, the deferral is logged. When the next maintenance run finds the predicted fault and corrects it, the closed loop is visible to the operations director.