How We Cut Dispatch Compute Time from 90 Seconds to 5
A look at the solver rewrite that made same-day re-dispatch viable for our largest customers.
Most fleet apps drain a driver's phone in three hours. We rebuilt our location pipeline so drivers finish a 10-hour shift with battery to spare.
Driver complaints about battery life are almost universal in this industry. The naive approach — sample GPS every second and post to the server — burns 25% battery per hour on an iPhone 13. By hour five the driver is hunting a charger instead of finishing stops.
Three changes that actually moved the needle
- Adaptive sample rate: 5s when moving, 60s when stationary, paused when geofenced at a stop
- Batched uploads: 6 samples per request, gzipped, retried with exponential backoff
- Background mode discipline: significant-location-change instead of continuous tracking when the app is backgrounded
Net effect: average driver battery at end-of-shift went from 12% to 64%. Customer-facing ETA accuracy stayed at 96%. Sometimes the right number of samples is smaller than you think.