Logistics & supply chain

Fleet dispatch optimisation with machine learning ETA prediction

We built a dispatch platform that plans routes against real constraints and predicts arrival times from historical telemetry, so dispatchers plan faster and customers get reliable delivery windows.

less time spent on daily route planning
70%less time spent on daily route planning
lower mean absolute ETA error versus the previous tool
31%lower mean absolute ETA error versus the previous tool
fewer kilometres driven per delivered parcel
9%fewer kilometres driven per delivered parcel
A delivery driver checking packages beside a delivery van

The challenge

Dispatchers built daily routes for around 300 vans using a desktop tool and their own local knowledge. Planning took most of the morning, routes ignored live traffic patterns, and customer delivery windows were wide because nobody trusted the estimated arrival times.

Late deliveries triggered support calls and failed first attempts, both of which cost money. The operator had two years of GPS telemetry and stop-level delivery records but no pipeline to turn that data into anything usable for planning.

Our approach

We first built the data foundation: a streaming pipeline that ingests vehicle telematics and driver app events, cleans GPS traces, and joins them to stop-level outcomes in a warehouse. That gave a reliable record of actual drive times and service times by area, time of day and customer type.

Route planning uses Google OR-Tools to solve a vehicle routing problem with time windows, vehicle capacities, driver shift limits and depot constraints. Instead of static road speeds, the solver uses travel and service time estimates from gradient-boosted models trained on the historical data. Dispatchers review the proposed plan on a map, adjust it with drag and drop, and publish routes to the driver app.

During the day, the ETA service re-predicts arrival times as vehicles progress and pushes updated windows to customers by SMS. Models are retrained weekly with drift monitoring, and prediction error is tracked by region on an operations dashboard so the team can see where estimates are weakest.

What we delivered

  • Streaming ingestion of vehicle telematics and driver app events
  • Constraint-based route planning with time windows, capacity and shift limits
  • ML-predicted travel and service times in place of static road speeds
  • Dispatcher map console with manual overrides and one-click publishing
  • Live ETA updates pushed to customers during the delivery day
  • Weekly model retraining with drift and accuracy monitoring by region

More case studies

Travel & hospitality

A complete taxi booking platform: fleet ERP, booking website and Flutter apps

We built a taxi operator's full digital platform: a fleet and operations ERP, a booking website, and Flutter rider and driver apps sharing one real-time backend.

ERP, website and mobile apps on one live backend
3-in-1ERP, website and mobile apps on one live backend
apps for iOS and Android from one Flutter codebase
2apps for iOS and Android from one Flutter codebase
driver tracking for riders and the dispatch team
Livedriver tracking for riders and the dispatch team

Travel & hospitality

A restaurant management platform: ERP, online ordering website and Flutter apps

We built a restaurant group's ERP for orders, kitchen, menu and inventory, plus an online ordering website and a Flutter customer app, all on one shared system.

ERP, website and mobile app sharing one menu
3-in-1ERP, website and mobile app sharing one menu
commission on direct online orders
0%commission on direct online orders
sales and stock view across every outlet
Livesales and stock view across every outlet

Healthcare & life sciences

A remote patient monitoring platform that care teams actually use

We rebuilt a telehealth provider's video visit and remote monitoring stack so device readings, alerts and clinician workflows live in one HIPAA-aligned platform.

faster median time from out-of-range reading to clinician review
64%faster median time from out-of-range reading to clinician review
systems clinicians use to review a monitored patient
3→1systems clinicians use to review a monitored patient
to launch a new chronic care programme, down from months
11 wksto launch a new chronic care programme, down from months

Facing a similar challenge?

Tell us where you are and what you need. A senior engineer replies within 24 hours with next steps, not a sales script.