Data Department built a repair shop planning system for an automotive service network. A machine learning model predicts how long each job will take, and an optimiser uses those predictions to assign jobs to technicians. It raised average gross shop revenue by about 15%. It combines prediction and optimisation in one planning decision.
What was the decision?
Each working day, a service workshop typically decides which technician takes which job, and in what order. The workshop has a fixed number of technicians and bays. The jobs vary: some take minutes, some take most of a day.
A plan built on poor duration estimates fails in two ways. Some technicians run out of work early while others fall behind. Jobs promised for the afternoon slip to the next day. Both leave paid technician hours unused, and unused hours are lost revenue.
So the question was not "how long does a job take on average?" It was "given today's jobs and today's technicians, which assignment gets the most work done?"
What did we build?
We built two parts that work together:
| Component | What it does | Why it matters |
|---|---|---|
| Job-duration prediction | Estimates how long each incoming job will take | The plan is only as good as its time estimates |
| Technician-assignment optimiser | Assigns jobs to technicians using the predicted durations | Turns better estimates into a better schedule |
| Output | An assignment plan for the workshop | The decision the shop actually makes |
Why both parts. A duration model on its own produces numbers on a screen. An optimiser on its own works from guesses. Linking them means better predictions change the plan directly.
In general terms, duration models in repair shops learn from the history of completed jobs: what kind of work was done and how long it actually took. Assignment optimisers then match jobs to technicians within the constraints of the day, such as working hours and available bays. We describe the general method here; the details of this network's data and rules stay confidential.
Results
- About 15% higher average gross shop revenue.
- One planning flow: duration predictions feed the technician-assignment optimiser directly, rather than sitting in a separate report.
What made it work?
We started from the assignment decision, not the model. The goal was a better assignment plan. Duration prediction was the input that plan needed most.
Prediction and optimisation belong together. Many projects stop at a forecast. The value here came from feeding the forecast into an optimiser that makes the actual decision. This is the core of our optimisation and simulation service.
Revenue was the measure. We judged the system by gross shop revenue, a number the business already tracks, not by model accuracy alone.
Service networks face related questions on the parts side too; see our spare parts forecasting case and our work for the automotive sector.
Frequently asked questions
How does job-duration prediction improve workshop revenue?
Better duration estimates let the optimiser fill each technician's day more fully. Fewer idle hours and fewer slipped jobs mean more paid work is completed, which in this project raised average gross shop revenue by about 15%.
Does the optimiser replace the workshop manager?
No. It proposes a technician assignment based on predicted job durations. The workshop team still runs the day and can adjust the plan when something unexpected happens.
Can this approach work for other service operations?
Yes, in principle. Any operation that assigns jobs of uncertain length to a limited team, such as field service or maintenance crews, faces the same decision. A free 30-minute AI (artificial intelligence) opportunity review is a simple way to check whether it fits yours.