Data Department helps automotive distributors and service networks use machine learning (ML), optimisation and AI (artificial intelligence) for parts and workshop decisions: how much of each spare part to hold, how long a repair job will take and which technician should do it. Our automotive work includes a forecasting system for 100,000+ parts and a workshop planning project that raised average gross shop revenue by about 15%.
Where does AI pay off in automotive distribution and service?
The after-sales business runs on two scarce resources: parts on the shelf and technician hours in the workshop. Both are planned under uncertainty, and both are expensive when the plan is wrong.
| Decision | What ML, AI or optimisation does | Typical data |
|---|---|---|
| How much of each part to hold, per location | Probabilistic forecasting that sets stock by acceptable risk, with separate models for frequent, sporadic and high-value parts | Order history, vehicle park by region, part compatibility and substitution, seasonality, weather |
| Which part to offer when one is out of stock | Substitution-aware forecasts and recommendations | Supersession and compatibility lists |
| How long a repair job will take | Job-duration prediction from job type, vehicle and history | Work orders, labour times, vehicle data |
| Which technician takes which job, in what order | An assignment optimiser that uses the predicted durations | Technician skills, shifts, bays, appointments |
| Which customers may stop coming back for service | Churn models that rank customers for service reminders and offers | Service visits, vehicle age, contact history |
| Where an answer sits in technical and warranty documents | A RAG (retrieval-augmented generation) assistant that cites its sources | Repair manuals, technical bulletins, warranty rules |
Relevant case studies
- Vehicle renewal and brand churn: an automotive distribution group. Two linked models, built on each customer's full history of services, complaints, purchases and other interactions, estimated the chance of replacing the vehicle within a chosen number of months and, if so, of switching to another brand. Prioritised calls and tailored price offers gave about 8% revenue gain.
- Spare parts inventory forecasting: a large automotive distribution group with 100,000+ active parts across several brands and channels (authorised service, dealers and direct sales). We built probabilistic forecasts and segmented models on AWS (Amazon Web Services) SageMaker with daily updates. They gave ~30% better forecast accuracy than the legacy method and ~10% lower inventory holding cost.
- Repair-shop technician planning: an automotive service network. Job-duration prediction feeding a technician-assignment optimiser raised average gross shop revenue by about 15%.
What are typical first projects for an automotive business?
- Parts forecasting for one brand or channel, starting with the items that tie up most capital. Our article on probabilistic forecasting for spare parts explains the method.
- Job-duration prediction for the most common repair types in a few workshops.
- Technician assignment once durations are reliable, measured on shop revenue and on-time completion.
- A technical document assistant for service advisers and technicians.
Each starts with a two-week Discovery: ranked use cases, a data check and a go/no-go decision. The free 30-minute AI opportunity review is a quick way to test fit. For warehouse, fleet and delivery questions, see our distribution and logistics page.
Data and integration notes
- DMS (dealer management system) and ERP (enterprise resource planning) hold parts sales, stock, orders and work orders. Forecasts and plans are written back so planners and service managers use them where they already work.
- Master data such as part compatibility and substitution lists is often as valuable as sales history.
- Vehicle park data by region helps forecast parts demand for older and less common models.
- Infrastructure: we deliver on AWS, Google Cloud, Azure or on-premises.
Frequently asked questions
Why use probabilistic forecasts for parts?
Many parts sell rarely and unevenly. A forecast range lets you set stock by the risk you accept, part by part, instead of trusting a single number.
Is workshop planning a forecasting or an optimisation problem?
Both. In our service network project, predicted job durations fed an optimiser that assigned jobs to technicians.
Do we need a data warehouse first?
No. Most first projects run on exports from DMS and ERP systems.
Can you work with several brands and sales channels at once?
Yes. Our parts forecasting system covers several brands and channels, including authorised service, dealers and direct sales.