Data Department built two linked prediction models for an automotive distribution group. The first estimates how likely each customer is to renew (replace) their vehicle within the next N months, where N is a parameter the business could set, for example to match a campaign window. The second estimates, if they do renew, how likely they are to switch to another brand. The scores were used to prioritise calls and tailor price offers. The result was about 8% revenue gain.
What was the decision?
A distributor knows its customers well: their services, complaints, purchases and accidents. But a sales team cannot call every customer about a new vehicle, and it cannot give everyone the same discount. The real decisions were two: who should we contact now, and how strong an offer does each customer need?
Contacting customers who are years away from replacing their car wastes calls. Offering a large discount to a loyal customer who would buy the same brand anyway gives away margin. Offering too little to a customer who is about to move to a competitor loses the sale.
What are the two linked probabilities?
| Model | Question it answers | What it changes |
|---|---|---|
| 1. Renewal model | How likely is this customer to replace their vehicle within the next N months (N set by the business)? | When to contact: who goes to the top of the call list |
| 2. Brand churn model | If this customer renews, how likely are they to choose another brand? | What to offer: how much the price offer needs to do |
Why they are linked. The second probability only matters if the first one is high. A customer who is unlikely to renew soon does not need an offer yet, whatever their brand loyalty. A customer who is likely to renew soon and likely to switch is the case where a call and a well-judged offer can change the outcome.
Read together, the two scores sort customers into practical groups:
- Likely to renew, likely to stay: contact in time, standard offer.
- Likely to renew, at risk of switching: call first, tailored price offer.
- Not likely to renew soon: no sales call yet; keep in normal service contact.
What data did the models use?
Both models were built on each customer's full history with the distributor: services, complaints, purchases, accidents and other interactions. A recent complaint, a pattern of service visits or an accident can all say something about when a customer will look for a new car and how open they are to other brands.
In general terms, projects of this kind turn that history into features per customer (for example, time since last purchase or the trend in service visits) and train classification models on what past customers actually did. We describe the general method here; the client's data and models stay confidential.
Results
- About 8% revenue gain.
- Prioritised calls and tailored price offers, driven by the two scores.
We report only the revenue figure for this project. We do not publish other metrics for it.
What made it work?
We started from the sales decision, not the model. The scores were designed to answer "who now?" and "what offer?", the two choices the team makes every day.
Splitting the question in two. Timing and brand loyalty are different behaviours. Modelling them separately made each score easier to act on and to explain to the sales team.
The approach travelled. The same family of churn and propensity-to-buy models was also used for our insurance renewal case. See also our automotive industry page and our churn prediction and customer analytics service.
Frequently asked questions
What is brand churn in automotive?
Brand churn is when a customer replaces their vehicle with one from a different brand. For a distributor, it means a lost sale at exactly the moment the customer was ready to buy.
Why use two models instead of one?
The renewal model says when a customer is likely to buy; the brand churn model says how likely they are to buy elsewhere. Keeping them separate lets the team set call priority and offer level independently.
What result did this project achieve?
The two models, used to prioritise calls and tailor price offers, produced about 8% revenue gain for the automotive distribution group. That is the only figure we report for this project.
Was the renewal window fixed?
No. The renewal model took the time window, N months, as a parameter. The business could choose N to fit the purpose, from a short sales campaign to longer-term planning.