What do stormy days, weekend evenings, and the last day of the month have in common? They are all good times to negotiate a good price for a new car. Inclement days yield less customer traffic in auto showrooms, which is good for the buyer. Weekend evenings, just before closing time, may make sales people more amenable to a deal. End of the month sales targets that have not been met generate similar pressure. But industry consultants are hoping that machine learning and analytics can curb discounting and extract more from the customer.
When it comes to artificial intelligence and the auto industry, autonomous driving, electric cars and ride/car sharing have captured most of the glamour, money and attention. McKinsey, though, forecasts that manufacturers could add $22 billion in revenue by “reducing vehicle discounts.” To put this in perspective, it is nearly triple the gain forecast from applying vehicle-generated data to research & development.
How to reduce discounts? Mainly by better predicting customer behavior and
- not offering big discounts to a customer who is going to buy anyway, and
- not offering discounts to low-value customers when better customers are available
The first goal suggests traditional predictive-modeling techniques – using customer attribute and behavior data (demographic information, data from any prior transactions, whether a customer is on a first, second, third, etc. visit, and more) to predict whether a discount offer is accepted.
In the second case, the calculus depends less on the attributes of individual customers, and more on general demand forecasts, using time series methods.
It is not a simple matter of setting an optimum discount level, however. Discounts have various sources (manufacturer incentives, dealer rebates, and affinity programs to attract groups, to name a few). Sometimes incentives discounts overlap; a Google blog reports that true net prices incorporating discounts can vary by as much as $6750. Google touts its cloud-based storage and analytics services as a way of bringing together different information sources (dealer, manufacturer, purchased customer data) to rationalize discount practices and make predictive modeling feasible.
In addition to optimizing discounts, predictive modeling can also optimize up-selling. Using the same predictor variables, analysts can predict which customers are likely to opt for add-ons to their purchase. This helps direct sales efforts at the dealer level, but also helps the manufacturing side match inventory to demand, avoiding the situation where the dealer has the car the customer wants, but not with the features the customer wants. McKinsey sees $9 billion to be gained on this front.