It’s rare these days that a prospective car buyer would wander onto a car lot without first looking on the web to find information on models and features. Either by accident or intention, that buyer may find their self at Edmunds.com. Edmunds began as a company that printed a series of informational books on cars but now primarily lives online as not only an informational resource, but an inventory tool that connects buyers and sellers.
The value proposition for Edmunds on the consumer side is providing the right information for a car buyer, potentially guiding them to a decision on the make and model they want, then providing a link to a seller that is offering that product. Machine learning helps in this case by providing a personal experience, showing a customer only the information they want or need to know.
Machine Learning to Better Advertise Vehicles
Where Edmunds is really seeking out value from machine learning is on the dealer and manufacturer side. For example, that value comes from providing ways for dealers and manufacturers to better show their products. Edmunds is doing that through a machine-learning driven tool that can identify the best angles in photos of cars to ensure the pictures are most appealing to consumers, based on analysis of past customer behavior.
For manufacturers and dealers, such as Ford and Toyota, it’s about making sure advertisements are pointed at the people that are most likely to buy. Machine learning, driven by analytics, helps identify new levels of consumer behavior that go beyond typical advertisement strategy.
Traditionally, if a consumer is looking at a truck like the Ford F-150, advertisements on the page will point to other trucks, maybe a Toyota Tundra. Now, Edmunds’ analytics have identified that certain consumers looking at trucks may also be in the market for an SUV, so those analytics are used to inform machine learning algorithms to also advertise SUVs to truck shoppers.
Analytics-Powered Machine Learning
At Edmunds, the philosophy of building machine learning on the back of analytics insights is about linking machine learning to specific business problems, says Punnoose Isaac, executive director of analytics development at Edmunds.
“The biggest problem facing people who are trying to start in machine learning is that they don’t know the data well enough—they don’t have an understanding of the business,” says Isaac, who presented recently at Predictive Analytics Innovation Summit in San Diego, California. “Without that, the model is not going to perform. It’s a natural fit for analytics to be part of the solution.”
The solution of using machine learning to advertise not only trucks, but SUVs, to truck shoppers came out of a business problem that Edmunds faces: A highly competitive market for dealers’ marketing budgets.
Using Auto Dealer Marketing Funds More Efficiently
Dealers have about $500 to market each car, Isaac explains, and Edmunds must stand out among about 15 companies looking for a piece of that marketing pie. Edmunds solution was to create a high value audience by identifying the simplest variables, one of which was return visitation.
“Just the fact that they return to Edmunds raises the probability of completing a purchase,” says Isaac. “We took one more step to build a model out, integrated with Google and Facebook to continue to buy audiences and retarget audiences.”
Propelled by analytics, Edmunds deploys a machine learning algorithm that targets returning customers and then also offers them cars, trucks, and SUVs that they might be interested in but haven’t yet researched. Through this, closing sales for dealers has “definitely improved,” says Isaac.
How to Know if You Are Ready for Machine Learning
Isaac lists three boxes an organization must check before embarking on a machine learning program: data availability, analytics maturity, and cultural discipline.
Data availability is about more than just access to data, it’s about having data that is useful and applicable to business problems, he explains. Analytics maturity means a company is already using data to solve simple business problems. Cultural discipline refers to being smart about applying machine learning—to identify areas where machine learning could provide business value.
Machine-Learning Data in the Cloud
Moving to the cloud was also an important step toward machine learning, Isaac explains. He stresses the importance of partnering with the right cloud infrastructure provider. Edmunds went with Amazon Web Services (AWS), and it runs several analytical products on top of AWS.
The cloud helps when it comes to building business rules into data, which can help keep the data clean for future use, rather than applying data rules during the analytical process. Scaling out data with those business rules is not a problem with cloud computing, Isaac says.
Finally, Isaac says at Edmunds, the driving force behind identifying when and where to use machine learning comes back to the fundamental question of why the company exists in the first place, which is to make car-buying easier. Any other company looking to apply machine learning may want to ask themselves the same question: Why does our organization exist, and how can machine learning drive that purpose?