Industry
  • Automotive
Location
  • Headquarters: Cleveland, OH
Use Cases
  • Building and deploying model-driven data products and solutions that increase dealership parts and services sales and customer retention
  • Tire Trigger predicts when individual consumers will need new tires
Impact
  • Increased client loyalty and new industry recognition as analytics/data science thought leader and service provider
  • Automation of the product makes it easy for dealerships and manufacturers to implement
  • Improved data science talent acquisition and development
  • Tire Trigger has delivered a 16% increase in raw conversion rates over the control group, which corresponds to a 21% increase in revenue over the same control group
Data Science Maturity
  • Each data science product, e.g. Tire Trigger, comprised of numerous models
Users
  • Team of five data scientists
  • End users include dealerships and auto manufacturers
Technologies in Use
  • Data Science Tools: Python
Algorithms in Use
  • Random forest
  • Gradient-boosting regressors
  • Linear regression

Data Science at Dealer Tire

Dealer Tire’s roots date back 100 years; the company began in 1918 as a single retail shop selling car tires, and grew to become a very large regional retailer. In 1998, the company pivoted away from the retail business to instead provide value-added, tire distribution services via car manufacturers and dealerships. Their mission is to help automotive manufacturers and dealers grow their tire and light maintenance business by providing tools and services that increase customer satisfaction and retention.

Why the focus on tires? Because they are a key defection point for customers, and selling tires drives dealer revenue and customer retention. Dealer Tire’s research indicates that people who buy tires at a dealership are 2.7 times more likely to return to the dealership for service, and are 1.3 times more likely to purchase a new vehicle from the dealership.

“We maintain partnerships with the majority of the major auto manufacturers, and we work with their franchised dealerships to make it easy and profitable to sell tires and light maintenance products,” explained Chris Schron, director of Data Science at Dealer Tire. “We consider ourselves a value-add distributor. Our whole business model is predicated on working with our partners to identify new opportunities that will increase sales and customer satisfaction.”

In 2015, the leadership team saw another opportunity: to leverage the company’s rich data and analytics capabilities to revolutionize its business yet again. The century-old firm invested to build a data science team that has since two revenue-generating data products to date, with a third currently in development. These products include consultative data projects, a predictive modeling suite built specifically for dealership fixed operations and a performance and coaching tool for dealership Service Advisors.

Challenge

Schron was asked to lead the data science effort and knew he had to produce results quickly for his team to demonstrate value. He wanted his team to focus on analytics and building products, but the early team struggled with DevOps challenges faced by many: limited access to infrastructure and analytics tooling that would allow them to build effective models. They could access a BI reporting system via drag-and-drop interface and use a API to pull data, but that was limited in both functionality and flexibility. The team also lacked data engineering resources.

Dealer Tire saw an opportunity to leverage technology to both allow existing data scientists to do data engineering, and also to maximize the scalability and effectiveness of their team’s work by automating infrastructure-related tasks like environment management.

Solution

Schron turned to Senior Data Scientist Chris McPherson, who in 2017 set out to create a collaborative space for data scientists to share and engineer data, experiment, build, and deploy models. A key goal was to automate the process of managing environments, employees, and data science projects as much as possible.

“We needed a solution that could balance flexibility with some standardization,” said Schron. “We didn't have the resources and we didn't have time to build a data science platform. Domino was the best solution we found to balance those needs.”

Dealer Tire implemented Domino in 2017 to form the foundation powering several key data science workflows that would ensure the data science team’s success:

Data engineering

Dealer Tire leverages a custom-built tool to author data migrations and pipelines either through SQL queries or Python code on Domino. Domino solved the complicated data pipelining challenge, allowing data scientists to combine raw data from multiple disparate data sources and then clean and transform it for exploration and modeling.

“With Domino, we now have managed environments per project and we've pinned all of the different libraries to specific versions,” noted McPherson.

Deploying data pipelines and models

Domino facilitates the seamless deployment of what has become a large tree of dependent models. Dealer Tire particularly values the ability to schedule multiple runs and to operationalize models through APIs or apps using Domino Launchpad.

“We’re fairly advanced in our technical maturation and Domino has been an accelerator that's enabled us to get there. I thought the value proposition was obvious from the start, and after a short trial, the rest of the team agreed,” said McPherson.

Can we predict when each consumer will need tires?

That was the first question the data science team was tasked with answering, and momentum snowballed from there. “It was the first ‘a-ha’ moment,” said Schron. “We realized if we could build some predictive models to answer that question, it could really add value.”

The team merged data from car manufacturers and dealerships to model individualized tire wear to identify the optimal time to drive each consumer to their dealership. This solution – “Tire Trigger” – was the first product Dealer Tire’s data science team delivered to the market.

Unlike traditional tire replacement models that are based on universal average miles, Tire Trigger is an ensemble of models that deliver individualized predictions. One model predicts the mileage at which a customer will need replacement tires. Another model predicts the number of miles per day that customer will be driving. And each of those models is comprised of additional models that ensure the most accurate predictions. Traditional models that treat all drivers as the ‘average’ driver results in many consumers being communicated too early or too late. The Tire Trigger model ensures that each consumer receives an individualized prediction that is consistent with their driving behavior and vehicle type.

“In total, there are some five different models in those two camps,” said McPherson, “and we put those two together to get a date. That’s the prediction. Then we run an experiment from it – we will randomize a control group and refrain from sending them any marketing, even though they’re predicted to need tires. And that’s how we’re able to measure the effectiveness of Tire Trigger.”

Once Tire Trigger predicts a date when each consumer will need tires, Dealer Tire experiments further to explore different marketing engagements at different times. The objective of this experimentation is to validate the most effective way to drive customers into the dealership for their replacement tires.

“We’re exploring a lot of things and we’re doing it all through the lens of an experimental design, so we can say with some certainty that Tire Trigger is what is driving these individuals to the dealership,” McPherson added.



The modeling before the modeling

Before the data even gets to Tire Trigger, Dealer Tire applies machine learning models to classify it because it’s often messy and inconsistent. For instance, much of the data from dealerships is human-entered, which is prone to errors, inconsistent nomenclature, and misspellings. Using natural language processing (NLP) classification, Dealer Tire flags parts that are likely to be relevant for Tire Trigger and other projects.

Then, Dealer Tire leverages a variety of models just to determine which model should be used in Tire Trigger for each individual’s prediction, based on the amount of relevant data they have about each individual. They use a model to determine which model to use!

The Domino Effect


Increased sales and engagement

Using Tire Trigger, Dealer Tire can provide manufacturers with a list of every individual that has likely hit a specific tire tread depth, which indicates it’s time for new tires. The team worked with a high-profile OEM to develop the product and test its accuracy, and is now focused on scaling Tire Trigger for widespread adoption, engaging more car manufacturers and dealerships and becoming an embedded tool that increases loyalty to Dealer Tire.

“We’re driving additional revenue, which they love,” explained Schron. “But a lot of the car manufacturers are focused on asking, ‘how can I get people interested and engaged in selling tires?’ So we’re bringing them new volume, we’re bringing them new customers, and we’re increasing that funnel on the front end.”

Since bringing the product to market, Tire Trigger has consistently delivered a 16% increase in raw conversion rates over the control group, which corresponds to a 21% increase in revenue over the same control group.

Another effect of Tire Trigger’s success is that Dealer Tire is now gaining traction as a trusted advisor for analytics and data science insights in the industry. Manufacturers are starting to turn to Dealer Tire for help with anomaly detection and other use cases.



Talent acquisition and development

Dealer Tire is better equipped to attract and hire talent with Domino.

“I don’t think you’d be able to hire the same people without a tool like this. It’s attractive to say, ‘You can use your language that you like, you can manage your environment the way you want to manage it, and it provides a lot of that functionality out of the box,” said McPherson.

Schron added, “It’s a good retention tool to show people that we’re serious and we’ve been thoughtful about how we’re giving people the tools to do their job.”

Dealer Tire’s junior data scientists appreciate Domino because it allows them to learn things they hadn’t otherwise been exposed to, such as working with containers and Docker, and it teaches them best practices for environment management.

“It's helped accelerate junior data scientists’ development. They're up and running in no time and they can add value right out of the box,” summarized McPherson.