Edmunds sets stage for AI with data infrastructure consolidation

For a decade, Edmunds, an online resource for automotive inventory and information, has been struggling to consolidate its data infrastructure. Now, with the infrastructure side of its data house in order, the California-based company is envisioning a bold new future with AI and machine learning (ML) at its core.

“We’ve solved most of the consolidation challenges,” says Greg Rokita, assistant VP of technology at Edmunds. “Now, how do we stay ahead in this AI landscape? What foundation frameworks should we develop to make our product teams more productive and gain on our competitors?”

Rokita has been with Edmunds for more than 18 years, starting as executive director of technology in 2005. His role now encompasses responsibility for data engineering, analytics development, and the vehicle inventory and statistics & pricing teams.

The company was born as a series of print buying guides in 1966 and began making its data available via CD-ROM in the 1990s. The shift to online started not long after. Rokita came onboard as the company launched its first free online magazine, and several years later, his team launched the company’s first mobile phone apps.

Today, Edmunds’ website offers data on new and used vehicle prices, dealer and inventory listings, a database of national and regional incentives and rebates, as well as vehicle reviews and advice on buying and owning cars. The company was purchased by Carmax in 2021 for $404 million.

One of the ways Rokita is looking to stay ahead in the AI landscape is the creation of a new ChatGPT plugin that exposes Edmunds’ unstructured data—vehicle reviews, ratings, editorials—to the generative AI.

OpenAI, the company behind ChatGPT, trained the generative AI on a corpus of billions of publicly available web pages called Common Crawl. But in a world that moves at internet speed, that data rapidly falls out of date. The idea behind Edmunds’ new plugin is to give ChatGPT the ability to draw from its large collection of specialized and constantly updated data.

“If you ask it, ‘How does the Toyota Camry 2022 drive?’ you’re going to get nothing,” Rokita says. “By developing a plugin, we’re exposing our most recent data.”

For Edmunds, the hope is that users of the generative AI who want more details or pictures of a vehicle will click on a link to its site, driving traffic.

Much like the internet revolution of the 2000s that transformed nearly every industry, Rokita firmly believes we now stand at a new inflection point.

“Twenty to 30 years ago, the internet became entrenched within every company,” Rokita says. “We believe the same thing is happening right now with AI. It doesn’t matter if you’re an agricultural company, an industrial company, or a construction company, AI will be embedded within your company to optimize how you order materials, how you determine whether the crops need to be watered or not, and so on.”

If AI doesn’t become part of the fabric of the company, Edmunds will fall behind.

“Part of the challenge for my team is to create frameworks and jumpstart the company on that path,” he says.

Rokita believes the key to making that transition is to stop thinking of data warehousing and AI/ML as separate departments with their own distinct systems.

“People need to understand that these are really different manifestations of the same system,” Rokita says. “The data warehouse is about past data, and models are about future data. Imagine a table where you have past behavior and future behavior that’s predicted so it’s all one timeline.”

That idea drove Rokita’s determination to consolidate Edmunds’ data infrastructure, and like many companies that saw the advantage of new data technologies early, Edmunds’ data infrastructure grew as a series of best-of-breed point solutions.

“We started off with dedicated data warehouses built on Oracle racks, progressing through specialized systems like Netezza and Teradata,” he says. “We used to have Hadoop to process the data and then we would load it into Netezza for people to query it.”

About 10 years ago, Rokita grew determined to start consolidating that infrastructure. The first step was moving to the cloud. The team replaced Netezza with Amazon Redshift and later added the Databricks cloud platform for data science and AI. But the consolidation still hadn’t gone far enough: with different systems for data science, data warehousing, and data processing, the team still had to worry about data going out of sync.

“When you work with analysts and they see data in two different spots, and that data doesn’t match, they lose trust,” Rokita says. “It’s critical that users within the organization have a consistent view of the data.”

As Databricks added new data warehousing capabilities to its platform, Rokita made the decision to move away from Redshift and Hadoop and do everything using Databricks as a layer on top of AWS instead. That change has not only helped bring costs down, Rokita says it’s also made things easier to manage operationally.

“Now we have one system that handles both data processing and serving with the additional benefit that you can create models on top of it without duplicating data,” he says.

Now Rokita and his team are working with one of Databricks’ newest features, Databricks Marketplace, a marketplace for data, AI models, and applications. As part of the offering, Databricks is curating and publishing open source models across common use cases like instruction following and text summarization. Third-party data providers are also joining the marketplace, including S&P Global, Experian, Accuweather, LexisNexis, and more.

Rokita believes the ability to join third-party data to Edmunds’ data at the click of a button, without any development time, will open new vistas for the company and its use of analytics and ML.

“You can search for what you need, say, demographics data for potential shoppers for your cars, and then you can use it in your ad campaigns,” he says. “All you do is click on a box and then this data set appears in Databricks.”

In particular, he notes that Edmunds’ parent company, Carmax, runs its own instance of Databricks, but it runs on Microsoft Azure, while Edmunds’ instance runs on AWS. With Marketplace, there’s no need to unify infrastructure.

“Often, we want to share data between each other,” he says. “Now, without development costs we can share a data set with them and they can share a data set with us. We’re really excited not just about data sharing, but what’s coming next, which is model sharing and dashboard sharing.”

Artificial Intelligence, CIO, Data Management, Generative AI, Innovation, IT Leadership, Machine Learning