CarMax drives business value with GPT-3.5

Generative AI such as ChatGPT has of late captured the imagination of business leaders across industries. While enterprise IT orgs by and large are taking a measured approach, some early movers are showing impressive results.

CarMax’s IT team, for one, has been working with Microsoft and OpenAI to leverage GPT-3.x for business value even before ChatGPT became a household name.   

That is why the omnichannel used-car retailer earned a coveted spot on the 2023 CIO 100 Award list: for its early, innovative use of a nascent AI technology that led to a spike in page views as well as higher SEO ranking and placement that drove substantial business growth.

CarMax EVP and CITO Shamim Mohammad, the brains behind his company’s digital transformation and AI push, would not specify how much money CarMax has netted from its AI investments and Microsoft Azure OpenAI Service to date, but the increased customer traffic no doubt has had an impact on revenue, while the cost implications of leveraging AI have also contributed to the bottom line, he says. 

“We would have had to have hired tens or maybe hundreds of content writers and taken years to generate this content,” says Mohammad. “We were able to do this literally in a matter of hours.”

The combination of CarMax’s revolutionary digital business model applied to the used-car business with revolutionary AI tools available to all makes for powerful and profitable business outcomes. Despite the current overall economic slowdown, CarMax’s Q4 2022 revenues rose 48.8% to $7.7 billion compared Q4 2021, with revenues for fiscal 2022 increasing 68.3% to $31.9 billion overall.

First-mover AI benefits

CarMax’s IT leaders and IT staff were experimenting with OpenAI’s GPT-3.x natural language model on pilots before Microsoft’s well publicized $10 billion investment in the nonprofit at the outset of 2023.  

In this early use case, the CarMax team employed GPT 3.5’s enhanced “iteration on prompts” to feed scrubbed and formatted data for thousands of used cars into a DaVinci model. Following that, a small dataset was sent for editing and fine-tuning and the content was pumped into the DaVinci model for mass publishing and consumer consumption.

Because use of the DaVinci natural language model itself requires fewer data scientists , which are scarce and super expensive, CarMax was also able to realize IT cost benefits, in addition to content creation savings.  

“The model’s ability to learn with just a few examples of intended outputs, a process called few-shot learning, helps CarMax’s 60-plus product teams use the models without requiring additional teams of data scientists,” according to a CarMax representative.

Creating car research content using AI eases the car purchasing process for consumers and the content creation process for editors. But perhaps most important, it drives an order of magnitude more inventory, eyeballs, reviews, and sales for CarMax.com. This is due to CarMax’s use of the cloud-based OpenAI API natural language model, which enables the extraction of millions — not thousands — of keywords.

As a pioneer, CarMax is reaping the early benefits of what will likely be a major business driver across the globe, one analyst says.

“As the use of generative AI becomes more widespread, it is causing significant disruption in many industries and sectors,” says Ritu Jyoti, group vice president of AI and automation research at IDC. “CarMax was at the forefront of embracing generative AI responsibly in partnership with Microsoft and has been a successful industry disrupter that has transformed the process of buying a used car.”

Looking forward

This is just the beginning, says Mohammed. Although he would not specify how CarMax is currently using the enhanced GPT-4 models released in March 2023, he is moving full steam ahead streamlining content creation for car research pages and scaling its use of GPT to help make buying a used car “effortless,” he says, adding that CarMax programmers are working on new aspects of the customer experience and will use GPT to gain efficiencies within the company. 

“My team is using the latest version for some other use cases, but we haven’t gone public yet,” he says. “That’s the way my team works. They like to experiment and try new things in a controlled way.”

Mohammed started the cloud journey and digital transformation when he became CIO of CarMax in 2014. Everything CarMax does that is new is done on the cloud but the company still has a small data center that will eventually be phased out.

Customer security is critical for CarMax, Mohammad says. The company claims Microsoft Azure’s security, compliance, reliability, and other enterprise-grade capabilities are what enables the company to scale its use of AI to enable the extraction of keywords and “filter out any harmful content in the user reviews.”

As a Microsoft Azure shop, CarMax relies on Azure Data Lake, an essential component of the company’s AI output, the CIO notes. “Data is the core of everything we’re doing because it feeds our machine learning algorithm that feeds our AI capability,” he says.

In fact, the blueprint CarMax has created within Azure can also be used by other companies, according to CarMax representatives.

CarMax is continuing to experiment and innovate using Azure OpenAI and GPT but Mohammad insists he is putting in place strong governance to ensure that machine learning and automation is used by its programmers in a disciplined manner to fulfill the company’s business goals.

“It’s not the wild west,” he says. “We have a good sense of what are the guardrails and what are the ways we’re going to be leveraging AI. How we deploy and utilize AI needs to be very much consistent with who we are a company.”