A hypothetical shopper, Hans, is in the market for a car, but he’s not sure exactly what he wants. New or used? Which make, model and year? Hans can’t define his search by these parameters, but he does know he wants something in the $30-35K range, and he wants fuel efficiency.
Before the release of the CarGurus ChatGPT plugin, Hans would have applied various filters to narrow down his options. But today, he can become a ChatGPT Plus subscriber (if he isn’t already one) and use CarGurus’ plugin to ask, “What are the top three best deals on fuel-efficient cars under $35K near me?” As if in conversational response, the screen would then show him the most relevant vehicles on CarGurus’ site.
“We are always looking for ways to support our dealer customers and people who use our site to buy, sell and finance vehicles,” said CarGurus VP of Product Ben Kasdon. “By integrating natural language processing into search, we’re enabling consumers to seamlessly discover options that may not have been on their radar earlier in the shopping process.”
In addition to elevating the car-buying experience, the new plugin also helps CarGurus better understand what matters most to shoppers, distilling this valuable insight to enhance the solutions it offers its dealers.
“All of this will serve us well as we continue to roll out AI-powered offerings,” Kasdon noted, alluding to future technological advancements on the horizon for the company.
Built In sat down with Kasdon to learn more about the development of the CarGurus ChatGPT plugin, including the culture, teams and considerations that brought this effort to fruition.
How did the CarGurus ChatGPT plugin enter your product roadmap? What were your areas of focus in development?
We’ve used machine learning algorithms for years to optimize our user experience and site operations, so it was a natural progression for us to incorporate generative AI.
But we wanted to make sure it provided value to our users and delivered a reliable experience — a unique challenge with a technology that’s been evolving in real-time.
“We wanted to make sure it provided value to our users and delivered a reliable experience — a unique challenge with a technology that’s been evolving in real-time.”
To do this, the team focused on three key areas:
1. Product discovery: Harnessing generative AI’s innate ability to translate human language or subjective requests like “fast,” “big” and “safe” to concrete recommendations.
2. Efficiency: Responding to customer inquiries with accuracy and speed.
3. Insights: Growing the number and type of actionable insights we can learn from and apply to optimize this offering and inspire others.
From a tech standpoint, this first version of the tool is pretty simple. We spent most of our time fine-tuning the initialization prompt that we send to OpenAI to ensure we get high-quality, relevant responses. Next, we leveraged OpenAI's API to pass that initial prompt along with the user input and some additional guardrails, and then mapped the response to specific inventory for sale on our site.
Did you run into any obstacles?
We spent considerable time optimizing our prompt and building guardrails around the potential responses. This ensured we could give people helpful answers when they asked questions relating to our core capability of connecting people with vehicles that meet their specific needs — but it also ensured we could manage responses to queries on topics that are beyond our scope. We can help you find a car that is fun and sporty, but not your missing sock!
We also had, and continue to have, a lot of dialogue around balancing performance and cost. Having strategies in place that help us pace our development are key to managing the growth of this exciting new offering.
A TEAM EFFORT
“Collaboration is a core value at CarGurus, and it’s key to our ability to innovate and move quickly across functions and teams,” Kasdon said. Here’s the role each team played in the ChatGPT plugin product:
- Engineering and product teams work “in lockstep” to design, build and test.
- Analytics team members set the company up to identify and extract insights that would encourage users’ future interactions with the tool.
- Marketing and PR teams help generate excitement around the offerings while building users’ trust in the nascent technology that powers it.
- Legal and information security teams are instrumental in ensuring that data use and privacy challenges are implicit in the design and thought process.
- Sales and revenue teams help the company better understand how the tools and the insights they generate can help CarGurus’ dealer partners in the long term.
When you think of other companies in your industry, how does CarGurus compare when it comes to how you build and launch new products?
Six years in, I’m still impressed by how we get things done at CarGurus. We’re thoughtful but fast, and highly collaborative with a voracious appetite for data and emerging technologies, and an unyielding commitment to building a better experience for customers and users. There’s something really unique and energizing about the way those things combine at CarGurus to help us unlock the potential of this growing industry.
