Matt Marcus, senior VP and executive creative director at R/GA, is pitching Cars.com on the benefits of having an AI strategy. "If you're not thinking about AI, you're going to get disrupted by someone else," he says to a room of the brand's marketing executives, who gathered at R/GA's Chicago's office late last month. "But you clearly are thinking about it and we'd love to help you out."

Cars.com is already developing a Facebook Messenger chatbot and voice apps for Google Home and Amazon Alexa. But its efforts, which haven't been released to the public yet, have a check-the-box quality. Though they are working as designed, it's hard to see them "moving metal," as car dealers say.

Marcus and his partner at R/GA's new Brand AI practice, Executive Tech Director Michael Morowitz, have seen dozens of companies go the same route. They're trying to convince brands that R/GA can bring them AI on a different level. But there's a gap between marketers' desire for something called AI and their ability to fund projects that swing for the fences with agencies' young AI units. "The No. 1 mistake clients make is thinking about a narrow solution for their consumer-facing AI," Marcus says in an interview later. "Narrow solutions get used once and abandoned."

Marcus and Morowitz are tasked with growing R/GA's Chicago-based Brand AI practice, responsible for developing AI solutions and strategies across the agency's portfolio of clients, as well as potential new ones too. Brand AI has nine clients so far, and 30 dedicated employees, according to R/GA.

This is not a test

Other agencies, such as Atlanta-based Moxie, provide similar services. Wunderman has a team of more than 100 dedicated to AI. Epsilon's Agency AI practice includes 25 data scientists, technologists and experts in "multimodal user experience" (that's voice, vision and touch).

But AI is expensive: A good program can run nine figures. And the results aren't easy to measure.

The R/GA team cautions clients that they won't get far if they consider their AI projects a "test," according to Marcus. "Positioning these as tests is tantamount to failure," he says. "If there is no marketing or support, they will naturally fail."

Marcus and Morowitz made their mark on an assignment for The Cosmopolitan of Las Vegas. The hotel has a music venue, stores, restaurants and spa-like services spread across multiple floors, often leading guests to overlook them and spend their money elsewhere. The answer was Rose, an AI chatbot that interacts with guests through text messaging. It got more people using the in-house facilities, increasing revenue from guests who interacted with Rose by 30 percent—measurable by comparing spending by those who used Rose with those who didn't.

But you can likely count on one hand the number of consumer-facing AI products that clearly lift revenue—as opposed to other, less fungible results, such as lifts in brand awareness or engagement.

Brandon Purcell, principal analyst at Forrester, who says he would never tell a client to do AI just for the sake of doing AI, believes there's a right way to approach it. "You need to start with a business objective because the kind of AI you might develop could cost millions," Purcell says. "People want to tick the AI box without thinking about what the customer needs."

Epsilon Chief Digital Officer Tom Edwards says understanding data is the key. "Without a solid understanding and corpus of data as a starting point for AI-driven experiences, consumer-facing experiences have the potential of falling flat," he says.

For R/GA, that meant learning the 300 pieces of art on display at the Cosmopolitan's underground parking lot—and what customers were asking for. "You'd be surprised how many people show up to Vegas without a toothbrush," Morowitz says.

The agency uses that sort of information to create a chart showing each customer query and the response that goes with it. With Rose, Morowitz says, "When they saw this and the map of how Rose looks at their business, it became a training document."

Again, however, AI often means more than a chatbot. (R/GA's other AI services include computer vision, traditional customer support and marketing communication.) The Brand AI team isn't even that hot on the term. "We prefer 'augmented intelligence,' " Marcus says. "Chatbots suggest limited functionality and reference the first wave of these consumer-focused AI experiences. In general, they are narrowly focused marketing tools. They are used once to deliver new content about a new product. Or they do one thing, like help you order flowers or pizza."

Remember the people

After more than an hour, R/GA's meeting with Cars.com comes to a close. Figuring out an AI tool that consumers actually use will take many sessions, plus a granular view of the brand's needs, what rivals are doing and what people really want—and deserve.

"AI-based technologies are going to cause a large-scale disruption across the board, including the automotive industry," says Deepak Kaimal, VP of software engineering at Cars.com.

"The one thing we'll all need to be considering as AI technologies become mainstream," he adds, "is the need for more governance over how algorithms behave, to make sure they have the best interests of the users."

~ ~ ~
CORRECTION: An earlier version of this article said Cars.com has created a Facebook Messenger chatbot and voice apps for Google Home and Amazon Alexa. Those efforts remain in development and haven't been released to the public. The article also referred to a gallery at The Cosmopolitan of Las Vegas. There is art throughout the resort, but no standalone gallery, according to the hotel.