Artificial intelligence: Where humans and bots collide

From the September issue:

Artificial intelligence is becoming the ideal balance between “art” and “science.”

Imagine a future where customer service call centers are devoid of ringing telephones, and agents serve multiple consumers simultaneously. Imagine a system where each step of the customer journey is logged and tracked, eliminating the constant repetition of account and identity confirmation.

In auto finance, that future is not far off.

While lenders initially turned to AI to bolster fraud detection or enhance underwriting, some are quickly realizing that the technology will have a bigger impact on customer experience. GM Financial, for one, is leveraging tech powerhouses like IBM and Apple to augment its message-focused customer service strategy. Meanwhile, Kia Motors Finance is expanding the use of its virtual assistant, called Kian, with the help of chatbot-developer CarLabs.AI. Subprime lender Tricolor Auto Acceptance plans to use AI to help customers shop for cars online.

No matter the application, robust customer service is paramount in auto finance, said Marco Ambrosio, head of conversational commerce and marketing at AI-powered communication company LivePerson. Like home-buying, auto purchases are part of a narrow category of products that entail an involved process. “It’s going to take more than, ‘Here’s the product, here’s a picture, here’s the price, give me your credit card and [you’re] done,’” he said. “What better way to get through [the purchase process] than having a conversation with the car company itself — with the financing arm — to feel comfortable about your decision?”

In fact, LivePerson — which works with more than 18,000 brands, including banks and captive financiers — estimates that conversational commerce increases sales 25% and reduces attrition rates by as much as half. Further, customer satisfaction can increase 20%. “Customers vote with their fingers,” said Client Partner John Kelly, who oversees GM Financial’s account. LivePerson’s tech has resolved 20% of conversations without transferring to human agents, he added.

To that end, all AI applications are useful for auto lenders when it comes to customer service, said Ernie Garcia, chief executive at Carvana. “I always feel compelled to define what we mean by AI,” Garcia said, explaining it as a system that leverages data to make “good decisions.”

“You build your models to get more data, many times for many different customers,” Garcia added. “Then, it gets smarter.” But, for individual auto lenders with varying customer bases, the devil, as they say, is in the details.

The tail wagging the dog

At the helm of GM Financial’s customer experience strategy, Bob Beatty lets the technology drive the strategy.

“It’s almost like the tail wagging the dog,” said Beatty, the captive’s executive vice president of customer experience. “The availability of this technology that wasn’t here before has opened so many doors in my mind.”

Throughout Beatty’s two decades in auto finance, the customer service process has, on the whole, remained the same. “You call in, you identify yourself, you wait, wait, wait — then you identify yourself again, you tell somebody your problem,” Beatty described. “They look it up in the host system and hopefully they solve it and satisfy your needs. Then you go about it the same way next time, and you have to take a lot of time out of your day.”

Now, the industry is at a “watershed moment,” Beatty said, and customer service is hinging more on messaging than on phone conversations. As such, Fort Worth, Texas-based GMF is investing “heavily” in AI, along with data and advanced analytics, Beatty said. The captive’s most recent initiative, the pilot of Apple Business Chat, or ABC for short, will be rolled out this month.

With ABC, consumers using Apple devices to search for “GM Financial” will be able to launch a secure messaging platform with the captive rather than prompting a call to an 800-number. The customer will be connected directly with one of the company’s agents. “You can avoid that phone call nightmare that I just described,” Beatty said.

And that’s just the initial launch of ABC. “Very quickly,” Beatty said, the lender will incorporate that into the IBM Watson experience, which will launch on the consumer-facing messaging app as soon as quarter’s end. Around that time, GM Financial will debut interactive voice-response deflection, enabling customers to conduct call center business via text.

With the launch of IBM Watson and ABC, GM Financial plans to create a hybrid customer service model in which AI, messaging technology and live agents all play a role. “The very first interaction they have with GM Financial is with a bot that validates who they were speaking with and delivers that account to another bot, Watson for instance, that would then be able to handle their situation,” Beatty said. “If not, it would pass that onto an agent who could take care of anything.”

That interaction would be asynchronous, too, Beatty said, meaning that customers could message the captive at 8 a.m., have GM Financial respond immediately, and then pick up the same conversation hours later, at the customer’s convenience.

The differentiation between the messaging described above and chat — which is usually a single box on a website that requires the user to stay at the desktop to complete the conversation — is an important distinction in AI-powered customer service, said LivePerson’s Kelly.

“I really see it as an orchestrated way to give customers their life back,” Beatty said. “Let us work into [the customer’s] schedule instead of [the customer] working us into [their] schedule.”

Already, GM Financial has experienced a decline in call center volume. The captive averaged 23,000 messaging conversations per month in the past year, with a customer satisfaction score above 80%. By comparison, the satisfaction score for voice service was 76%. Further, if most customer service communication is achieved through text, Beatty said, theoretically one agent can handle multiple accounts.

“I really believe that 10 years from now we’ll look back and say, ‘You know, right around 2020, things changed,’” Beatty said.

Predicting the future

Meanwhile, after witnessing the success on its OEM’s website, Kia Motors Finance decided to leverage artificial intelligence to improve the user experience on its website through the use of a chatbot.

In the first quarter of 2020, the captive plans to launch phase two of its AI-powered chatbot, which will enable consumers to make account-specific queries online in an authenticated manner behind the firewall, said CarLabs.AI Chief Executive Martin Schmitt. CarLabs is developing similar capabilities for BMW Financial Services, which declined to comment.

Already, the bot has improved Kia Finance’s web satisfaction score and customer experience. “To offset call center volumes and operational costs for the captive, you have to be able to address account-specific questions per each customer,” said Schmitt, the bot’s mastermind and developer.

In addition to scaling call center volumes, captives — including Hyundai Capital America — use CarLabs’s technology to reduce costs. “Good customer support is expensive, it’s not scalable, and no matter how much you have it’s never enough,” Schmitt said, adding that cost reduction is often a lender’s top priority.

However, the most valuable benefit, Schmitt said, is the insight into the consumer’s mind. The software for CarLabs’s virtual assistants is rewritten every day based on customer queries. The process of crunching that data reveals intents or topics that may go unnoticed when looking at individual user messages, he explained. For example, CarLabs recently found increasingly more requests for mobile apps, buried in call center transcripts.

“It lets us know what we and our clients need to focus on,” he said. “What are the functionality gaps that they’re missing?” As such, what started as a simple initiative to improve user experience
online afforded the captive more data and insight than expected.

Further down the road, that customer data will be the basis for phases three and four of CarLabs’s technology to predict consumer behavior, not just intent, Schmitt said. Phase three will focus on the predictive side of artificial intelligence for customer management — to respond proactively to patterns in customer activity and behavior. “If a consumer has been consistently late on payments month after month, maybe that triggers a payment plan and restructuring so [the captives] can help and be supportive, as opposed to waiting until the loan goes to default,” he explained.

“Everybody wants to look into the future,” Schmitt said. “We’re using these tools, to a large degree, for prediction.”

With that in mind, phase four will address risk scoring and classification. For instance, lenders will be able to use analytics to look for correlations between customer attributes and their creditworthiness. “That way, you don’t have separate initiatives within the organization, each working on their own models and using their own tools,” Schmitt said, though he declined to offer a timeline for the phase three and four rollouts.

Balking the bots

For Tricolor Auto Acceptance, the use of AI with machine learning in underwriting and risk segmentation has provided access to a segment of consumers that would otherwise be credit-invisible, said Don Goin, president and chief operating officer.

“Having seen the success of that, we think there’s more opportunity across pretty much all of our business to get these concepts in play to produce either a better customer experience or to drive out operational inefficiencies,” Goin said.

However, Dallas-based Tricolor has taken a different AI path than GMF or Kia Finance. “Our customers tell us that they don’t like chatbots,” Goin said. “They’re not natural — the language processing isn’t quite good enough to seamlessly interact like a human being.” As such, Tricolor is “reticent” to put robotics in front of its customers because “it’s not going to be easy to
answer all their questions” due to regulations, he added.

Further, Tricolor’s customer base primarily interacts with the lender via mobile text messaging, such as WhatsApp and Facebook Messenger, Goin said. “If we had a global call center to have easy calls or easy interactions, such as payment assistance or, ‘Hey, what’s my balance?’ I could see where chatbots could help with efficiency,” he added.

Instead, the buy-here, pay-here lender is interested in AI for conversational commerce. “We’re working on the ability for consumers to engage in online shopping and auto buying through a multichannel, text-based interface,” Goin noted. For example, say a customer engages with Tricolor’s interface looking to buy a red Ford F-150. Tricolor can present inventory that matches
the request — complete with pictures, location, contact information and closest dealer — without the customer ever having to browse online, he said.

“We think this is the natural point of interaction with our customers on their preferred channel,” he added.

Scratching the surface

Carvana, which acquired Propel.AI last year, utilizes the tech’s AI to better serve customers online by engaging with incoming SMS leads. Mojo, as it’s called, engages with customers “within seconds” to provide vehicle model, mileage, and aesthetic information, according to Propel’s website. In addition, the company estimates that 60% of incoming leads engage with Mojo for a 23% increase in incremental sales.

In fact, Propel’s technology supports several features within Carvana’s online support menu. Similar to Kia Finance, customers can enter into chat conversations online for customer support. Beyond that use case, Tempe, Ariz.-based Carvana deploys the same underlying technology to offer up FAQs and explainer videos based on the customer’s clickstream, Garcia said.

“The more time you’re spending on the site, the more we’re learning about what questions you might have, and we’re putting questions in front of you to try to answer the questions you’ve got,” Garcia said. Explainer videos are also updated and prioritized based on clickstream.

Feedback from customers and within data has been encouraging, Garcia said. Customer service agents have reported that customers are better informed by the time they get to the live agent because initial answers have been supplied in an automated fashion.

AFN’s investigation into lenders’ uses of AI in customer service just scratches the surface of the technology’s potential. However, before companies can reap the rewards, they will need to ensure all departments are on the same page, said GM Financial’s Beatty.

“Business owners don’t realize the power of these advanced analytics and the data scientists don’t know what the business problems are,” Beatty said. “If you can get them to talk to each other, light bulbs go off.”