How Comerica Bank is Tapping Voice AI to Improve Its Call Centers

As part of a broader push to upgrade its technology, Comerica has implemented voice artificial intelligence tools to automate much of its contact center’s interactions with consumers. The regional bank has also launched voice-based authentication.

Both initiatives are continually adjusted. There’s no “next generation” or “2.0” in Comerica’s approach. User experiences dictate changes, sometimes as often as daily.

While Comerica’s overall voice efforts have already paid off in increased efficiencies, and voice authentication has been well received, the Dallas-based regional bank says it is still early days and hopes for continuing improvement in consumer service and efficiency.

“We consider voice as a fledgling space,” explains Nishanth Varanasi, Vice-President for Platform Management, at Comerica, which has $73 billion in assets. “There’s a lot of promise in voice AI but it is still evolving at a very rapid pace.”

Conversational AI Is Not An Upgraded IVR

The effort to add voice AI and voice authentication to Comerica’s contact centers began in 2018 as part of TechVision 2020, which is Comerica’s technology strategy. Between 2016 to 2018, Comerica Bank went through a major digital transformation to enable growth in earnings and revenues. As part of that transformation program, the new technology strategy of the bank, TechVision 2020, was established. This consisted, first, of key initiatives for strengthening the core such as providing high systems availability, improving cybersecurity, and upskilling talent. Second, the company set out to modernize its computing, create enterprise-wide agile capabilities and enable a richer customer experience. The latter is where voice AI and related efforts fit in.

The first stage, voice AI, was fully implemented for consumers for retail banking and tech service desk queries in 2018 and thus far has improved customer support efficiency by over 50%, according to a Comerica investor presentation. Voice authentication adoption began in October 2019. The voice-related tech implementation dovetailed with investments in the bank’s customer relationship management platform.

“We consider voice as a fledgling space. There’s a lot of promise in voice AI but it is still evolving at a very rapid pace.”
— Nishanth Varanasi, Comerica

Transforming its contact centers began with the decision that Comerica’s approach would be deep and broad, including creation of application programming interfaces (APIs) and updating chatbots, creating a conversational interactive voice response system and adopting voice biometric solutions.

Varanasi prefers to avoid using “IVR” when speaking of the system because he believes that label still carries associations with historical approaches — old, clunky touchtone systems with robotic voices and limited functionality. That term “suggests how technology worked in the past,” he explains.

“Conversational AI is the next stage, and it is what is making voice way more popular than it was before,” explains Philippe Dintrans, Chief Digital Officer, Banking and Financial Services, at Cognizant, the consultancy Comerica has worked with on aspects of its voice efforts.

Ultimately, combining conversational AI with people in contact centers puts more of the routine on the machine. “At the end of the day,” says Varanasi, “this allows our agents to have more high-value conversations.”

Varanasi adds that the point is not just cutting the costs of customer service nor improving “containment rates” — the “batting average” that indicates the percentage of calls not transferred to contact center staff. It’s about improving the customer experience.

Read More:

Benefits of Using an ‘Ecosystem’ Approach

Varanasi explains that Comerica strives to balance the drive for agility with the need to control the risk of adopting tech that could wind up in a dead end. This means avoiding putting all of the bank’s chips on one product or concept.

“What we’ve done is make strategic investments into key ecosystems,” says Varanasi. “We took the ecosystems route primarily because it gives us the ability to create a sustainable solution that’s also a progressive solution. We try to stay away from one-off technology.”

The idea is to set up a stable tech foundation that can be transformed through modules. “We can rebuild or replace pieces if we decide we want our approach to evolve differently,” Varanasi explains. “The benefit of using an ecosystem approach is that where you see ‘white spaces’ there are several providers trying to rapidly address them. So the ecosystem gives us the advantage to quickly hop on and try something new. And we can do it in a very short time frame, observe results, and make future tech choices based on what we learn.”

Voice Authentication ‘Learns’ Daily

Building a strong dictionary for the AI to begin working with was a key early task and one that doesn’t really end, in Varanasi’s view. The bank’s system learned as testing progressed, and the intent is that this will continue.

“We actually do updates on almost a daily basis at times,” Varanasi explain. “It depends on what the system is currently learning.”

One example of ongoing learning concerns VoiceSafe, the voice authentication system referred to earlier.

“We redesigned the whole dialog that customers have around identifying themselves,” says Varanasi. New skills had to be coded into the system to accomplish this.

“When we launched VoiceSafe we started learning more from some of our initial calls,” he continues. “We found there were gaps in our dialog design. So we started making updates almost nightly, in order to make sure we captured some of the low-hanging fruit and addressed other gaps right out of the gate.”

This is a small illustration of the overall challenge of bringing up a voice user interface for a banking institution. There is much that is very specific to banking, especially in the area of identification. Varanasi points out that it’s not just a matter of voice recognition and verification, but associating the right voice characteristics with the proper account IDs and more.

Better to Build a Voice Library from Scratch

Another key factor to appreciate: Conversational AI is about enabling the customer to have just that, a conversation. Part of voice AI is using natural language processing to translate words, which are ultimately symbols, into what practitioners call “intents.” These intents, according to a Cognizant backgrounder, are a key aspect of voice AI. While whole libraries of intents can be purchased, building a library from the ground up is the approach Comerica followed. The idea is to tailor the technology to the bank’s own consumers.

Humans don’t communicate the way computers do. We zig, zag, and change our thinking in midsentence. We interrupt each other.

One of the early design gaps that needed to be remedied in the first phase was prompted by what Varanasi says was the lack of a “‘speech interrupt feature,’ which allows interruption of a thought as a conversation progresses.”

“Business banking conversations are often very consultative in nature, and they tend to be particularly ‘context heavy’.”

Consumer conversations can also become complex enough that staff intervention may be necessary. Varanasi explains that calls concerning mortgages may require human conversation. One capability that Comerica has been working on to help in such situations is enabling its system to tap into past experiences with a consumer to help direct conversations to the right staffer, such as a mortgage specialist.

While consumer banking conversations can become complex, Varanasi points out that business banking conversations present even more challenges.

“Business banking conversations are often very consultative in nature, and they tend to be particularly ‘context heavy’,” says Varanasi. At this time Comerica is not using voice AI on the business side. When the bank addresses this, he says the machine learning approaches used will be set up exclusively for the business side. Using a “domain-centric” approach would be logical, he explains, rather than trying to solve everything with one tool.

Read More:

Getting Consumers on Board in Different Ways

Comerica approached the two voice initiatives differently when it came to promotion.

The implementation in 2018 of voice AI was done promoted gradually with some branding and other communication to create awareness of the new features of contact centers. Authentication looked like a more definitive sell.

“When we introduced Comerica VoiceSafe we spent a lot of time and effort branding it — it’s trademarked. Our marketing teams were heavily involved in getting the message of the capability out to customers,” says Varanasi. Before the launch, numerous adjustments to policies and procedures, as well as staff training, also had to be in place.

Varanasi says the bank was a little uncertain about how customers would adapt to voice authentication. “Authenticating voice is not something you see a lot of banks doing yet,” he says. Without giving specifics, Varanasi says Comerica has seen dramatic acceptance and continues to make progress.

“We had to be absolutely sure that our customers fully understood how we can enroll them in VoiceSafe and how we protect them,” says Varanasi. “So we actually spent a lot more time building up that awareness than building the technology integrations themselves.” (Very sensitive transactions may require a second level of authentication.)

Voice authentication requires an initial call to set up verification and voice recognition, with the system taking a voiceprint. One factor he thinks has helped sell people on voice authentication is that afterwards the bank can do much more for consumers via the contact centers.

You Don’t Have to Be a Tech Nerd to Take to Voice AI

“Comerica Technology’s strategic goal is to become a fully digitally-enabled, data-driven relationship bank with a reputation for exceptional quality and reliability that will drive extraordinary colleague and customer experiences,” says Sangy Vatsa, EVP and Chief Information Officer, Comerica Bank. “This strategic goal will serve us well as we think beyond 2020 with an eye on 2025.” What’s next on the voice front, particularly given consumers’ growing interest in smart speakers and digital assistants? Varanasi could not discuss specific plans, but he does note that the appeal of voice user interface spans across demographic groups.

“I think we’ve found that there isn’t a ‘tech-centric’ customer type, or that every consumer can be a tech-centric customer and that’s something we’ve learned to appreciate as we build this technology,” says Varanasi. “Voice is very convenient, native and humanized way of interacting. The onus is on us to create the technology that gets the customer’s sponsorship.”