Adoption of chatbots is on the rise. So are questions about how they can do more.

Bank of America reported 6.3 million users of its virtual assistant, Erica, in the first quarter of 2019, up from 4.8 million the previous quarter. The AI-powered chatbot has completed 39 million interactions since its launch, the bank said.

TD Bank‘s Clari, an AI-powered chatbot added to the Canada-based bank’s app in January, already gained more than 100,000 registered users by its Q1 earnings report in February. Other traditional banks, such as Capital One, JPMorgan Chase and Wells Fargo, have chatbot deployments with various capabilities and use cases, and digital banks are implementing them, too.

But not everyone is impressed.

Josh Sutton, CEO of Agorai, a provider of AI tools, said natural language processing is an area that large banks, especially, are rolling out efforts in and pushing into, but with “pretty dismal results.” Even at BofA, just 23.2% of the bank’s active mobile users are also Erica users.

“The concept of a chatbot, while it eventually will be outstanding, hasn’t quite jumped the chasm yet into something that is intuitive and easy to use the same way that a mobile banking interface is,” he said.

Asked where chatbots are falling short, Sutton said, “Common sense.”

Also Read: Why Bank of America’s Erica Cannot Help with Complex Transactions

“Natural language tools, in general, right now, are very good at enacting specific commands,” he said. “They’re very bad with things that are vague.”

He said one area ripe for optimization, however, is how to take customer feedback, from sources such as call center logs, and use that data to optimize the customer experience.

“Call center optimization is a huge area of opportunity for banks right now because it’s a massive cost,” Sutton said. “It’s really about understanding how to optimize and route those calls properly, which entails understanding what the root causes are for what somebody is calling about and getting them to somebody [who] can address that quickly. Frankly, it’s a perfect use case for natural language understanding.”

Digital bank TymeBank and conversational banking technology provider Finn AI recently partnered to build a low-cost, AI-powered chatbot and financial literacy coach called Max, which resides within the bank’s TymeCoach app.

“The key for us is to find that perfect, teachable moment,” where the customer needs to know something and is interested in learning about it immediately, said Coenraad Jonker, co-founder and CEO of Tyme, and chairman of TymeBank in South Africa.

The bank offers its customers the option to speak to a human to resolve an issue, albeit as a last resort.

“That’s an expensive option for us,” Jonker said, adding that keeping operational costs down as his young bank scales up is critical to its long-term viability.

Jake Tyler, CEO of Finn AI, said humans are already no longer the default channel for customer service, but an escalation channel.

“You’re always going to have humans in the mix, but probably not for the routine, transactional, repetitive queries you’re getting hundreds or thousands of times a day, rather, more for the complex things,” he said. “Industrywide, we’re moving out of call centers and branches to messaging being the big channel.”

Jonker said he foresees a future in which humans will not be allowed, by regulations, to advise others on financial matters without having a chatbot open in the background. “Even when you choose to speak to a human, we’ll expect that the AI will actually lead the human advisors into action,” he said. “You’ll essentially have your own private banker who doesn’t get tired, doesn’t get grumpy, doesn’t forget the previous conversation with you, doesn’t make calculation errors, and isn’t incentivized to sell the wrong product to you.”

The margins for error in regulation, Jonker added, are narrowing. But he also acknowledged this chatbot-reliant future will take a while to come to fruition. “Our AI is still young and inexperienced,” he said.

Asked whether regulatory required chatbot-assisted financial advice could be the future, Christopher Harris, Business Transformation & Operations Improvement Leader, Americas Financial Services Advisory for Ernst & Young, said, “Maybe.”

“I mean, I haven’t seen anybody replicate Warren Buffett yet,” he told Bank Innovation.

While Harris said chatbots could likely provide material assistance to two-thirds of financial management advisors in the world and effectively satisfy the financial and investment needs of up to 80% of Americans, history shows that disruptive technologies — such as the computer itself — more often lead to new capabilities or new jobs altogether than a diminished role for humans.