TORONTO — AI agents are consistently performing as well as humans at handling customer service issues at scale. The support bots are continuing to improve, and could regularly beat human assistants by the end of this year, according to one firm that sells the technology.

Toronto-based Ada claims the best of its AI agents can resolve more than two-thirds of inquiries on their own, without having to call for human help, and that they get customer satisfaction scores that rival those of humans. “We’ve proven that we can be better than a human,” said CEO Mike Murchison. “Now we’re going to get everybody there.” 

Despite the hype around generative AI, many businesses are still struggling to find useful applications for it. Customer service is one area where the technology is already producing measurable results.

Take Ada client Trust & Will. Along with routine requests for refunds and password resets, the firm’s support staff also handle more complex questions like how to structure inheritances without offering legal advice. Trust & Will is growing fast, said Sophie Heller, head of member support. After joining the firm in November 2023, she added 10 new staff to the five-person support team to handle the volume of inquiries, and switched on Ada’s generative AI agent to help out.

Ada’s AI agent started out resolving about 45 per cent of the firm’s inquiries on its own, but it’s now up to 70 per cent. Heller is impressed by cases where the chatbot goes back and forth with people four or five times, correctly answering questions like whether a will or a trust is a better option. That frees up support staff to answer more complex questions that Heller said still require human empathy. “I want my team to be providing that real touch,” she said, adding that her human agents still score better for client satisfaction.

There’s a risk with automating customer service for a process that generates as much anxiety as estate planning. “It’s a huge life decision,” said Heller, and customers sometimes just want to talk through their options with a human being on the phone. Trust & Will needed technology “that wasn’t going to make our customers angry.” 

Ada clients seeing human-level results from its AI customer service agents include Trust & Will, and Monday.com, a maker of productivity software, and Checkr, an employee screening service. On average, across its clients, Ada’s AI agents currently handle just over half of customer questions autonomously, with satisfaction scores slightly below those of human reps. To get better, such chatbots need more practice dealing with real inquiries. They also need information about how businesses run and need access to the same software staff use to look up customer data.

By the end of this year, Ada expects that its AI agents will consistently outperform humans, according to Murchison.

Ada has raised US$190 million to date, according to PitchBook data. Its competitors, including Intercom and Zendesk, also use large language models to power AI customer service agents they sell to businesses. Some larger firms have built their own in-house automation to improve client interactions. 

Insurance giant Manulife, for example, uses AI tools to help contact centre agents who are dealing with customer queries. The system is just as accurate as human reps, according to Jodie Wallis, the company’s global chief analytics officer, and better at some tasks. The tools produce more accurate summaries of conversations with customers and more thorough lists of the topics they cover, said Karen Leggett, Manulife’s global chief marketing officer, adding that automating note-taking means human reps don’t have to split their attention. 

AI agents have other inbuilt advantages, Murchison claims. For example, they typically work in more languages than humans. They’re also cheaper, and don’t quit. That may mean the work of those in the field of customer service—which employed 2.86 million people in the U.S. as of May 2023—could change significantly. 

At Trust & Will, support staff currently make up just over a tenth of the company’s employees—and they’re pretty familiar with working alongside AI. The AI agents help manage their workload, according to Heller. For example, she gave members of her team time off over the recent December holidays, despite it being the busiest will-making time of the year. “We joke about the week between Christmas and New Year’s being our Q5,” said Heller; Ada’s AI agents were able to help keep up with the surge in business.

That’s today. But in a few years, the customer service unit at a similar-sized company “could be a few people and a series of bots,” Heller said. 

Ada wants its AI agent to be the “number one customer-facing employee” for clients, Murchison said. He sketches out a future in which the bots are just like other staff, active in companies’ Slack and handing off work to human colleagues. “Our view is that Ada helps a single person do the work of 500 people,” he said. The firm is now helping businesses convert their support staff into managers for its AI agents, overseeing teams of automated assistants that do the work they once did. 

Ada’s clients “are under extreme board pressure to drive efficiency with AI,” Murchison claimed. He argued that “smart leaders” will use the technology to improve the quality of their customer experience and that some clients have already spotted opportunities for new products and services because of all the extra client interactions, prompting them to grow their workforces.

Still, Murchison acknowledged that the total number of humans employed in customer service may fall as AI agents take over. It wouldn’t be the first time the profession has been disrupted by cheaper alternatives. In the 2000s, many large firms in advanced economies hired call centres in emerging ones, where well-educated, English-speaking workers could take customer service calls in a familiar vernacular. 

Tech firms today are similarly training AI agents to field customer service inquiries in ways that feel familiar—only this time what they’re imitating is a human rep rather than your neighbour down the street. 

At some point, companies are betting, customers won’t care whether they’re talking to a machine or a person, as long as whichever one it is can resolve their issue. “They just want help and to get on their way,” Heller said.