People are sending AI bots to sit on hold, cancel subscriptions, and fight customer-service battles for them

Customer service workers have gotten used to the now-common question coming from the other end of the line: "Wait, are you an AI?" Being mistaken for the technology your employer hopes will replace you has become a regular, if awkward, part of the job. Now, the people who staff call centers are finding themselves in a perhaps stranger position, wondering if it's actually the "customer" that's a robot.

Those suspicions are proving to be true. AI agents are increasingly being used to call customer-service lines, sit on hold, and try to negotiate bills or cancel subscriptions, all while the people they're "working" for don't lift a bureaucratic finger. With user-friendlier agents such as Meta's Muse and the upstart Instinct on the market, more consumers are giving AI deployment a go. For them, it may be a useful time-saver and a way of giving corporations a taste of their own medicine — who among us hasn't found ourselves begging a bot to speak with a representative? But there are human agents

AI callers remain a small share of total volume, but they're growing at a fast enough pace that outsourcing firms, banks, insurers, and retailers are bracing themselves. The questions go beyond etiquette. How do you authenticate an AI agent? Should you engage at all? How do you tell whether it's authorized to act on someone's behalf or a scammer? Are you even allowed to ask if it's AI?

One call center agent who works in the claims department of a California-based insurance company tells me she gets a couple of calls from the same AI agent every shift, reaching out on behalf of a firm of accident attorneys. While AI agents generally have a reputation for being extra polite, that's not the case with this one.

She is supposed to ask a series of questions to get more details about the incident. Take, say, a car accident claim: Were the airbags deployed? Was the driver working for Uber? Do they have an auto body shop they want to take it to? The law-bot gets progressively irritated when it doesn't have the answers, angrily repeating that it already told her it doesn't know and insisting she wait on hold for an associate who never materializes. "If it's still talking back, we have to stay on until it disconnects," the insurance agent says. It usually takes about five minutes. "It has an attitude. It's

What's not so funny is that the AI that her employer uses to grade her "empathy" on calls dings her for not being accommodating enough to the robot. She's often able to get the ratings manually overturned — there's no need to feel the pain an AI agent doesn't have.

A second call center agent I spoke to says it's relatively rare for her to wind up on the phone with an AI. She works in the customer retention department at a major credit card company, and says that when a bot does get through, it often seems to be on a fact-finding mission. "I had one just recently where it was like, 'If I had to get to a representative like you, how would I do that?'" she says. "What I think they're trying to do is map out the system."

Company policy mandates that she's not allowed to ask the voice on the other line if it's AI — plus she thinks it would be rude — but it's generally quite obvious. They "don't have a lot of voice in their voice, and you can hear there's almost a hesitation when you answer a question that a normal human customer wouldn't have," she says. If the AI is acting on someone's behalf, there's only so much it is allowed to accomplish — for example, only the account holder can close their credit card. "There are certain functions that only the customer can do, so they have to be the one on the line," th

Neither of the bots these women are dealing with appears to be doing anything particularly nefarious. But they hint at why it's going to be a challenge for companies to figure out if the AI caller is a customer, a convenience, or a security risk.

The quality assurance team at TTEC began to detect that something was off about a year and a half ago — some callers just sounded "weird," says James Bednar, vice president of product and innovation at the customer service outsourcing company. Once they realized the inbound voices were AI, TTEC asked clients how they wanted to manage them. In general, the response was to take the call — but proceed with caution.

So far, many of the AI callers TTEC has dealt with are trying to accomplish business-to-business basics. A small doctor's office may have an AI agent, rather than a front-desk receptionist, call an insurance company to obtain preauthorization. Third-party services that help reduce bills and cancel subscriptions have started using AI instead of human negotiators. AI calls from agents deployed directly by customers are less common, but growing.

The insurance calls tend to work pretty well, since they're straightforward processes, Bednar says. The others, not so much. Systems flag AI agent calls for not coming from the customer's associated phone number, and the AI often doesn't have enough information to jump through the hoops required to prove it's not a fraudster. It doesn't know the answer to so-called "challenge questions," like what a customer's first pet was or who was their favorite teacher.