I've been fighting with my health insurance company a lot lately. The mundane billing disputes are exactly the type of situation that, theoretically, AI should make easier. That, however, is not what's going on. The first point of contact is the AI-powered online virtual assistant, which asks what it can help me with but has, thus far, never been able to actually help. After some back and forth, it directs me to an allegedly real person who's supposed to be better equipped to handle the matter. A lot of the time, I get referred to a phone number to call instead. Once I call that number, I'm presented with a new robot — this time, one that talks. It's not any better at understanding my problem than the typing robot, but it's also not so sure I'm ready to get to an agent just yet. Yes, it understands I'd like to speak with a representative, but why don't I explain what about first? As my frustration grows, I can hear my voice rise to a Karen-level pitch I swore I'd never use.

By corporate America's (sometimes dubious) telling, AI is basically the answer to everything, including customer service. Businesses say it's the way to unlock efficiencies and improve customer "journeys" so people can solve their problems and get what they need on their own, and fast. The bigger, though less advertised, focus is how AI can save companies money and cut costs, whether by helping human assistants or, in likelier scenarios, reducing the need for human assistants at all. Corporations have long seen contact centers as cost centers, and ones they're constantly looking for ways to reduce.

"It's a lot of work, and it's expensive to think about customer experience and design your AI in a way that's going to be an enjoyable experience," said Michelle Schroeder, the senior vice president of marketing at PolyAI, which creates AI-based voice assistants. "And most companies that are thinking about cost cutting and the AI revolution are not really thinking about the customer."

Simply put, the AI still doesn't work that well. Many of these chatbots and virtual support agents are not ready for prime time. People don't want to use them, but they have to anyway.

"Companies are operating in the dark, in some sense. They have this idea that this technology is going to provide them with cost savings," said Michelle Kinch, an assistant professor of business administration at Dartmouth's Tuck School of Business. "They don't exactly know how to deploy it."

At the moment, customers are the guinea pigs in companies' experimentation with AI. We're the ones navigating the mishaps, overcoming the hurdles, and serving as case studies for what works and what doesn't. The hope is that all this testing will pan out, and the AI will get better as time goes on. But that's not the only outcome possible. We may just be consumers, standing in front of a chatbot, begging to talk to a real person forever.

Consumers are already suspicious of the whole chatbot thing. A recent Gartner survey found that nearly two-thirds of customers prefer that companies don't use AI for customer service. The main reason for their concern was that it would make it harder for them to reach a person. They also worried it would take jobs and give the wrong answers. A J.D. Power survey found bank customers aren't sold on AI. Some academic research indicates that when consumers hear "AI," it lowers emotional trust, and that consumers evaluate service as worse when it's provided by a bot versus a human, even when the service is identical. People think automation is meant to benefit the company — as in, save money — and not them.

Many of them use AI in their daily lives, to some extent, like using ChatGPT to research a product or ask a question about a warranty, said Keith McIntosh, a researcher at Gartner. They're just wary in a customer-service setting that it won't do the trick. "They know the tools can work, but they're just worried that service organizations will use it to just block access to a person and probably do not trust yet that the technology will actually give them a solution," he said.

Companies need to reassure customers that they're actually using AI to deliver a solution they can use in a self-service way and offer a clear path to an agent when necessary, he said. That sounds nice, but that's often not the reality. It's tough, if not impossible, to get a real person on the phone in a way that can be deeply frustrating and anxiety-inducing.

"When we do have that acute need to talk to a person, the chatbot becomes a hurdle," Kinch said.

Even setting aside the cost savings for companies, there are clear reasons that AI should be a good fit for customer service. When people reach out to a company, it's often with the same basic questions — when is my package arriving, where are my tickets, what is the balance on my checking account? Generative AI chatbots are good at distilling this sort of simple information and packaging it in an easy-to-read, conversational way — assuming they're not making stuff up.

"Most companies have tiered operations where they have tier-one, tier-two, tier-three support in increasing complexity, and that tier-one support is typically the sort of high-volume, low-complexity type questions," said Jason Maynard, the chief technology officer of North America and Asia Pacific Zendesk, a customer-service platform. "We're already seeing some customers that are really successful at automating a lot of what has been typically like their tier-one operations."

He pointed to DraftKings, which has millions of players, many of whom have basic questions about where to find their bonuses or how to work a promotion that would be expensive and inefficient for a human to answer on a case-by-case basis. It would be an "untenable cost" for the size of their brand, he said.

What gets more complicated is when people get up the ladder into tier-two and tier-three issues. When "Where is my package?" becomes, "You say my package is here and keep sending me a picture the FedEx guy snapped of the delivery, which shows ⅓ of my package is clearly missing," the robot's in a pickle. (A former coworker is in such a situation now.)

"Customer experience is so much more complicated than people realize," said Chris Filly, who heads marketing at Callvu, a customer-experience company. "The customer-service team has to deal with an infinite number of potential issues that come up across all these different touchpoints, all these different customer types. It's very, very complicated to make sure that every node in that network has perfect information from everything else."

No system, AI-driven or otherwise, is going to be perfect. But weighing on the corporate decision of what counts as "good enough" is money. Maynard, from Zendesk, spends a lot of time with chief operating officers and chief customer officers in his position, and they're under pressure to cut costs. They "know they're under the microscope," he said — some CFO reads a story about how a company cut 700 jobs using AI support agents, and they shoot over an email asking, "Why aren't we doing that?"

"We're in a macroeconomic environment where there's just much more scrutiny on costs these days for any organization," Maynard said, adding that thanks to increases in interest rates, there's a "real focus on profitability, and that puts pressure on margins."

This creates some misaligned incentives. Companies are inclined to implement AI broadly even if it's not appropriate and will make their customers miserable. They may see the immediate dollar signs they save by moving to an automated system — but they don't see the consumer on the line shouting at the AI agent and pleading to talk to a human.

"They tend to view contact centers as a cost center, not as a profit center, and the only thing you want to do in a cost center is reduce cost," said Jeff Gallino, the CEO of CallMiner, a software company that focuses on conversation intelligence and customer experience. "They're not looking for transformative, they're looking for incremental."

I recently found myself watching a panel at a conference hosted by Fortune magazine that was focused on unlocking the economic potential of AI, featuring executives at companies such as Santander and Siemens. The consensus was that AI was inevitable — bank tellers are out, robots are in, and everyone is just going to have to get used to it, including begrudging consumers who are often on the unfortunate end of it. Rodney Zemmel, a senior partner at McKinsey, said consumer acceptance is coming. "It's amazing how many people in the US were dead against any form of facial recognition until it saves them two minutes in the Delta security line in the airport," he said, or were "massive privacy advocates and for a free pizza online will give away all their personal information." As long as the benefits are there, people will come around to it.

That sounds lovely, except for a lot of consumers, the benefits aren't that evident yet, or at least not enough to outweigh the drawbacks. AI looks like just another measure companies put in place to boost their bottom lines. The bull case is that the AI gets better over time, that five years from now, the virtual agents will be lifelike enough that nobody can tell the difference, and we'll just be chatting away with robots all day to solve our problems. At the moment, companies are building the AI-enabled plane, in a sense, while flying it. Eventually, the plane will be built: The models will be trained, they'll have the right data, and there will be best practices in place for deployment.

Maynard compared the current moment to building a website in 1999 — everyone's guessing at what this is supposed to look like, but eventually, they'll figure it out. "That transition, we're just very, very early in it, and like all technology changes, it's sort of like things that you think are going to happen really fast tend just to proliferate out into the broader economy and have people adopt them and all these things, it just takes longer than anyone expects," he said.

"People are not enjoying that experience right now," Gallino said. "I very strongly believe that they will enjoy the experience probably soon."

Filly, from Callvu, said that a survey his company conducted on attitudes toward AI in customer-service settings shows consumers are coming around on it and are more willing to give it a chance. Still, they prefer to deal with a live agent in most situations.

"The honest truth is that the data is getting better, that there is hope that this will all resolve itself," he said. "We know that there are certain aspects of customer service that AI is doing well. Now, how long before the state-of-the-art AI makes it into that chatbot that's annoying the heck out of you? It might not be there yet."

The bear case is that significantly better doesn't come. There are no guarantees that this will all just work itself out. The conventional wisdom in business is that if customers have a bad experience, they'll vote with their pocketbooks and go elsewhere. But many industries are uncompetitive, and you can't easily pick up and walk away from your health insurer or your cable company. What's more, if every company has a mediocre AI experience, the bar might just be lowered across the board.

Many companies don't prioritize customer service and contact centers. They're a necessity, but the goal is to make them as cheap as possible.

"Everybody says, 'Oh, this is just going to get better naturally, and then thus conversational AI will get better naturally.' There's two huge flaws with that," Schroeder, from PolyAI, said. For one thing, Google Home and Alexa have been around for years, and they're not wizards. "Even that is, still years later, not getting the difference between 15 and 50," she said. That's a "dealbreaker" for a good conversation. "The second thing is that most of these companies are thinking about conversational AI purely as an efficiency play and as a cost savings and human replacement," she said. If the point of the AI isn't to do a good job, then why would it?

Companies' new favorite way to make — or, rather, save — money, is making consumers slightly more miserable. Hopefully, that will change, eventually. We've just got to wait and see.

Emily Stewart is a senior correspondent at Business Insider, writing about business and the economy.