AI is taking over call centres. Will that make service better?
![](https://thehalifaxtimes.com/wp-content/uploads/2023/06/ian_collins_jpg-780x470.jpg)
It should have been simple.
Ian Collins’ travel plans changed and he just wanted to update his check-in time, so he called the customer service number for the hotel to make what he assumed would be an easy fix.
“How can I help you today?” an automated voice asked him.
The voice — smooth yet stilted, clearly recorded — asked for his reservation number. He responded that he didn’t have the number handy.
“How can I help you today?” it asked him again, taking him right back to the beginning.
No matter what he said, Collins couldn’t get out of the automated voice loop.
It was the type of experience that leaves many of us thinking fondly of the days when you could quickly get a live human on the line, but digital customer service is not going away. In fact, it’s about to take over.
Thanks to the arrival of “generative” artificial intelligence (AI), a type of machine-learning technology that can create new content, such as text, images or music — or provide natural-language answers to customer service complaints — more than a third of Canadian companies are looking at how AI can improve their operations, according to a recent survey by KPMG Canada.
Over the next few years, telecoms, insurance providers, banks, utilities and government departments of all kinds are expected to lean heavily on AI technology for customer service and support.
Many consultants, corporations and researchers say the next generation of automated support will be much better than the chatbots of today. The new systems will be able to learn and adapt on the fly, instantly tapping into reams of information to deliver individually tailored responses.
But there are also concerns that if businesses rush in too quickly or implement systems poorly, the call centre experience could become worse. Customers could find themselves stuck in more frustrating automated loops, unable to reach a human, or even running into AI systems that are unintentionally offensive or succumb to “hallucinations,” where they firmly state incorrect information as fact.
Either way, millions of call centre workers will lose their jobs over the next decade or so as they’re replaced by machines — and the shorter-term scenario isn’t much better. Experts warn workers will face constant monitoring by the machines they work with, which will not only make suggestions for how to respond to queries but also report on workers’ performance, potentially making an already difficult job even more stressful.
Still, with the promise of significant savings and productivity gains — a McKinsey report just this week said improvements to customer service functions alone could create $404 billion (U.S.) in value globally — the pull of generative AI will be impossible for businesses to ignore.
For Collins, the CEO of Toronto software development company Wysdom.AI, encountering bad customer service in the wild, like he did with that hotel chain, may be frustrating, but it’s also a marketing opportunity.
His business is built on helping clients make sure their digital customer-support tools work properly and for years he’s used older versions of AI technology to do that.
If you fine tune today’s chatbots using analytics, you can keep people pretty happy, he says, but that’s nothing compared to what’s coming.
“That’s all gen one,” Collins says. “The second generation just emerged and nobody really saw it coming.”
ChatGPT enters the chat
It’s barely been half a year since the research lab-turned for-profit company OpenAI launched ChatGPT. The public-facing AI chatbot attracted unprecedented consumer adoption — hitting 100 million monthly active users within two months — and put school teachers everywhere on guard against fake essays.
ChatGPT and other tools released around the same time, such as Google’s Bard and Microsoft’s AI-powered Bing search engine, have become a pop culture sensation, spawning thousands of articles on how individuals can use the technology. A New York Times tech reporter conducted a bizarrely compelling “interview” with the Bing chatbot, and ChatGPT even showed up on an episode of “South Park.”
But generative AI has also grabbed the attention of the suits.
“There’s a significant amount of interest, across the board, in all types of industries,” says Uma Challa, senior director in the research and advisory group at management consulting company Gartner. “Most of our clients are in the discovery phase (right now).”
Collins says his own customers have also been calling with questions about how they could use this new generation of AI to change call centre and support work.
“Six months later, every one of our clients — senior level, the VPs and chief product officers — are asking us, ‘How soon can we get this stuff? What does it mean to us? How should we get ready and are our competitors doing it?’ ” he says.
“Everyone is so interested and a bit worried that it’s coming and they might be left behind.”
“AI has been around for a long time, but it’s been morphing,” says Andres Rojas, director of applied AI projects at the Vector Institute for Artificial Intelligence in Toronto.
Let me talk to a human
A 2021 online poll found Canadians reluctant to trust AI and digital assistants when seeking customer support.
74%
Say AI-based customer service (“chatbots”) have provided a worse customer experience compared with a live representative.
63%
Say they trust a live person more than AI.
6%
Say they trust a chatbot more.
63%
Percentage of Canadians who believe companies that switched to chatbots during the pandemic should return to live representatives after the pandemic subsides.
60%
Say they will view a company’s reputation more negatively if that company does not return to live agents in chatbot roles. Fewer than 10% see a positive impact on reputation.
Source: StrategyCorp
Earlier generations of the technology helped power applications like chatbots, computer programs that mimic human assistants, or voice bots, which can recognize and respond to customers’ voices over the phone instead of texting or typing online.
For customer service, that’s been helpful for the ability to quickly scan and analyze mountains of data.
But it has typically been limited to helping direct callers to the correct (human) agents, giving out information such as account balances, or delivering canned responses to simple customer queries, often from a predetermined list of answers.
“It used to be like a guided FAQ,” says Katie Bolla, a management consultant partner at KPMG Canada who specializes in customer-related services.
Generative AI is set to significantly improve the quality and personalized nature of those answers, expanding the number of customer queries that could be handled by chatbots, according to the McKinsey report.
About half of customer calls to banks, telecoms and utilities in North America are already handled by machines, the report said, adding that generative AI could “could further reduce the volume of human-serviced contacts by up to 50 per cent.”
‘Whisper bot’ on your shoulder
At this point, generative AI is mostly working behind the scenes when it comes to customer service and call centres.
Some companies are using the technology — typically not the publicly available tools, but models customized to their own needs and trained on their own corporate information and customer data — to help support their human customer service agents.
Researchers from Stanford University and the Massachusetts Institute of Technology recently published a report on what they say is the first real-world study of that kind of tool at work.
They followed the use of a generative AI-based conversational assistant by more than 5,000 customer support agents at an unnamed Fortune 500 enterprise software company.
The AI listened in on the agents’ conversations with customers — small- and medium-sized business owners who needed to troubleshoot problems with their software — and provided suggestions on how to respond.
Rojas calls this type of technology a “whisper bot.”
In an example of how that could work in practice, the Stanford and MIT researchers suggested a scenario where a customer calls in because they can’t log into their system. The AI quickly identifies why that might be and offers a common solution, suggesting that the agent ask, “Can you check that your caps lock key is not on?”
The AI is also trained to take the customer’s feelings into account and suggest language that can help with that, something like, “That wasn’t stupid of you at all! I always forget to check that too!”
The researchers found that by using the AI assistant, productivity increased by 14 per cent (that measured the average length of calls, whether issues were resolved and customer satisfaction).
The AI also helped newer, low-skilled agents do their jobs better, by sharing knowledge and suggestions it might otherwise have taken them weeks or months to pick up.
‘Not quite ready for prime time’
A key point made in the case study at the software company is that the agents did not have to take the AI’s suggestions and still had the power to use their own, human judgment.
But as the technology progresses, computers could soon be handling everything.
Some technology vendors are already creating virtual assistants capable of answering calls on behalf of an agent, resolving issues in some cases and speaking like a human in response, Challa says.
“We haven’t really seen that product in action yet,” Challa says. “But that could be in the near future when it becomes more and more common.”
Collins thinks it will be about six months or so before AI-based assistants are unleashed to deal directly with customers.
Right now, he says, they’re “not quite ready for prime time in the in the big enterprise market,” in part because companies want to be sure of “every word this thing’s going to say to my customers.”
AI is still unpredictable and prone to “hallucinations” — delivering false information with convincing confidence — or unintentionally making offensive comments.
There are other barriers too.
Getting generative AI right can be expensive and time-consuming. The MIT and Stanford researchers say that training the program on a company’s own data “requires thousands of GPUs and weeks to months of dedicated training time.” (GPUs are graphics processing units, the computer chips used in generative AI processing.)
“You need good routing capabilities. If the model doesn’t understand the specific type of question, how seamlessly does it transfer to a human agent?” says Gartner’s Challa. “Some of those issues have to be worked out.”
KPMG’s Bolla agrees that it’s crucial to get the right “plumbing” in place (such as data centres to manage customer data and meticulous mechanisms for routing data).
Without that, AI won’t help prevent that frustrating experience of getting bounced from question to question and department to department without getting your problem solved.
And customer expectations are constantly increasing, Bolla says, noting that businesses can pay a heavy price for disappointing people. “It typically costs more to acquire a customer than it does to retain a customer.”
AI use in call centres to surge
As companies look to cut customer support costs, Gartner predicts that by 2026, businesses around the world could save $80 billion (U.S.) on labour. Meanwhile, investment in conversational AI for customer support is set to grow.
1 in 10
Agent interactions that will be automated using AI by 2026, up from an estimated 1.6% of interactions today.
Source: Gartner
The Vector Institute’s Rojas says organizations need to balance that with the short-term savings they could realize by shifting to an AI-based customer support system that doesn’t satisfy customers.
There are also questions around how comfortable people are with AI, he says. “It’s not a technical issue, it’s more of a social norm.”
In 2021, an online survey of 400 Canadians conducted by strategic advisory firm StrategyCorp found almost three quarters of respondents said they received worse customer service from an AI-based chatbot compared to a real person. Sixty-three per cent said they trust live agents more than AI.
Still, says Rojas, as the technology improves, consumers are likely to see the advantages of getting an answer from a computer in 30 seconds with zero wait time.
“I think society will also evolve and people will get used to it.”
Worker surveillance and job losses
Once the technical snags are ironed out there are still some thorny ethical issues to consider.
AI tools designed to assist agents won’t just help them solve customer problems, they’ll keep a close watch on everything they do.
Customer service employees have long been subject to call monitoring and random performance checks, but some warn the new AI systems have the potential to take monitoring to a whole new level.
“Worker surveillance is intensified. Every interaction is very closely scrutinized by computers, which is something that no human manager could do,” says Valerio De Stefano, the Canada research chair in innovation law and society at Osgoode Hall Law School.
De Stefano worries there will be a spike in workplace discipline cases related to poor performance captured by AI.
“Workers are going to bear the brunt of this for some time unless something in the legal landscape changes,” he says.
He added that while unions are beginning to address surveillance issues in new collective agreements, there is limited unionization in the private sector, which employs many call centre workers.
Call centre jobs are hugely stressful — a 2020 report by Cornell University’s Virginia Doellgast and McMaster University’s Sean O’Brady found a large number of call centre workers experience emotional strain, sleep difficulties, use of anxiety medication and repetitive stress injuries.
Arif Jetha, a scientist at the Institute for Work & Health and associate professor at the University of Toronto’s Dalla Lana School of Public Health, says AI technology could help take some of the stress out of the job by taking over simple, repetitive tasks and properly routing calls to the right employee.
“But it also has the potential to displace workers,” he says. Call centre workers already tend to come from vulnerable backgrounds, he says, adding that the people most likely to lose their jobs as AI takes over are workers of colour, recent immigrants, women or those from an intersection of those backgrounds.
Virtual reality is also likely to play a part in what the customer support agents of the future look like, says KPMG’s Bolla (though this is much further down the road).
“You could (have a conversation with) a visual manifestation of an avatar, that you feel is most accessible and easy to talk to, with the language or the accents that you find most comfortable speaking in.”
There are about 17 million customer support agents worldwide, according to an estimate from U.S. management consultancy Gartner, which predicts that by 2026, 10 per cent of all agent interactions will be automated using AI, up from an estimated 1.6 per cent today.
About 16 million of those jobs are going to disappear over the next decade or so, Collins figures.
Joel Blit, an associate professor of economics at the University of Waterloo with a focus on innovation policy, similarly predicts humans will eventually be left handling only about 10 per cent of customer support issues.
“Once these automated systems are good enough, the guardrails are in place and they’re not hallucinating or saying unsavoury things, they’ll probably be able to resolve my issues a lot faster than humans, so I will be very happy to just deal with AI,” he says.
That isn’t necessarily a bad thing, in Blit’s view.
“AI is replacing a fairly monotonous job … As long as we can redeploy those individuals in other areas of the economy, it could be a win,” he says, adding, “We need a social safety net and retraining programs.”
Ultimately, businesses will adopt the technology when customers are comfortable with it and it makes financial sense to do so, says the Vector Institute’s Rojas, adding that computers will always have an edge on cost.
“Whenever you look at automation, the equation is always stacked against humans.”