AI Changes Customer Service, but Harder Work Moves to Humans

More than half of surveyed U.K. contact-center decision-makers said agent workloads had increased, according to MaxContact's 2025/26 U.K. Contact Centre KPI Benchmarking Insights Report. The share was around 40% in 2024; KPI means key performance indicator.

The survey does not establish what drove the increase, so it cannot show that artificial intelligence (AI) caused the rise. That distinction matters as operators deploy AI to handle routine requests while keeping people accountable for exceptions and system failures. MaxContact's result supports a workload warning, not a verdict on AI.

What changes inside a customer-service team?

Kaylee Harris, who works in customer service for a credit card company, told CNBC that her hours used to include calls, note-taking and lengthy card disclosures. With AI, she says, she can focus more on each customer, although occasional system glitches still create extra work.

Shannon, who works for a telecommunications company in Cardiff, Wales, offered a less favorable account. She withheld her surname because she feared identification by her employer and said AI-generated notes at her workplace are often riddled with errors. Her workaround is blunt: “so I just write my own.”

At her previous employer, a property services company, senior management told staff to “ChatGPT it” when they did not know an answer. AI agents, meaning automated software that handles customer requests, booked appointments but passed cases to a human when they could not offer the requested date.

Staff were “constantly fighting fires,” Shannon said. “All the complicated cases came to us, causing a lot of burnout.” For workers, automation can move the hardest cases up the queue instead of eliminating them.

What do vendors and operators expect?

In theory, customer-service automation should move administrative work away from people and give them more time for problems requiring judgment, empathy and experience. That is the case made by Nikola Mrkšić, CEO and co-founder of London-headquartered PolyAI, which develops voice AI systems, meaning software that automates spoken customer-service conversations.

Mrkšić said contact-center agents can face “maximum abuse for minimum wage.” They handle frustrated customers while working under strict performance metrics, or measures used to judge their work, sometimes without the authority or information needed to resolve a problem. He sees AI agents as a way to shoulder some of that load.

“It's a really, really difficult job,” he said. “I think with AI, we have a shot at bridging the gap to make it better for both [customers and employees].”

British telecom and media company Virgin Media O2 launched an AI voice agent in September. The company said the agent handles routine issues so human advisors can spend more time on complex or sensitive cases; CNBC's account provides no measured results for that deployment.

At U.K.-based customer experience company ArvatoConnect, CEO Debra Maxwell told CNBC the company aims for 70% of interactions to be handled by technology and 30% by people. This is an operating target, not a disclosed outcome, and CNBC's account gives no company-wide performance figure.

Maxwell also described an ArvatoConnect tool developed for the U.K. government's procurement agency, the Crown Commercial Service. It generates draft supplier viability assessments with 97% accuracy, she said, reducing the need for hours of manual data extraction, or copying information from records for review. CNBC's account does not state the test method or sample size.

Across these examples, the promised gains center on handling volume and efficiency. CNBC's account reports a target and a company claim, not independent results for either deployment.

Can automation make the remaining work harder?

Amanda Jones, who researches organizational behavior, meaning how work affects people and groups, and the future of work at King's College London, specializes in work intensification, the process by which jobs become more demanding over time. She told CNBC that AI could intensify call-center work if human agents receive a growing share of the toughest interactions.

“Some customers may already be frustrated by the time they reach that human,” she said. “One thing we sometimes overlook is that relatively straightforward tasks can provide a bit of psychological respite between more demanding ones.”

“If those disappear, the intensity of the working day could increase considerably, even if the overall volume of work doesn't,” Jones said. Her warning applies to workload even when total contacts do not rise.

Jones also cautioned that productivity gains may bring more intense performance targets. As routine tasks move to software, workers could lose opportunities to develop the experience needed for harder cases.

“So, we could end up expecting people to perform more demanding work while simultaneously reducing the experiences that traditionally helped prepare them for it,” she said. Workers could face greater demands without receiving the preparation that previously made those demands manageable.

More work for people does not automatically mean better work, because the remaining cases can carry more emotional strain and fewer opportunities to build skill. Jones's account therefore supports that caution.

Should handling-time targets change?

ArvatoConnect CEO Debra Maxwell argued that average handling times, meaning the time taken to resolve an interaction, may need to increase as employees take on more complex cases. Her alternative measure focuses on outcome quality rather than raw speed.

“We assess whether [human] agents understood the customer's needs, exercised sound judgement and reached the right resolution,” she said. “Not whether they followed a script or hit an arbitrary time target.” For contact centers, Maxwell's approach treats a time-only metric as inadequate when cases require judgment.

What does automation do to customer experience?

Kaylee Harris sees benefits in AI tools despite occasional glitches, but she worries that the push to automate will come at the expense of customer experience. Her criticism targets both speed and the spread of AI.

“As a country, I think we are focusing on getting things done faster and choosing quantity over quality … resulting in customers being fed up with the way AI is being pushed on everyone,” she said. Her complaint concerns a national emphasis on throughput rather than service quality.

Shannon also sees potential benefits in AI but does not think it should ever replace real customer service. Her concern is not task completion alone, but whether a customer can experience a genuine human relationship.

“I believe people buy from people; with AI you don't get to experience that,” she said. “I go the extra mile. Can AI give that feeling? Can it feel genuine?”

Neither Harris nor Shannon supplies a controlled comparison of customer outcomes, so neither account settles whether AI improves service. The practical test for further deployments is whether they reduce routine handling without transferring enough complexity, burnout and error to customers or the agents left behind.