Last week, a flashy McKinsey report on customer service AI landed in my inbox, confidently predicting the imminent takeover of contact centres by automation. It reminded me of dozens of similar customer service AI reports I’ve seen over three decades in digital transformation.

Each new customer service AI wave – from IVR systems to chatbots to today’s generative customer service AI – arrives with the same promise: dramatic cost savings through human replacement.

Yet buried in the McKinsey customer service AI report was a striking contradiction that nobody seemed to notice: despite decades of digitisation efforts and customer service AI investments, human-to-human customer service interactions have actually grown by 2% annually since 2010.

This isn’t how the customer service AI story was supposed to unfold. Why aren’t humans being replaced?

The digital-by-default delusion: lessons from Universal Credit

In 2012, I was part of the team tasked with creating Universal Credit’s digital service, essentially an early customer service AI solution before we used that terminology. Our explicit brief was to design a customer service AI system so intuitive and comprehensive that nobody would need human support through phone calls or job centre visits.

We threw everything at this customer service AI challenge, including integrated help text, progressive disclosure, plain language explanations and extensively tested user journeys. For its time, it was truly groundbreaking for GOV.UK standards.

Multiple ministers proudly announced how this customer service AI approach to digital transformation would dramatically reduce staffing costs. That customer service AI ambition had to be abandoned almost immediately.

Despite our best customer service AI design efforts and user research, the human support need remained stubbornly persistent. People still called. They still visited job centres. They still needed reassurance from humans about life-changing financial decisions, things our customer service AI systems simply couldn’t provide.

And this wasn’t a customer service AI design failure. It revealed something fundamental about human behaviour that efficiency-focused customer service AI strategies consistently misunderstand.

The gap between customer service AI promise and human reality

The pattern I’ve witnessed repeatedly shows customer service AI following a predictable cycle:

1. New customer service AI technology emerges with promises of dramatic efficiency gains
2. Companies invest heavily in customer service AI self-service
3. Human contact fails to decline as customer service AI predicted (and sometimes increases)
4. Executives blame customer service AI implementation rather than questioning the premise

This cycle repeats with each new customer service AI technology. But what if the customer service AI premise itself is flawed?

McKinsey’s own customer service AI data shows this disconnect clearly: digital interactions have grown by 6% annually since 2010, but human-to-human interactions have still grown by 2%. The promised replacement by customer service AI isn’t happening.

The social function beyond customer service AI transactions

The Dutch supermarket chain Jumbo offers a fascinating counterpoint to the customer service AI efficiency-above-all narrative. In 2019, they introduced “kletskassas” (chat checkouts) specifically designed for elderly customers and others who value conversation during their shopping experience.

These lanes operate counter to every customer service AI principle. They’re deliberately inefficient. They prioritise human connection over transaction speed, the exact opposite of what customer service AI optimisation aims to achieve. And they’ve proven enormously popular.

This acknowledges something most customer service AI transformation strategies miss entirely: many service interactions serve a social function beyond the transaction itself. People seek validation, reassurance, and simple human connection that even the most advanced customer service AI struggles to provide authentically.

When we reduce customer service to its purely transactional elements suitable for customer service AI, we fundamentally misunderstand what many people are seeking from these interactions.

The ethical quicksand of simulated customer service AI humanity

The uncomfortable question this raises for customer service AI: if people want human connection, should we train customer service AI to simulate it more convincingly?

Several large customer service AI companies are already working on making their voice assistants more “empathetic” and “relational”, essentially creating the illusion of human connection where none exists.

This crosses into deeply problematic customer service AI ethics territory:

1. Inherent deception: Creating customer service AI that makes people feel they’re connecting with something that understands their emotions when they’re not is fundamentally dishonest.

2. Exploiting vulnerability: It’s often the most isolated and vulnerable who most seek human connection in service interactions. Creating customer service AI that targets this need feels potentially exploitative.

3. Eroding social capacity: If we replace genuine human connection with customer service AI simulation at scale, what social skills and capacities might we be losing collectively?

There’s something dystopian about the logical customer service AI conclusion: “Research shows customers want human connection, so let’s create customer service AI that tricks them into thinking they have it.”

Reimagining the human-customer service AI relationship

Rather than persisting with the replacement narrative, a more honest approach to customer service AI would acknowledge that different types of interactions call for different solutions.

The most successful customer service AI implementations I’ve seen position AI as augmenting human capabilities rather than replacing them:

• Using customer service AI to handle truly transactional elements, freeing humans for more complex or emotional interactions
• Recognising that some customers will always prefer human interaction for certain services, regardless of customer service AI advances
• Measuring customer service AI success beyond simple efficiency metrics to include customer satisfaction and resolution quality
• Being transparent about when customers are interacting with customer service AI versus humans

This requires a more nuanced understanding of what drives customer service interactions beyond mere problem resolution, something customer service AI alone cannot address.

The future of customer service AI isn’t replacement – it’s recalibration

The persistent growth in human customer service interactions despite customer service AI advancement suggests something fundamental: customer service AI doesn’t simply replace human needs, it transforms them.

As customer service AI systems handle simpler issues, human agents increasingly deal with more complex problems that customer service AI can’t solve. This requires higher skills and better training, the opposite of the deskilling often predicted with customer service AI implementation.

Rather than preparing for the mass replacement of customer service workers by customer service AI, organisations would be better served focusing on how customer service AI and humans can work together effectively, leveraging the unique capabilities of each.

Because if three decades of digital transformation have taught me anything about customer service AI, it’s this: the predicted death of human customer service has been greatly exaggerated. The future won’t be customer service AI or humans; it will be finding the right balance between them.

And despite what the customer service AI consultants keep promising, that balance will likely involve more humans than their PowerPoint slides suggest.