The Human Touch in a Digital Age: How Generative AI Can Make Customer Service More Humane

It's a near-universal experience: the gnawing frustration of a customer service interaction that is inflexible, inefficient, and strikingly impersonal. For years, the pursuit of cost-effectiveness in customer service has led to automated systems that, while reducing labor expenses, often leave customers feeling unheard and undervalued.

With the arrival of Generative AI (GenAI), the next wave of technological innovation in customer service stands at a crossroads. While it threatens to extend the ruthless cost-driven experiences of the past, if done well, Generative AI also presents a paradoxical opportunity: to make customer service decidedly more attentive, transparent, and respectful of customer time and effort, or in a single word, more humane.

Cost, Automation, and Dissatisfaction
The economics of customer service have long been a tightrope walk for businesses. Staffing phone lines and support desks with human agents is expensive, and unless stringently controlled, these costs can significantly outweigh the perceived benefit to the business.

Consequently, service provision is often restricted, from limited hours to intentionally complex and opaque phone menus, leading to widespread customer dissatisfaction.

In this environment, early automation efforts, emerging in the 1990s, were overwhelmingly geared towards reducing operational expenses rather than genuinely elevating the customer experience. These initial systems often presented a poor return on their high set-up costs, hampered by a lack of flexibility and an inability to adapt to new contexts; they delivered shallow or rigid functionalities, such as rudimentary call routing, and at their (regrettably common) worst, trapped users in frustrating phone menu labyrinths.

Even as later iterations in the 2010s began incorporating elements of early deep learning to automate specific narrow tasks, or to assist human agents with real-time data, the primary objective remained largely unaltered: cost efficiency over empathetic customer engagement. Against this backdrop, the latest generation of AI changes what’s economically and experientially possible.

Why This Time Is Different
Generative AI presents the opportunity for a significant departure from these earlier limitations. The underlying Large Language Models (LLMs) are rapidly improving in affordability, speed, and capability. We've now reached a point where their emulation of human interaction is so convincing that LLMs even pass demanding versions of the Turing test—a milestone for machine intelligence—leaving human interlocutors unable to distinguish AI from human.

This evolution is not just about conversational plausibility; GenAI now demonstrates sophisticated understanding and even empathy (or at least the appearance thereof). For instance, in the demanding field of medical diagnostics, systems like Google's AMIE have already shown the ability to outperform human doctors in both diagnostic accuracy and empathetic communication.

Coupled with their capacity to rapidly process extensive information via large context windows, these foundational AI capabilities are poised to redefine (and greatly expand the scope of) human-machine interactions.

This technological leap is further complemented by advancements in low-latency text-to-speech (TTS), automatic speech recognition (ASR), and natively multimodal models that collectively allow synchronous, open-ended voice chat1. Coupled with 24/7 availability and easy scalability, it’s clear that GenAI offers a powerful toolkit to reshape customer interactions.

For human customer service agents, GenAI means less repetition, as routine tasks and interactions become increasingly automated. Crucially, empirical studies are beginning to reveal not just efficiency gains but qualitative shifts for workers.

An early, notable study of GenAI adoption in a call center found that, with what would now be considered a significantly dated LLM (a GPT-3 finetune), human call center agents showed an average 15% productivity improvement when augmented with GenAI.

More notably, GenAI resulted in a significantly improved experience for both customers and agents: customers became more polite and less frequently escalated issues to managers. This points toward a future where GenAI doesn't just improve efficiency, but can actually enhance the customer service experience on both sides of the interaction.

However, this technological shift brings a "digital dilemma". Credible forecasts point to a decrease in the overall number of customer service representative jobs: the BLS projects a 5% decrease in the roughly 3 million US customer service rep jobs between 2023 and 2033, while Forrester predicts that in 2025 alone, GenAI will displace 100,000 frontline agents from top global outsourcers.

The nature of the remaining jobs will inevitably change too. They may involve more interpersonally challenging work, requiring a different skill profile and perhaps a greater emphasis of the remaining roles on the handling of complex cases and AI oversight. These shifts preview the risks and design choices leaders must confront.

Risks to Manage—Deliberately
The opportunity is real, but so are the new risks and failure modes.

Klarna's experience offers a prominent illustration. The company's ambitious automation drive, announced in early 2024, featured an AI assistant reportedly handling the workload of 700 full-time human agents. Initial metrics were indeed impressive: 2.3 million conversations managed in the first month with claims of customer satisfaction (CSAT) on par with human agents, and significantly reduced resolution times.

This aggressive strategy, coupled with a widespread hiring freeze and CEO Sebastian Siemiatkowski’s pronouncements on AI’s expansive potential, initially signaled remarkable efficiency gains.

However, customer dissatisfaction with the AI assistant soon surfaced—challenges later acknowledged by Siemiatkowski who admitted that an aggressive focus on AI-driven cost-cutting had resulted in "lower quality" service. Following the CEO’s public "epiphany" in early 2025 about the enduring value of human interaction, Klarna announced the resumption of hiring human customer service agents.

This isn’t an isolated incident. The Commonwealth Bank of Australia’s chatbot-first shift increased call volumes and frustration, prompting a pullback, and demonstrating that over-zealous automation can raise, not reduce, demand (not to mention reduce customer satisfaction). And in an early and well-publicized incident, Air Canada’s chatbot hallucinated incorrect policy information for which the company was held liable.

The lesson is clear: when AI speaks for you, you own the outcomes, both good and bad. Even with Turing test‑level performance, there is no "just add AI" solution for customer service; durable results will only come from a thoughtfully designed system, disciplined oversight, and clear standards you monitor and uphold.

The best outcomes will be realized when you automate only what you can deliver without friction, and proactively escalate to human agents before failure results in frustration. Anything else reads as cost-cutting gatekeeping.

Transforming the Interaction
So what does a humane design look like?

The potential of Generative AI in customer service extends far beyond merely automating, expediting or achieving marginal improvements in existing, unsatisfactory processes; it offers an opportunity to entirely redefine customer interactions.

Where traditional approaches, rigidly constrained by operational costs, struggled to deliver consistently attentive service, GenAI stands to reshape these interactions to become not only more efficient, but more flexible, responsive, and deeply personalized.

Customers will increasingly come to expect as much: frictionless, on-demand interactions and self-service options (no more waiting on hold, no more having to repeat the same information to multiple agents); more readily available information and more proactive communication presented in a way that caters to their specific preferences, circumstances and interaction history.

Given the ongoing decline in model inference costs, businesses can design for capability and customer benefit first, then optimize for efficiency—an inversion of the old playbook.

Realizing this potential, however, is not a simple matter, and it will not happen overnight; it requires navigating significant technical and social complexities.

Businesses will likely have to revisit build-versus-buy decisions, and will need to confront varied challenges ranging from siloed data and privacy protections to retraining frontline staff for heavily AI-augmented roles.

Businesses will likely also see shifts in organizational structure, such as a deeper integration of customer service with business functions like sales and user experience (UX).

The payoff for businesses that get GenAI adoption for customer service right is a distinct competitive advantage: a more humane customer experience that exceeds evolving customer expectations, and produces a more empowered and efficient customer service workforce.

Unlocking this advantage, however, goes beyond the simple decision to adopt GenAI. It demands a deliberate exploration of a greatly expanded universe of possible interaction patterns that is deeply informed by an understanding of the core purpose of the business, its unique data and capabilities—or indeed, limitations—and most importantly, the fundamental needs of its customers.

Done well, GenAI will make customer service more humane: demanding less customer effort while delivering higher quality service with greater efficiency.