Human + AI: The Winning Contact Center Model

Ask 10 contact center executives what AI means for their operation, and many will quickly ask the same question: How much operating expense can we eliminate—or, more specifically, how many agents can we replace?

I believe that's the wrong question. Not because AI won't reduce the number of people required to operate a contact center—in many environments, it absolutely will—but because it frames AI primarily as a headcount strategy when what we're really experiencing is a fundamental redesign of the contact center operating model.

The better question is: Which customer interactions should never require a human in the first place?

Password resets, order status, simple account inquiries, appointment changes, and basic troubleshooting are predictable, transactional contacts that often never require human judgment. They only required humans because, until recently, there wasn't a viable alternative. AI is changing that by allowing more of these interactions to be resolved quickly and consistently without human interaction.

Bank of America's virtual assistant Erica is a good illustration of the scale involved. Since launching in 2018, Erica has handled more than 3 billion client interactions and now fields roughly 58 million a month, including spending analysis, balance alerts, appointment scheduling, and investment guidance. Approximately 98% of users find what they need without ever reaching a live agent. That's no longer a chatbot experiment; it's a core service channel. Bank of America has also been clear about the purpose: freeing financial specialists to spend more time on complex financial conversations with clients.

That's the pattern I expect to continue. Predictable volume moves to AI, while human capacity gets redirected toward more complex interactions. What I find particularly interesting—and what I don't believe gets enough attention—is what happens after those easier interactions disappear. The work left for human agents gets harder, and that has enormous implications for how companies think about their people, technology, outsourcing partners, delivery locations, training, and ultimately the economics of the contact center.

The Human Agent Becomes More Valuable, Not Less

As AI absorbs more predictable and repetitive interactions, the agents who remain will spend less of their time answering simple questions and more of their time handling the interactions AI cannot resolve. Those include complex problems and exceptions, escalations, emotionally charged situations, retention and loyalty conversations, high-value customers, complicated troubleshooting, sales opportunities, and situations requiring judgment, empathy, or negotiation.

That changes the workforce equation considerably. Companies may need fewer agents, but the people they retain will need to be more capable, better trained, and potentially better compensated because the work they are being asked to perform will be more difficult and more consequential.

Klarna's experience provides an interesting cautionary example. In 2024, the company built an AI assistant that was reportedly doing the equivalent work of roughly 700 full-time agents and handling about 75% of customer service chats. For a period of time, it appeared to demonstrate how far an AI-led service model could go. In 2025, however, Klarna began hiring humans again. CEO Sebastian Siemiatkowski acknowledged that while AI delivered cost advantages, quality suffered in more difficult conversations, and customers still wanted the ability to reach a person. Klarna didn't prove that AI doesn't work. It demonstrated the limitations of an AI-only model, which is an important distinction.

This is also where outsourcing strategy becomes very interesting. For years, one of the fundamental questions in outsourcing has been: Where can I find 500 qualified agents at the right cost? Increasingly, I believe that question will become: Where can I find 250 highly capable people who can successfully handle the interactions my AI cannot?

That's a significant shift because it increases the importance of English proficiency, communication skills, education, cultural alignment, recruiting profiles, training investment, leadership quality, and the overall depth of the labor market. It also challenges a location strategy built primarily around low labor costs and large quantities of available agents. As the complexity of human-assisted interactions increases, the cheapest location may no longer produce the lowest cost to serve—or, more importantly, the best business outcome.

AI Shouldn't Just Face the Customer

Much of today's AI discussion centers on customer-facing technology such as chatbots, voice bots, virtual agents, and self-service. Those capabilities are important, but I believe some of the biggest opportunities are coming from AI that surrounds and supports the human agent rather than simply trying to replace them.

Consider what AI can already do throughout an interaction. Agent Assist can provide real-time answers and recommendations, while AI-powered knowledge management can surface the right information without forcing an agent to search multiple systems. Automated quality management can evaluate nearly every interaction rather than relying on the traditional small sample, and AI can identify coaching opportunities based on actual performance patterns.

Octopus Energy's UK operation is a good example of this in production. Its generative AI tool, Magic Ink, built by its technology arm Kraken, doesn't talk directly to customers. Instead, it drafts responses and next-step recommendations for human agents to review while also summarizing the conversation history behind each case. The tool has summarized more than 6 million customer calls, and roughly a third of customer emails are sent with its assistance. Those AI-assisted emails have also generated higher customer satisfaction scores than emails written without AI assistance.

That's an important distinction. The human stays in the loop while AI removes friction from the interaction.

AI can also summarize conversations, automate dispositions, identify sentiment, provide real-time compliance prompts, and improve/recommend the next-best action while the customer is still engaged. None of those capabilities require the customer to ever interact with a bot, yet collectively they can make the human agent faster, more knowledgeable, more consistent, and more confident.

Don't just automate the customer experience with AI. Instead, augment the employee experience.

Research supports this approach. Economists Erik Brynjolfsson, Danielle Li, and Lindsey Raymond studied approximately 5,000 customer support agents and found that agents using an AI assistant resolved 13.8% more inquiries per hour, with a modest improvement in quality. The more interesting finding was who benefited most. Lower-performing agents improved significantly more than experienced, higher-performing agents, and new employees using the technology reached proficiency benchmarks much faster.

That's not simply a story about AI replacing agents. It's a story about AI reducing the time and variability required to develop a capable agent, which has direct implications for how quickly newly recruited teams—whether onshore, nearshore, or offshore—can become proficient.

From the customer's perspective, all of these capabilities are connected. The winning model looks less like "a customer talks to a bot instead of a person"  and more like an integrated ecosystem:

Customer → AI → Human → AI-Assisted Human → Analytics → Continuous Improvement

AI can operate across the entire customer interaction lifecycle. Before an interaction, it can support intent recognition, authentication, intelligent routing, and self-service. During the interaction, it can provide knowledge, recommendations, compliance guidance, and Agent Assist. Afterward, it can handle summarization, quality management, coaching, analytics, and sentiment identification. That is a much broader transformation than simply adding a chatbot to a website.

The Economics of Outsourcing Are Going to Change

This may be one of the most significant implications for the BPO industry. For decades, a substantial part of the outsourcing value proposition has been labor arbitrage: accessing large pools of qualified talent at a lower cost. AI doesn't eliminate that advantage, but it changes the equation considerably because technology is increasingly capable of absorbing much of the transactional work that made large-scale labor arbitrage so attractive.

The sourcing decision increasingly looks something like:

Labor Cost + Talent Quality + AI Capability + Technology + Automation + Governance + Business Outcomes

That's much more complicated than simply comparing hourly rates or cost per FTE, and it raises some important questions about how outsourcing relationships will be structured in the future.

Does traditional per-FTE pricing continue to make sense when an increasing percentage of customer interactions are resolved without an FTE? Should companies eventually pay providers based on transactions or outcomes rather than seats staffed? When AI reduces average handle time, improves first-contact resolution, or eliminates contacts entirely, how should those productivity gains be shared between the client and provider?

Here's another question that I believe will become increasingly important: Does a $16-per-hour delivery location still outperform a $22-per-hour location if the higher-cost workforce is substantially better at resolving the complex, high-value interactions that remain?

These aren't just hypothetical questions. Teleperformance, one of the world's largest contact center operators, has publicly committed to what it calls a "Human + AI" strategy, equipping its workforce with generative AI knowledge assistants, real-time call guidance, and AI-assisted routing. At the same time, Gartner has projected significant reductions in contact center labor costs from conversational AI and expects agentic AI to autonomously resolve a growing percentage of common customer service issues over the next several years.

Those trends will inevitably affect sourcing decisions. Markets built primarily around large volumes of relatively simple voice work could face significant pressure as AI adoption accelerates. Conversely, locations and providers capable of delivering highly skilled, AI-enabled agents may become considerably more valuable.

The winners won't necessarily be the lowest-cost providers. I believe they'll increasingly be the providers that combine great talent, effective technology, intelligent automation, strong operational leadership, and measurable business outcomes.

The Winning Model Is Orchestration

Ultimately, this isn't a debate about AI versus humans. The future contact center is an operating model where AI handles more of the predictable, repetitive, and transactional work while humans increasingly own the complex, emotional, judgment-based, and revenue-impacting moments that matter most.

The organizations that win won't necessarily have the most AI. They'll be the ones that understand when to automate, when to augment, when to humanize, and when to eliminate.

Automate: High-volume, predictable, low-emotion, rules-based interactions where AI can increasingly own the customer journey from beginning to end.

Augment: Complex and knowledge-intensive work that still requires human judgment. In these situations, AI should work beside the employee, providing better information, guidance, and tools that allow the individual to perform at a higher level.

Humanize: High-emotion, high-value, retention-sensitive, or relationship-driven moments where human judgment, empathy, and connection can materially affect the outcome.

Eliminate: Perhaps the most overlooked category. Some contacts shouldn't be automated or handled more efficiently; they shouldn't happen at all. If customers repeatedly contact a company because of a broken process, confusing policy, recurring billing issue, or poorly designed experience, the goal shouldn't be to build a better bot to handle the problem. The goal should be to eliminate the reason customers need help in the first place.

Air Canada learned an important lesson about these categories when a grieving customer asked the airline's website chatbot about bereavement fares. The chatbot provided incorrect information about the airline's policy, and when the customer later tried to claim the discount, the dispute eventually went before a Canadian tribunal, which held Air Canada responsible for the information provided by its chatbot.

I would argue that this was fundamentally a “humanize” moment: a customer dealing with the death of a family member and was treated as an “automate” moment. It demonstrates why getting these decisions right isn't simply about efficiency. It's also about customer experience, judgment, and risk.

Conclusion

I believe the contact center of the future will employ fewer people, but the people who remain will matter more. As AI absorbs more simple and transactional work, the interactions left for humans will increasingly require judgment, communication, empathy, problem-solving, sales ability, and cultural fluency. Those capabilities aren't commodities, and that means AI may actually increase the importance of talent quality in outsourcing decisions rather than decrease it.

The next generation of outsourcing decisions may therefore look very different from those made during the previous 20 years. Companies will need to look beyond hourly rates and available seat capacity and ask harder questions about talent, technology, AI readiness, leadership, and the business outcomes a provider and delivery location can actually produce.

Organizations treating AI primarily as a headcount-reduction exercise will undoubtedly find opportunities to lower costs, but I believe they will miss the larger opportunity. The real transformation comes from redesigning the contact center around what technology does best and what people do best.

Let AI handle the interactions that never needed a person, while giving employees AI tools that make them significantly better at their jobs. Then put highly capable people in the moments where judgment, empathy, relationships, and revenue truly matter.

That's not simply a human contact center or an AI contact center. It's a Human + AI contact center—and I believe that's the model that's going to win.