A customer calls your contact center after working through your app's issue resolution process to fix a missing item from a $50 order. She has been a loyal, enthusiastic user until recently, spending around $2,500 a year, with five years of consistent history and almost no use of a competitor. This is exactly the customer you want to keep. Your AI processed her case correctly and followed the right steps, but her situation had a variation that required a human. Your systems know she contacted support twice last week, that her most recent satisfaction score is the lowest of her entire relationship, and that customers with similar patterns who leave rarely come back. She is transferred to a queue and told the wait will be at least fifteen minutes due to high call volume. She waits. She speaks to an agent who has been running at 94% occupancy for their entire shift and has three minutes left before the end of their hour. She gets an apology and a credit. The next time she needs a delivery, she opens a competitor's app.
The failure here is not the AI. The failure happened further upstream and is a strategic one. The company automated the wrong work and never asked a more important question: what should we do with the human capacity that automation releases?
AI will not only reduce routine work. It will create a capacity dividend: human time, attention, and judgment that most companies are about to take as margin without realizing what they are giving up.
Part I - The Question Most Leaders Are Getting Wrong
The first generation of AI transformation in customer care has been dominated by an efficiency narrative. How many contacts can be deflected? How much after-call work can be eliminated? How quickly can average handle time fall? How many agents can we eliminate? These are reasonable questions for a cost-intensive function. They are also incomplete ones. I have also been in a room with a leader who saw AI as the opportunity to move from 20K agents to 200, with only cost reduction as the celebration goal.
Gartner predicts that by 2029, agentic AI will autonomously resolve 80% of common customer service issues and contribute to a 30% reduction in operational costs. McKinsey estimates that generative AI in customer care could create productivity value equivalent to 30 to 45% of current function costs. These are material numbers. But a more important strategic question is not what cost can be removed. It is what new forms of customer value become economically possible when routine demand is absorbed by AI.
I have spent 25 years leading global support operations at companies including Capital One, Chime Financial, Uber, and DoorDash, organizations with millions of customers, even more millions of interactions, complex contact center networks, and P&L accountability for service delivery. In every one of those environments, I watched smart leaders optimize relentlessly for cost. And in every one, I watched the same backlog accumulate: dissatisfied customers who were noted but not recovered, repeat-contact patterns that were reported but not investigated, trust that was damaged in plain sight and never repaired. The old operating model simply could not afford to do anything about it. There was no bad intent here, just actual team capacity.
AI changes the economics, but only if leaders make a deliberate choice about what to do with the capacity it releases.
What the AI Capacity Dividend Actually Is
The AI Capacity Dividend is the usable human capacity created when AI reduces the volume, duration, administrative load, or complexity of routine service work. It may appear as fewer contacts, shorter interactions, lower documentation effort, faster knowledge retrieval, or less time spent on repetitive policy explanation. However it appears, it represents a strategic choice.
Important to remember that a dividend is not a guarantee of value. It is a strategic resource-allocation decision. Leaders can bank it as margin gain. They can spend it on technology infrastructure. Or, how about they reinvest part of it in higher-value human work that the old cost structure could never support. They could design new process that can add tremendous value to both the company and the customer? What is more likely and what I already see happening in organizations today is they are allowing the entire dividend to disappear into celebrated (yet undifferentiated) cost reduction before anyone has experimented with what a different allocation might produce.
The research suggests this is a decision worth making carefully. A large field study of more than 5,000 customer support agents by Brynjolfsson, Li, and Raymond found that access to a generative-AI conversational assistant increased productivity by 14% on average, with the largest gains for newer and lower-skilled workers. AI improved not just speed but customer sentiment, employee retention, and knowledge transfer. Meanwhile, PwC found that 86% of consumers say human interaction remains moderately or very important to their brand experience. Further, 58% said they were only somewhat or not at all comfortable using AI tools to engage with brands. Salesforce reports that 61% of customers believe AI advancements make company trustworthiness more important, not less.
The implication is profound. AI will take the easy work first. What remains for humans will be more emotional, more ambiguous, more relationship-sensitive, and more commercially valuable. The future agent should not be an escalation endpoint. The future agent is a relationship asset, and most companies are not designing for that future.
Three Ways to Reinvest the Dividend
Much like we do with our portfolio dividend gains, service organizations should think about reinvesting dividends instead of cashing them out. Here are three reinvestment moves that senior leaders can pilot within existing structures. Each targets a different form of relationship value that the old contact center model consistently left on the table.
1. Recovery before churn
Most organizations already know which customers are at risk. They have complaint records, sentiment signals, repeat-contact patterns, escalation histories, and survey verbatims that describe, in detail, where trust has broken. What they often lack is the capacity and process ownership to act before dissatisfaction hardens into departure. Sometimes it's a lack of creativity or experience in designing an added-value process where there is no history of such design.
A Relationship Recovery Team would use AI to identify damaged relationships and prepare a recovery brief, what happened, where trust broke, what has already been promised, and what authority is needed to repair it. Human specialists would then conduct informed, proactive outreach. Not to apologize faster. To repair trust with context, accountability, and follow-through. Service recovery research is clear on one point: a recovery call is not enough on its own. Recovery must be connected to visible resolution and learning. A company that calls a six-year customer before she cancels, with full knowledge of her history and genuine intent to make it right, has a fundamentally different relationship with that customer than one that lets her quit quietly.
2. Prevention before the complaint
Most contact centers are still structurally reactive. They respond when the customer reaches the queue. AI permits a different model: identify friction before the customer has to complain. Abandoned self-service journeys, repeated knowledge-base searches, failed chatbot sessions, payment errors, usage anomalies, and early sentiment shifts can all signal that a customer is approaching a breaking point.
A Proactive Friction Prevention team would not call every customer. It would focus on moments where a small, human intervention can prevent a larger relationship failure. The business case is not lower handle time, this is not aligned to the process value. The more relevant metric is Contact Avoidance through trust preservation: fewer repeat contacts, lower escalation rates, reduced churn. In high-volume BPO environments, even a modest reduction in avoidable inbound contacts has significant economic value for companies, and it produces an experience that customers remember: the company reached out before I had to call.
3. Turning agents into a business intelligence system
The contact center is one of the richest data sources in the enterprise. It captures product confusion, policy friction, broken processes, emerging dissatisfaction, competitive comparisons, and compliance signals in real time, at scale. Too often, these signals are summarized into dashboards that do not change the business.
An AI-enabled Voice-of-Customer Intelligence function would combine interaction analytics with human interpretation. Agents and analysts would convert repeated friction patterns into product recommendations, journey fixes, knowledge-base improvements, and policy reviews. This is not a new idea. What is new is the economics: when AI handles routine volume, the capacity exists to do this work consistently, not just during a quarterly listening session. The agent of the future will not only help customers. They will help the company hear what customers have been saying all along. In the past this might have been a few dozen cases, but AI can push this number to the thousands in human terms, a massive opportunity to strengthen retention.
Part II - Building the Agent Who Can Do This Work
The strategic case for reinvesting the dividend is only half the argument. The harder operational question is: do you have the people who can actually do this work, and if not, how do you build them?
The counterintuitive reality that most leaders miss is that AI raises the bar for human hiring, it does not lower it. The instinct is to assume that as AI handles more routine work, the agent role becomes simpler and easier to fill. The opposite is true. When AI absorbs password resets, order status checks, and policy lookups, what remains for humans is precisely the work that is hardest to standardize, the emotionally charged conversation, the ambiguous situation without a clear SOP, the customer whose trust has been damaged and who needs to feel genuinely heard before any solution matters. That work demands more of a human, not less and will require a serious review of what agent archetype should be the future of the BPO industry, but should not be ignored as an opportunity. The Complexity Curve is shifting, as the leftovers for humans require a higher performing agent in empathy and comprehension, not less capability.
This means selection has to change before training can. Companies have historically hired for contact center availability, typing speed, and script adherence. The Contact Centre Agent of the Future is selected for situational judgment, how they read context without being told what to do. For depth of comprehension, whether they actually understand what a frustrated customer is communicating beneath the surface of what they are saying. For verbal reasoning under ambiguity, the ability to hold a difficult conversation without a script and still move toward resolution. For true empathetic connection to what is going on in the minds of the customers, as if the issue was also the issue for the agent. These are qualities that assessments built for the old model were never designed to surface. Organizations that continue hiring the same way will staff their new, higher-stakes roles with people built for the old, lower-stakes ones.
Training has to change too. The old model trained compliance: follow the script, hit the metric, escalate on schedule. The new model must train judgment: read the situation, choose the approach, own the outcome. That requires scenario-based development, simulated conversations with real emotional complexity, not role-plays designed to confirm that agents know the refund policy. It requires supervisors who coach for empathy and discretion rather than manage for occupancy. And it requires a progression model that gives the best agents a visible career path into recovery, prevention, and intelligence roles, because the dividend disappears quickly if the people best equipped to do relationship work feel no incentive to stay.
Two new roles emerge from all of this, and they work as a pair. Neither is fully effective without the other, and together they form the organizational nucleus that makes every reinvestment play in this article operationally real.
The first is the Customer Intelligence Analyst. This is not a traditional data analyst. Their job is to sit between the raw output of AI systems, interaction patterns, sentiment shifts, friction signals, repeat contact clusters, churn indicators, and translate it into human-readable insight that operations can actually act on. They are not building models. They are interpreting what the models surface in the context of a service relationship: why is this pattern appearing, what does it mean for this customer cohort, and what should happen next? That requires a blend of analytical fluency and customer behaviour intuition that is genuinely rare and does not yet have a clean pipeline into most contact center organizations.
The best candidates for this role today are likely already inside the building, underutilized. Workforce Management Analysts who understand contact center data and pattern recognition but have never been pointed at customer behaviour. Quality Assurance leads who have deep interaction-level knowledge and understand what a conversation in distress sounds like, but who have never had a seat at the design table. Voice-of-Customer Analysts who are already thinking about customer signals but are siloed from operations and rarely influence what happens on the floor. Each of these profiles needs professional development, in AI tooling literacy, in behavioural interpretation, in the discipline of writing an insight brief that someone else can act on within days rather than quarters. But the foundational instincts are often already there.
The second role is the Service Process Designer. This is someone who takes the Intelligence Analyst's brief and builds what comes next: the pilot, the knowledge documentation, the success metric, the scaled program. They can read a friction pattern, design a small intervention, test it against real customer data, and write the operating procedure that a team of ten can run without them in the room. They sit between operations, service design, and customer strategy, and like the Intelligence Analyst, they barely exist yet as a named role in most organizations. I have worked at companies that are especially good at this step, however the majority are not in my opinion.
What makes this pairing significant is not just the individual roles, it is what they can accomplish together with AI tooling at their disposal. A Customer Intelligence Analyst and a Service Process Designer, working in tandem with access to interaction data and AI-generated pattern recognition, can do at scale what previously required expensive consulting engagements or large internal strategy teams operating on quarterly cycles. They can identify a relationship failure pattern on a Tuesday, design a recovery brief by Wednesday, and have a pilot running by the following Monday. That speed of iteration, from signal to action, is a structural advantage that compounds over time. Organizations that build this capability early will be harder to catch.
Measuring What Actually Matters
Even the right people in the right roles will fail if the measurement system does not change around them. Relationship-value work will be quietly killed by the wrong metrics long before it has a chance to prove its economics. It will also require internal alignment with executive buy-in to have the best chance of long-term success. This is building new sensibilities and capabilities for teams not normally entrusted with these tasks, as they are seen mostly as operators.
A Recovery Specialist should not be judged primarily by contacts per hour. An AHT-free relationship conversation should not be flagged as inefficient simply because it took longer than average. A prevention team's success is measured in contacts that never happened and customers who renewed, neither of which shows up in a standard queue dashboard. The future contact center will need multiple operating modes and a scorecard that can distinguish between work meant to be fast and work meant to be valuable. Legacy metrics, average handle time, occupancy, deflection rate, remain relevant for routine resolution. They are the wrong instrument entirely for relationship work.
What replaces them is behavioral measurement over time: did the customer stay? Did the repeat contact stop? Did the complaint reopen? Did the at-risk account renew? Did adoption improve after the education call? These outcomes take thirty, sixty, or ninety days to appear. Organizations used to measuring service in real time will find this uncomfortable. The discomfort is worth tolerating. It is the difference between knowing whether your agents completed a process and knowing whether the process actually worked.
The Leadership Choice Forming Right Now
The coming years will divide customer care organizations into two groups. The first will automate routine work, reduce headcount, and declare victory when cost per contact falls. Their service model will become cheaper, but that may not translate into a better experience for your customers.
The second group will also automate. But they will treat automation as the beginning of a redesign, not the end of one. They will ask what customers have always wanted that the old cost structure could not afford to deliver. They will build recovery capacity before they need it. They will convert their front line into a proactive intelligence system. They will measure relationship value alongside queue performance. Who you place in these roles will be important: the qualifications should be more creative, analytical and especially an ‘out-of-the-box’ thinker.
For BPOs specifically, this distinction is existential. A BPO positioned primarily as a labour-arbitrage provider for routine contacts faces genuine compression as AI reduces addressable demand. A BPO that can say, we will help you automate the right work, and we will help you reinvest part of the savings into work that makes your customers harder to lose, is offering something different: not cost management, but relationship management at scale.
The companies that win the AI era of customer care will not be the ones that only reduce the number of people in service. They will be the ones that redeploy their best people into relationship work that used to be too expensive to perform at scale.
The agent of the future is not the last human left after automation. The agent of the future is the human expression of a more ambitious service promise: when the issue is complex, when trust is at stake, when the customer needs help becoming successful, or when the business needs to learn from its customers, a capable human will be there, informed by AI, empowered by data, and measured by the value they protect and create.
AI will expose the difference between companies that view service as a cost center and companies that view service as a relationship system. The first group will become less expensive. The second group will become more valuable. The capacity dividend is available to everyone. What leaders choose to do with it is the strategic question of this decade in customer care.