Picture one customer call. A person has a problem. An agent picks up. For most of the next several minutes, the agent is barely talking to the customer at all. They search a knowledge base while the customer waits. They click through systems to work out who this person is. They write the whole thing up after the line goes dead. Somewhere in the middle, they quietly do the thing the customer actually asked for.
Verint's State of Agent Experience 2026 put a hard number on that hidden work. Surveying 1,000 frontline agents at companies with 300 or more seats, it found that across a typical call, agents spend roughly nine minutes on tasks that never touch the customer — searching mid-conversation, gathering context, completing the transaction, and writing the after-call summary.¹ It is the clearest single data point behind a conclusion the biggest research houses have now reached independently: for three years, most organisations have been automating the wrong end of the contact centre.
This is not one vendor's number
The instinct to distrust a single survey is healthy. So test it against the firms with no product to sell you in the contact centre. Accenture, studying agent work across its client base, found that 65% of agents' working hours are, on average, spent on tasks that could be automated or augmented by AI.² McKinsey sized the economic potential of generative AI across sixteen business functions. It put the figure at 60–70% of employees' time in activities the technology can now absorb — and named customer operations one of the four areas holding three-quarters of all the value on the table.³
Three independent bodies, three methods, one finding: roughly two-thirds of frontline effort is not the conversation. It is the scaffolding around it. Verint simply told us where the minutes go on a single call. Accenture and McKinsey confirmed the scale of it across the whole workforce.
So the question this edition exists to ask: if two-thirds of the work is search, context-gathering, documentation and processing, why did we spend three years pointing AI at the customer instead of the agent's screen?
The wrong end
The instinct was understandable. AI arrived in force in 2023, and the most visible cost in any contact centre is the queue of customers trying to get in. So the industry aimed automation at the front door — chatbots, virtual assistants, deflection — machinery built to stop the customer reaching a human at all.
That machinery has a ceiling, and the analysts have found it. Forrester's Predictions 2026 is blunt: this will not be the year of dazzling transformation but of "gritty, foundational work," and it forecasts that service quality will actually dip as organisations over-automate the front end and crash into operational reality. Three in ten firms, Forrester expects, will harm their total-experience growth by rolling out AI self-service before it is ready.⁴ Push automation at the emotional, complex conversations and you do not remove friction — you manufacture it.
The busywork sits at the opposite end. It is behind the agent, invisible to the customer, and it is where the minutes actually go. None of it needs empathy, judgement, or the ability to calm an angry customer — the three things only a human brings. Every one of those minutes is friction — Effort and Time, two of the six sources we track. It reaches the customer as dead air, as "let me just check that," as the second call they make because the first agent ran out of runway. Friction is a profit-and-loss line. The busywork tax is friction you pay for twice: once in the agent's wage, again in the experience the customer remembers.
Point AI at the agent, and it manufactures competence
The proof that this is the higher-return move is now peer-reviewed, not promotional. The National Bureau of Economic Research studied over 5,000 support agents. An AI assistant that surfaced answers and drafted responses raised issues resolved per hour by 14% on average — and by 34% for the newest, least-experienced agents, while barely moving the veterans.⁵ McKinsey, reporting the same programme, records the fuller result: alongside the productivity gain, it cut agent attrition and requests to speak to a manager by roughly 25%.³
Read that twice. The AI did not replace expertise — it manufactured it, closing the gap between a rookie and a veteran, and it made the job survivable at the same time. That matters, because Verint's second finding is a retention crisis hiding in plain sight: 31% of agents say they are likely to quit within six months, and only 8% name AI as the reason.¹ They are not afraid of being automated away. They are exhausted by tools that fight them. Take the grind off the desk and you keep the people — and customers stop meeting a revolving door of green agents.
The strategy the analysts now agree on: redeploy, don't remove
Here is where the tier-one consensus becomes a mandate. Gartner surveyed 321 service and support leaders. Only 20% had cut agent headcount because of AI; most held staffing stable while volumes rose, and 42% are creating specialised new roles to run it.⁶ And Gartner's sharpest warning to the cost-cutters: by 2027, half of the companies that plan AI-driven headcount cuts will abandon them as fully agentless service proves too hard to reach.⁶ As Gartner's Eric Keller frames the choice, leaders must decide "whether to simply do the same work at lower cost, or to redeploy human agents into roles that AI cannot replace and that customers value most."⁶
Forrester says the same thing from the operating-model side: the hierarchy flips, AI handles the majority of the routine, and the human becomes the exception layer — the one who diagnoses the complex case, recovers the angry customer, and coaches the machine.⁴ Both houses land in the same place. The winning move is not fewer people. It is the same people, freed from the scaffolding, aimed at the part of the call that is a relationship.
And there is a reason to move now rather than admire the problem. Medallia's 2026 State of CX Report — 552 practitioners, 1,522 consumers — finds experience quality plateauing: only 13% of practitioners believe experiences are improving, and just 17% of consumers agree.⁷ The front-door automation of the last three years has not moved the number customers feel. The capacity to move it is sitting behind the agent, unspent.
The fix, sequenced
This is Dimension 05 of the VALORICE model — People & Enablement — carrying the load, supported by Dimension 04, Operations & Journeys. A sequenced piece of work, not a platform purchase:
- Mine where the time actually goes in your centre. Verint's nine minutes and Accenture's 65% are a mirror, not a target — measure your own.
- Automate the busywork first, by avoidable time: real-time knowledge retrieval, automated context at the start, after-call summarisation, routine transactions.
- Redesign the journeys and policies that manufacture the work. A share of that search exists only because something upstream is broken. You cannot automate your way out of a bad policy.
- Measure it as an experience outcome: capacity returned to the customer, resolution quality, agent retention. In the contact centre, those move together.
The lesson of the last three years is not that AI failed in service. It is that we pointed it at the customer when we should have pointed it at the agent's screen. The organisations that win the next phase will stop trying to keep customers away from their people, and start freeing their people to serve them. We automated the wrong end. It is not too late to turn around.
