CX Operations · AI

The customer wants to be recognised. The agent wants to be able to act.

By Roberto La Rosa · Fractional CX Director · November 2026 · 9 min read

Originally published in CMI Customer Management Insights (Italian), in the series I Quaderni della Relazione con il Cliente, no. 4/2026. Republished with permission.

A customer opens a chat. Their lunch never arrived, and they rightly ask for a refund.

The answer they get doesn't convince them. They open another chat. They find another agent and another answer, different again. On top of the hunger comes anger, and a growing distance from the brand.

When I had just been hired at a European delivery platform, I saw this scene many times. The agents were kind, fast and well trained. The block was somewhere else.

That customer had been flagged by a fraud system that cross-checked a huge number of datapoints: history, orders. It was a false positive. A cautious algorithm had flagged an honest customer.

The system belonged to another department. The CX Operations team saw the effect. The cause stayed out of its reach.

The agents could override the block.

On paper they had the authority. In practice, using it was risky. Every override was read and tracked, and it weighed on their metrics. And an override that came too easily could look like an equally easy way to buy a good CSAT.

So the customer stayed stuck in the middle. The cautious system saw a suspect. Nobody managed to see a person and make them feel like one.

The personalisation paradox

According to McKinsey, 71% of consumers expect personalised interactions and 76% get frustrated when that doesn't happen [1].

Yet 56% of customers say they often have to repeat the same information to different people [2]. In my view, it's one of the worst experiences there is.

How did companies respond? With more data, obviously.

Richer CRMs, complete histories, AI suggestions on the agent's desktop. The customer still tells their story from the start.

In my experience, personalisation gets stuck in two places.

  1. The context arrives at the wrong moment.
  2. The agent doesn't have the permission, or the time, to use it.

What a good restaurant knows how to do

In my training courses I often use a metaphor: the restaurant you are always happy to go back to.

The waiter recognises you and remembers what you had last time. If the kitchen is running late, they decide on their own to bring you something while you wait. All with a warm smile and a manner that makes you feel recognised.

These gestures only need memory, context and permission to act.

People who run top restaurants separate service from hospitality [3]. Service is the technical delivery of a product, to a standard the company sets. Hospitality is how that delivery makes the person receiving it feel.

The general manager of one of New York's most awarded restaurants tells a story that became famous [4]. At the end of a dinner, a table of European guests say they have never eaten a New York hot dog. He buys one from the cart on the corner. The kitchen plates it and serves it between courses.

That gesture was outside every manual. It was possible because someone had permission to do it. Going the extra mile? Inventiveness? Creativity? For me it all starts with listening well.

Place matters too. For more than thirty years, research on services has shown that the setting affects customers and employees [5]. And we get attached to "third places": the café downstairs, the shop you keep going back to [6].

Those of us who love CX and live it every day should remember one thing. In digital, the physical place is missing.

Consistency takes its place: the same communication style and the same way of deciding, on every channel. When every chat gives a different answer, the customer loses their bearings, and their place.

More data, fewer decisions

The temptation is to fill the agent's screen. Next Best Action, real-time speech analytics, order history, flags, scripts, AI suggestions. I am a strong supporter of all these tools, when they are designed with the agent at the centre. That means efficiency, so handling time and number of clicks. It also means Happiness at work.

Cognitive psychology reminds us of a simple limit: working memory is small [7]. Every extra piece of information has to be read while the customer waits. The result is an agent who knows more and decides less. And who has less attention left for the person, who needs to feel recognised.

According to Gartner, 45% of agents avoid new technology and stick to the tools they know [8]. I read it first of all as a sign of overload.

That's why I suggest treating context as editorial work.

I call it the 30-second card: three pieces of information that change the agent's next action. Everything else stays one click away.

This is where AI does its best work, as an editor. It reads the history, picks the three lines and puts them in front of the agent before the case opens. If it shows everything instead, it becomes a fire hose.

AI that works alongside the agent delivers, and now we have solid data. A study of 5,172 support agents, published in 2025 in the Quarterly Journal of Economics, measured the effect of an AI assistant. Issues resolved per hour rose by 15% on average. The less experienced agents improved in both speed and quality [9].

This is the AI I call augmentation. It strengthens the person and leaves the decision with them.

The moment the AI hands over

There is one point where trust is won or lost: the handoff from bot to person.

Gartner asked 3,566 customers what they think of AI in customer service. 87% consider it essential to be able to reach a human agent. And people who need more attempts with AI before reaching a person use that tool less [10].

So customers accept AI. They do ask for an open door, at least for now. And they ask not to start from scratch when they walk through it.

That's why the handoff has to be designed before launch, with three decisions in writing.

When the AI hands over: at what level of uncertainty, on which topics.

How quickly a person replies.

And what that person receives: the 30-second card, already filled in.

If one of these three decisions is missing, the customer feels it straight away. They repeat their story. They wait. They reach the agent already irritated.

If all three are there, the agent opens the conversation like this: "I can see you've already tried to sort this out with the assistant. Let's look at the refund." In that moment the customer feels recognised.

Giving authority, within a written boundary

Back to the false positives. The solution we found was simple.

We calculated the average false positive rate. That average became the override threshold for each agent. Within the threshold, the agent decided without worrying about their numbers. Beyond it, the case was reviewed.

In parallel, we took the problem to the fraud team, together with the group's headquarters. More checks and more data, to block fewer honest customers.

Autonomy worked because it had a measurable boundary.

In psychology this has a name: locus of control [11]. It's the belief that you can influence what happens to you.

An agent who flags a problem and gets no answer learns that the effort is pointless. An agent with a clear boundary learns the opposite.

The difference comes from the environment the company builds.

A second story goes further back: retention at a pay-TV company, years ago. One of the first teams I led called customers who had decided to cancel. The offers were written down: a few free months of football, of cinema, or of pay-per-view events.

One agent realised the offers could be combined. He gave the customer two months of their own team's football. On top of that came the pay-per-view events, so that customer could follow their team's other matches too.

Technically, he was stepping outside the guidelines. In practice, he had built an offer made to measure for that person.

That combination worked so well it became the rule for the whole team.

There was also an effect we never measured. The customer discovered pay-per-view. They could come back to it for films not yet in their subscription. A new habit, born from a retention call.

The agent as Advocate

In the previous issue of CMI's Quaderni, Donato Dileo picked up an idea I care about: the customer care advocate, the agent as the customer's lawyer.

In the Customer Happiness Center, the model I created in 2015, the agent is called an Advocate and does more than close tickets.

Advocates carry the customer's voice into the company: to product, to finance, to logistics. Every department has a single point of contact in the team, with data and real cases.

I add one condition, though.

The Advocate needs trust and a written mandate, with clear limits. Without a mandate, advocacy stays a heroic act by a few.

Many leaders now have this on the table. According to Gartner, 85% of customer service leaders are adding tasks to agents' roles, and 75% are moving agents into new roles [12]. AI takes the easy cases. Agents are left with the cases that need a decision. It makes sense to give them the mandate to make it.

Where gratitude comes from

This issue of the Quaderni is about trust and gratitude, and I think they grow in the same place. Researchers who study relationships between customers and companies have measured it. Gratitude weighs on the relationship as much as trust and commitment. It also starts an exchange that keeps feeding itself [13]. People who feel helped tend to give back.

In my experience, gratitude comes from the gesture the procedure didn't plan for. And the chances for that gesture are countless. In the 1980s Jan Carlzon, then head of the airline SAS, made a calculation that became famous. About ten million customers a year, five contacts each with an employee, fifteen seconds each time. Fifty million "moments of truth" a year, in which the customer forms an idea of the company [14].

Customers don't remember them all the same way, though. From an experience they mostly keep the most intense moment and the way it ended [15]. The refund finally unblocked after several chats. The two months of their favourite team's matches, born from an agent who put two promotions together. These are the moments that stay. They become possible when the agent has the right context and permission to use it.

Three questions before you buy another tool

Before you add another screen to your agents' UI, answer three questions.

  1. What does the agent need to know in the first thirty seconds to change their next action?
  2. What can they decide alone, and with what written threshold?
  3. Who owns the case to the end, even when the cause sits in another department?

Sources

  1. McKinsey & Company, "The value of getting personalization right, or wrong, is multiplying", 12 November 2021 (71% and 76%).
  2. Salesforce, State of the Connected Customer, 6th edition, 2023 (56% often repeat information to different people; base: consumers and business buyers).
  3. Danny Meyer, Setting the Table, HarperCollins, 2006 (definition paraphrased).
  4. Will Guidara, Unreasonable Hospitality, Optimism Press, 2022.
  5. Mary Jo Bitner, "Servicescapes: The Impact of Physical Surroundings on Customers and Employees", Journal of Marketing, 56(2), 1992, pp. 57-71.
  6. Mark S. Rosenbaum et al., "A Cup of Coffee With a Dash of Love: An Investigation of Commercial Social Support and Third-Place Attachment", Journal of Service Research, 10(1), 2007, pp. 43-59.
  7. John Sweller, "Cognitive Load During Problem Solving: Effects on Learning", Cognitive Science, 12(2), 1988, pp. 257-285.
  8. Gartner, press release, 31 January 2023, survey of 888 agents (August 2022).
  9. Erik Brynjolfsson, Danielle Li, Lindsey Raymond, "Generative AI at Work", Quarterly Journal of Economics, 140(2), 2025, pp. 889-942.
  10. Gartner, press release, 4 August 2026, survey of 3,566 customers (87%).
  11. Julian B. Rotter, "Generalized Expectancies for Internal Versus External Control of Reinforcement", Psychological Monographs, 80(1), 1966.
  12. Gartner, press release, 28 April 2026, survey of 321 customer service leaders.
  13. Robert W. Palmatier, Cheryl Burke Jarvis, Jennifer R. Bechkoff, Frank R. Kardes, "The Role of Customer Gratitude in Relationship Marketing", Journal of Marketing, 73(5), 2009, pp. 1-18.
  14. Jan Carlzon, Moments of Truth, 1987 (original Swedish edition 1985).
  15. Donald A. Redelmeier, Daniel Kahneman, "Patients' memories of painful medical treatments", Pain, 66(1), 1996, pp. 3-8.
  16. Donato Dileo, I Quaderni della Relazione con il Cliente, no. 3/2026, CMI.

Related: Why contact center agents disengage · What is a Customer Happiness Center?

Roberto La Rosa Roberto La Rosa works in CX advisory for European scale-ups with Happiness Harbor (happinessharbor.io). He started as a phone agent and led customer service operations in several European countries. He studies organisational psychology. His book "AI Alone Is Not Enough" is in progress. LinkedIn

Can your agents decide without fearing for their numbers?

45 minutes to look together at where your customers get stuck. No cost, no deck.

Book a free call