Cisco Live: AI rewires customer experience

Customer experience is often reduced to the moment you need help. Carlos Pereira, chief architect for customer experience at Cisco, wants to widen that lens.

“The best way for us to start is to ask what you mean by CX, the customer experience,” he told Gadget during the Cisco Live Europe conference in Amsterdam last week. “There is usually an association of CX from two angles. One is contact centres: people read CX equals contact centre. The second is AI, and AI equals chatbots.”

That definition may hold in many industries. Inside Cisco, he says, it barely scratches the surface.

“Cisco CX deals with the life cycle end to end. When a customer buys a solution, there is plan, design and implementation. Then there is a team responsible for customer success, which is more about adoption. Then we have renewals. And in addition to that, we obviously have support.”

While support is visible, Adoption and renewals are not. Yet those stages determine whether a customer stays.

From that broader view, several clear shifts are emerging. Pereira outlined four clear trends in the changing shape of the customer experience:

Trend 1: AI has to live inside the workflow

The first lesson Pereira draws is that AI cannot sit on the sidelines.

When Cisco introduced an AI-driven workflow to support a team handling a complex internal process, the results were measurable: time taken was reduced by more than 40% and accuracy increased above 90%.

Yet adoption lagged.

“When you checked the adoption, it was not what we anticipated,” says Pereira. “They checked it once, and then they ghosted.”

The reason was straightforward: “It was not embedded in the daily workflow. It was a tool that they could consult. There was optionality.”

Once leadership integrated the AI into the core process and aligned incentives, adoption rose above 90%.

For Pereira, that marked a shift in thinking. AI that improves a task is useful, but AI that becomes part of the task changes behaviour. As a result, the industry’s early fascination with chatbot interfaces is giving way to deeper integration.

“Agentic will become less of the hype and more of a foundation, because what actually matters is the business workflows that the companies run.”

Trend 2: Users want freedom, even when they don’t use it

Another internal experiment revealed how strongly perception shapes acceptance.

Cisco built an AI assistant for a defined group of experts. The team identified a recurring set of questions, and the interface was designed around guided prompts. Generative AI  then produced new answers, and accuracy improved sharply.

“Accuracy went through the roof,” says Pereira. “People loved the accuracy. But they said, I want my freedom to put a question. We said, these were the questions that you gave us up front. They said, yes, but I don’t like that it is scripted.”

The interface was opened to allow free-text input.

“Guess what happened? They wrote the same thing.”

The logic remained intact, but the experience felt different. Pereira sees that as a reminder that AI systems must balance control with the perception of autonomy. Guardrails can sit behind the scenes, but the surface should feel open.

Trend 3: Voice will thin the front line

As AI becomes more conversational, a more disruptive question surfaces: does it erode customer experience as a field?

Pereira separates contact centres from the broader CX function.

“If you look at CX as a synonym of contact centre, your question is very appropriate. But for CX at Cisco, this is just a modality of consumption for support. It has nothing to do with adoption, nothing to do with renewals, nothing to do with the life cycle.”

The contact centre layer is already shifting. Cisco handles about 16-million support interactions a year. Roughly 1.5-million become formal cases, implying a deflection rate above 90%.

“for the 1.5-million that we have, we use AI to actually for classification, from what we call Severity One, which is like full catastrophe, and the network is down. Severity Four is like when we have 48 hours to respond. Everything that is Severity Three and Severity Four is already answered through AI, because I can offload those, because they are less critical. My expert engineer resources focus on what matters. So from that lens, the business workflow changes.”

While AI classifies and routes, human specialists step in where judgement is required. As a result, the front line becomes leaner, but the need for expertise remains.

Trend 4: Retention is becoming predictive

The most financially significant shift lies in renewals. More than 55% of Cisco’s revenue is recurring, placing retention at the centre of its model.

“If you have higher adoption, you have a higher likelihood of renewal,” says Pereira. “It’s like a gym membership. If you never go to the gym, you are not going to renew.”

Customer sentiment also plays a role, since service failures erode trust, while consistency strengthens it.

Cisco built a predictive model using traditional machine learning to score renewal risk.

“LLMs don’t go there. They are very bad on predictions.”

The predictive engine produces a risk score. A fine-tuned language model then interprets that score, correlating adoption data, sentiment signals and historical benchmarks into an explanation.

“I don’t care if the risk is low, high or medium,” he says, reflecting internal feedback. “I want to understand the reasoning why, and what is the guidance.”

As a result, the system generates mitigation plans automatically. It highlights both negative signals to address and positive factors that can support renewal or expansion.

For renewal teams without deep technical backgrounds, that context is critical.

“Imagine a renewals person with a financial background,” says Pereira. “If I have a high-risk renewal, the likelihood for that person to explain why my product is worse or better is very low.”

AI narrows that gap by assembling relevant signals into a coherent plan.

Confirming instinct

Has AI contradicted any long-held assumptions about retention?

“It reinforces some sentiments that we already had,” says Pereira. “But it also helps to summarise and correlate some of the things.”

Support data spread across regions can now be aggregated, and recurring issues become visible more quickly.

“We can go back to engineering and say, this is happening multiple times across multiple regions. You need to take care of this.”

In other cases, AI exposed opportunities that had not been formalised, such as automatically generated mitigation plans for renewals.

“Nobody thought about that before. When you bring this together, it is very helpful.”

The visible face of AI in customer experience remains the interface. The deeper change lies in how organisations predict risk, interpret behaviour and reshape internal processes. The customer may never see that machinery, but they are likely to feel its effects.