Lourdes Saez

I wanted the system to understand not only formal requests, but expressions such as “I want out”, and recognise that the customer was probably thinking about leaving.

Today, this sounds almost obvious. In 2013, it was not.

We spent months defining use cases, process flows, customer intentions, synonyms and semantic fields. The technology could identify words and some related terms, but it could not truly understand what the customer meant.

Then the company was acquired and the project stopped.

Thirteen years later

This year, at Hey!Broadband, I returned to that ambition.

The technology was finally ready.

What had once required months of preparation could now be built and launched in a few weeks. AI made it possible to interpret natural language with a level of flexibility that we could only imagine in 2013.

The most striking change was not simply that the technology had improved. It was how dramatically the distance between an idea and its development had shortened.

From idea to something real

This website is another, smaller example.

Throughout my professional career, I led the development of many websites. Even the simplest ones required weeks of coordination between copywriters, designers and developers. With AI, I have been able to build this one myself in a few days.

That does not mean pressing a button and accepting whatever appears.

I defined the audience, selected the evidence, challenged the language, rejected ideas that did not represent me and reviewed every iteration. AI helped turn those decisions into a working website and helped me develop the content.

It did not replace my judgement. It allowed me to apply it directly.

And that may be one of AI’s most important effects: it gives more people the ability to move from thinking to making.

A world worth exploring

AI is making experimentation faster, cheaper and more accessible. Ideas no longer need to remain in a presentation while we wait for the resources required to test them.

This does not make every idea good.

It does make it easier to discover which ones are.

Thirteen years ago, I had a vision that the technology could not deliver. This year, I launched it in weeks. And now I have created the website where I can tell the story.

AI has opened up a new world.

We should be brave enough to explore it.

I tried to change it back through the app. No availability.

I called the clinic and spoke to a bot. It first offered me an even earlier appointment. When I explained that the test had to happen at the end of August, it suddenly found the original date and asked whether I really wanted to wait that long.

Yes. That was the whole point.

Another bot later called me to confirm the change. Three interactions, involving two bots and one app, simply to return to the correct starting position.

The automation appeared to work perfectly. The decision behind it did not.

Technology scales the rule it is given

The clinic may have been trying to fill an empty slot, improve asset utilisation or reduce waiting times. Each objective sounds reasonable in isolation.

But the process optimised one operational measure while ignoring the reason for the appointment date. It created avoidable work for the customer and, presumably, for the organisation too.

AI can make an existing process quicker, cheaper and more consistent. If the process is badly designed, however, it will also reproduce the wrong outcome more quickly, cheaply and consistently.

Efficiency is not the same as effectiveness.

Start with the basics

The experience reminded me of a fault-resolution transformation I led at ONO.

We did not begin with technology. We began by simplifying the existing telephone process and questioning the journey itself: what did the customer need, which decisions were necessary and how could we remove avoidable effort?

We then created a consistent fault-resolution process across channels, turned it into a digital self-service journey and connected it to network systems so diagnostics could support decision-making automatically.

The result was a faster digital journey that became customers’ preferred way to resolve faults—both in actual usage and in their stated preference. The same network integration also improved assisted resolution over the telephone.

This work formed part of a wider digital-care transformation. Online care became the company’s preferred service channel, handling 65% of customer interactions—more than the telephone. Digitalisation, self-service and contact deflection generated €2 million in annual savings, while company NPS improved by 20 points.

Technology mattered. But it worked because we had first understood and redesigned the process.

Find the need before the use case

AI gives us capabilities we could only imagine a few years ago. That makes it a powerful enabler—but still an enabler.

Instead of looking for AI use cases, start with a problem worth solving.

Otherwise, we are not transforming the business.

We are just making bad decisions faster.

I complained. After some discussion, the supermarket refunded both the missing item and the delivery charge. It was a small claim, but it mattered to me: the minimum order value meant I could not simply replace the steak, so lunch plans had to change.

A few days later, I found it in another part of the fridge.

The supermarket had not made a mistake. I had.

I felt embarrassed—but something else happened too. My trust in the company increased dramatically. They had believed me when, as it turned out, they had every reason not to.

The cost of being right

The experience reminded me of a billing-compensation policy I introduced earlier in my career.

For claims below an agreed threshold, we allowed the customer’s requested refund without carrying out further checks.

At first sight, this might sound commercially naive. It was not.

Investigating a small claim consumes employee time, creates additional contacts and increases the likelihood of escalation. Even when the company proves itself right, it may spend more defending the charge than the charge was worth—and lose the customer’s trust in the process.

Our policy reduced escalated complaints and lowered the total cost of handling them.

The objective was not to give customers everything they requested. It was to make a rational decision about where the company should invest its effort.

Trust still needs judgement

A trust-based policy does not mean ignoring fraud or removing every control.

It means designing controls proportionately.

The value of the claim, the customer’s history, patterns of repeated behaviour and the cost of investigation should all inform the decision. A high-risk or unusually large request may require scrutiny. A small, credible claim may not.

Technology and AI can make that judgement more informed. They can identify patterns, assess risk and help distinguish an isolated problem from systematic abuse.

But the starting question remains commercial:

Is the cost of challenging this customer greater than the cost of resolving the issue?

Look beyond the transaction

The supermarket refunded a relatively inexpensive item. In return, it gained something more valuable: my confidence that, if a genuine problem occurs in the future, it will be handled fairly.

Not every customer will later discover that they were mistaken. Not every act of trust will generate loyalty.

But businesses should understand the full economics of a service decision: the refund, the handling cost, the risk of escalation and the future value of the relationship.

Sometimes trusting the customer is not generosity.

It is simply good business.