Dear customer, we are sorry but your browser doesn't support all necessary features for good site view. Please switch to one of the modern browsers (Chrome, Safari, Firefox).
Jindřich Fremuth
Jindřich Fremuth

At PPF, we are exploring various synergies and using AI to change the way the entire company operates, says O2 CEO Jindřich Fremuth

O2 Czech Republic

28/5/2024 | 23 minutes to read

Copy link

Mobile operator O2 has set its sights on artificial intelligence, quietly building its own language models and various AI assistants. And now it is deploying them at scale in its day-to-day workflow. “AI is merely the next step in the evolution of technology. At O2, we are now putting an enormous amount of in-house effort into explaining this to people, getting them on board for this change, and actively engaging them. We are trying to gain a competitive edge,” the company’s CEO, Jindřich Fremuth, confirms in an interview with e15, citing numerous examples of how new technologies have revolutionised the internal workings of the company, which is currently PPF Group’s most important profit generator.

In the O2 headquarters lobby, you have a great board called Faces of Change, where you praise employees’ projects and achievements. One team there caught my eye – they were lauded for “O2’s collaboration with Air Bank”… I dare say there could soon be praise there for some other team for working with other entities. Which ones?

For sure, we have plans in the works. I don’t know where you got that information, though.

It’s a matter of logic, isn’t it? You and Air Bank are hardly alone in PPF Group…

You’re right. We recently officially launched the Unity brand. This is the formalisation of a collaboration that has been under way for some time, almost a year now. It is a platform for our joint customers to benefit from additional added value. Lots of synergies can be found between the business of a bank and an operator. The introduction of Unity offers a host of benefits to the customer, whether that’s simply more pleasant service or tangible rewards. I would say that the collaboration with Air Bank is just the beginning.

How many people have already signed up for Unity?

We’re talking hundreds of thousands of customers.

I’m guessing most of them are redeeming that three-hundred-crown cashback when they use their Air Bank account to cover their O2 spending?

Yes, this is the most democratic benefit available. We’ve just added another exciting benefit – when customers buy a new Samsung phone from us, they get a television with it. Then you have smaller, popular and, in my opinion, very practical benefits, such as the option of getting protective glass applied to your phone screen for 99 crowns. We’re aiming to come up with more bonuses and figure out how we can continue to grow the Unity platform in terms of the membership base.

Can you divulge the name of another partner? Nova? O2 Arena?

Not yet, we’re still sorting it all out. Obviously, PPF Group is uniquely positioned on the Czech market. It has a broad portfolio and there are many synergies in there. But we need and want to offer them in a way that adds value for customers. Plus, we are planning to be unconventional, we definitely don’t want to create yet another “discount club”. There are dozens of those on the domestic market already, and their success is often debatable. It may sound like a cliché, but the bottom line for us is customer experience. We want clients to see logical coherence when they use our services.

Can you give me a “non-clichéd” example?

Well, today, if you’re an Air Bank and O2 customer, you can set up payments for services with a single click directly in the My O2 app. You don’t need to ferret around in the My Air banking app menu for the “direct debit” box.

That’s pretty good news from a business and customer base management perspective for your retention department, right?

That’ll probably be brought to the table for review later, when we know the longer time line, but yes, it’s a logical conclusion. The more services a customer has, the more loyal they generally are and the less they will switch. And that’s precisely because they’re leveraging benefits and synergies.

Are we likely to see less churn among domestic customers, who are traditionally very price-sensitive?

Yes, that’s one approach. On the other hand, and I don’t know if this is common knowledge, but churn in this sector is not particularly high. We’re talking less than 10 per cent a year in telecoms.

Going back to that board with the Faces of Change, you are currently changing O2 quite dramatically with your emphasis on AI, on working with data and on automation. When do you think you yourself will be appearing on the board?

You won’t be seeing me there, it’s more for my coworkers. But, of course, O2 is evolving and changing. We’ve always tried to profile ourselves as a tech leader in the market. We’re typically the first to bring new things to market, whether that’s devices, new technologies or services. You’ll soon see us moving the brand forwards a bit again.

And as for artificial intelligence, the proverbial AI, it’s been a huge talking point for us for eight years now. More than a decade ago, we set up a very solid analytics department, which we beefed up about five or six years ago with the division. Since then, we’ve been working with machine learning and “artificial intelligence”, and we’ve spent a long time using open source language models to create a range of tools for ourselves and our customers.

The division I mentioned is made up of people who are different from your average telco company employee. There are more academics, people from applied research. We started by developing our own machine learning and language analytics models – mainly to support marketing and sales. However, in recent years, we – like everyone else – have been swept up in the general AI boom. We’re currently looking to integrate AI into all the company’s processes. But my take on all this is quite multidimensional.

How do you mean?

Every day and every week new tools are being developed that can be used for business. It’s always been like that and always will be. But what I find more interesting is that all of this quite fundamentally changes how the company operates, how we do business and how we tackle problems.

But does AI generate new business?

Absolutely, but the change in the way the company operates is a pivotal factor.


Let me try to illustrate this with a simple example: a junior analyst comes to a meeting and shows me a routine economic analysis of product profitability. This tends to be a simple chart showing the trend over time. To carry out such an analysis, the analyst needs to process particular primary data. But that chart in itself is not interesting. For me, as a business leader and decision-maker, it doesn’t offer much added value. What I need to know is why the profitability has been changing and what’s behind the increases or decreases. I want to hear the business implications of an analysis like this. And if I were to take that reasoning and run with it, I also need to know if we’re analysing the right product at all. Shouldn’t I be concerned with the profitability of something else instead?

So AI has improved the analytics available to you?

First of all, our analysts are doing things completely differently as a result of AI. Instead of collecting primary data, compiling a contingency table, and extracting a chart from it, they basically ask our analysis tool to perform a certain type of analysis, and they make that request in “natural language”. And that analysis is done in seconds. This means they save time that would otherwise be spent on such a routine task. But the main thing is that, within the framework of the conversation tool we have built, they will be able to engage in a “business” dialogue and work on a solution to the problem at hand.

So they’ll be using AI as a sort “sparring partner”…

Yes, as a constructive critic who helps to deal with business fundamentals. To me, that’s a much more important aspect, it shifts people’s work from routine tasks to interesting, creative, business-like activities. It may sound mundane on the surface, but it is not trivial, it comes with a whole set of challenges and requires a comprehensive change in our mindset and how we approach our work.

So you’re starting to recruit and train people who “know how to interact with AI”?

These people need to be able to do this, and this is a need that is only going to grow. Eventually you’ll get to the stage where you can ask AI questions as though it were a “normal human”. Except AI will be smarter because it will have the data it needs immediately to hand. As a result, there will be instant feedback on the selected hypotheses that you confront it with. This is not about automation, but a shift from addressing piecemeal issues to resolving fundamental business problems.

So are you confirming that AI doesn’t destroy “human” jobs, it just repurposes them?

Absolutely. AI is merely the next step in the evolution of technology. Historically, people have always worried that technological innovation would destroy something and put them out of work, but in reality it has always tended to create new opportunities. And that’s what we’re seeing here, too. At O2, we are now putting an enormous amount of in-house effort into explaining this to people, getting them on board for this change, and actively engaging them. What I have just described is something all companies will eventually have to do. We’re just trying to get ahead of the curve so we can gain a competitive edge.

AI has “infiltrated” not just your in-house communication, but also your customer communication…

In marketing, typically, we are able to develop better products that are more relevant to the customer segments we are targeting. If you come to a branch as a client and I, as the sales guy, have good-quality analytical AI that quickly and qualitatively assesses the information we, as a firm, have about you, the chances are we are more likely to make you an offer more relevant to you. This is in the interest of both sides – customers don’t want to be offered things that are irrelevant to them, and companies don’t want to offer them anything unnecessarily either, because that’s a waste of resources for them.

But we will see – and in fact we have already seen – an increase in the speed and quality of customer care itself when we communicate with customers. And that’s right from the start.

You mean that whenever I call you I no longer have to navigate my way through a bunch of number-based options, but just say what I want to discuss when prompted to do so, and I’ll be put through?

Yes, Eva, our virtual assistant, will automatically organise your call after you introduce yourself and tell her what you need to sort out. I would say the selecting-by-numbers approach, i.e. the automated voice constantly repeating “for service A, press 1” and so on, is something everyone has come to hate over the past few years. We ourselves ditched it a few months ago. But that’s just step one. Step two is that O2’s in-house virtual assistant, Buddy, will be better at helping the operator talking to you or handling your issue.

Buddy straddles customer care, education and training. This is basically a robot that helps our consultants and salespeople to navigate the products and technology solutions we offer. It’s been a real game-changer. During testing, we noticed that people who’d just joined O2 and hadn’t done any training, but just a one-day course on how to work with Buddy, were able to resolve problems much more accurately and quickly than colleagues who’d been trained for a month in the traditional way.

So we’re talking huge savings in terms of time and money…

On top of that, there’s a much more motivating aspect for the worker. They get great results more quickly and they don’t have to do all that cramming. Instead, they can focus on what they enjoy and what they’re inclined towards, like communicating with customers in a more “human” way.

Buddy is now up and running in all our call centres, and will be rolled out to all our stores by May. But that’s just stage one.

And what exactly is stage two going to involve?

Buddy will listen in on conversations between the consultant and the customer, work out itself what the issue is, and instantly prompt the assistant with solutions and appropriate suggestions and answers. The consultant then decides whether or not to use the solution offered by Buddy. 

When is that going to happen?

We’re about a year away from that.

And in stage three you chop the live salesperson?

At that point, we’ll have to decide where to use a live consultant and where not.

What will happen to the “live element”?

Typically, they’ll be moved to more engaging positions with activities that deliver more value. We’re not at the stage where a robot can replace humans completely in customer communications. It just frees them up from having to perform routine activities that don’t add much value.

It’s a pretty big change, all the same…

We’re already preparing for it, for instance, by looking for different types of people than before. Generally, we’ll be recruiting people who are tech-savvy and who are more communicative, creative, and flexible. Obviously, it’s unrealistic to expect that you can learn something once and that it will last you for the next five years or more at work. The pace of change here is going to be extremely fast.

I don’t mean to sound pessimistic, but do people like that even exist out there?

They do, or will. Not everyone, of course, but that’s what we as a company – or, to be more precise, what we as a sector – have to adapt to. I sometimes say, with a touch of hyperbole, the quote “it’s not AI that will take your job; it’s the person who knows how to leverage AI”. I see this as a good thing. For me, it means being able to solve problems more effectively, getting things done faster, saving money, and having more time to focus on more exciting and creative activities.

How much of a cost-saver is AI across O2?

I can’t give you an exact answer, but I’ll tell you some examples, because we calculate savings separately for each application. Some benefits take the form of faster processes, others make sales more efficient or free the hands of salespeople, resulting in increased revenues – here we’re looking for an effect of, say, tens of per cent.

In other applications, we might completely replace human activity with robotics, and here we’re talking about very high tens of per cent in cost savings. There are sure to be activities that will be done by a single person instead of a team of ten, as we currently have. That one person will course, have a completely different skill set. Broadly speaking, we are looking at tens of per cent in cost savings over the next few years.

On the other hand, all of this will incur new types of costs that we haven’t had until now. For instance, AI “transactions and calculations” are not free. This activity runs on hardware infrastructure, typically in the cloud, which needs to be purchased and maintained, and the data on it has to be protected. New licensing fees for selected management tools will kick in, the energy to run all this is also a major cost…In public debate, we’ve gone from the energy intensity of mining bitcoins to the energy intensity of powering AI tools…

How much have you invested in the development of these AI tools?

Today, the overall investment over the last few years has run into the hundreds of millions. For example, an entire “AI engine” will be used to transform all customer centres. We have been working on this project for two years and are now putting the finishing touches to it. Call centres have been transformed into digital contact centres fully powered by AI tools. The costs involved are not only the creation of our assistant, Buddy, but also the purchase of suitable equipment for consultants and their training. Our IT architecture has undergone major change.

  How do you mean?

AI tools only work as well as the data you make available to them. We had to first bring huge amounts of data into one place, and continuously collect and describe it. And what we’re talking about here is not just customer or transactional data, but also, say, signalling data from our network traffic. AI tools now enable us to transcribe our conversations with customers into structured data that other trained AI then works with.

So AI keeps a record for the salesperson of what I, as a customer, am discussing with the salesperson?

Yes. Of course, you are warned at the beginning of the conversation that it is being recorded, but basically, based on a five-minute “chat” with you, the AI tool will be able to generate a logical extract, bullet points or a table with structured data. From that, we can see who you are, where you live, and what products you use – including those of competitors, if you mention them in your conversation with us. And it’s in a format that AI technology can continue to work with. Perhaps next time it will suggest to a consultant that they try to offer you our product as an alternative to the one you currently have from a competitor.

What I have described is a big challenge for companies that have hundreds of thousands of customers and dozens of different interactions with each of them. We can already do this, it’s viable and relatively straightforward. We’re planning to make this information available to our other service channels. So if you’re dealing with something on the “shop floor”, the call centre can know about it within seconds, minutes at most, in a way that’s understandable to them.

Operators don’t have to read through your account history and the like?

Yes, they’ll get a better idea of what happened in the last interaction or a proposal to make a relevant offer.

What you’re describing, is that what you’re doing now?

Yes. Incidentally, it took us only a few days to program call transcriptions into structured data output.

So it’s fair to say that the world and business are moving at an ever faster pace.

Obviously, scaling this to all interactions is another matter. But the “intelligence” itself can be produced really quickly. We may be a company generating revenues in the order of 35 billion a year, but our investments in IT and technology are in the higher hundreds of millions, sometimes around a billion or more. Of course, that all depends where we are in the technology replacement or project cycle.

Does this mean that the most valuable asset you have right now, after customers, is data? How do you protect it?

I would say that our employees are even more valuable than data. But what has been a major change is that it is only now, in recent years, that we are finally seeing how this huge amount of data can be put to use.

Are you referring to the buzzword that was “big data”?

We’ve always worked with big data, but it’s only thanks to tools that have been developed in the last few years, if not the last few months, that working with such data has become much easier and faster. Quite simply, we have planted AI tools on a robust data base and they have figured things out for themselves.

How have you handled the security of the whole process?

Everything I have described has taken place under the tightest of security. To make all AI tools relevant to us, we had to let them learn on our own data. All this takes place in a closed offline environment or dedicated cloud. Simply the traditional security measures we would employ in the development of our own language models on our own knowledge base.

But, as well as your own data, you’re also protecting your customers’ data on their devices against active external threats. You profile your network as “secure”. Leaving aside the traditional DDoS attacks, spamming, spoofing and the like, how much are advanced AI tools – that can do things like make a fake video call based on a photo and voice emulation – factored into this virtual battle? These threats must flow through your network, but privacy concerns prevent you from proactively detecting and eliminating them. Isn’t this pushing you towards some kind of new approach?

It is a huge problem, but one where we also see a huge opportunity. The number of cyber threats and attacks is growing at an exponential rate. The data on this is alarming and when you first look at it, you’d think we got it completely wrong. In our network alone, O2 Security intercepts 250 million threats targeting our customers every month.

We therefore offer DNS-level security. For customers who order this service from us, we monitor the traffic at the domain layer and filter out “malicious” traffic. We can stop you from entering domains that are known to be malicious and are in the threat database. You simply can’t click through to them. A database like this contains hundreds of millions of records. Changes amounting to tens of thousands of entries are made every day. Because we’re not allowed to monitor the content of private data streams, we can only react to an action by you. For instance, if you want to click somewhere, we can warn you. Whether you choose to go ahead is then, ultimately, up to you.

Typically, this is how we block malware sites (sites that can install malicious software on the customer’s phone – ed.), sites that try to gain control of the user’s device (for example, in a subsequent DDoS attack – ed.), or sites that specialise in phishing (deceiving people into revealing information and data – ed.).

How many people currently use your security service?

O2 Security currently has over 1.5 million customers signed up and activate. Hence the huge number of threats that we know we have stopped. We block more than two billion of them a year.

Is this the future of an operator’s business? To be a shield, a checker, a verifier that the data flowing towards me as a customer is not faulty or somehow tampered with?

It’s a logical progression. We can do it at network level. From our perspective, this is a natural step that doesn’t require customers to install any additional software or apps. That makes it very handy. Of course, that is not to say that users no longer need to consider the protection of their devices.

So what is an operator’s business going to look like in ten or fifteen years?

I’m not looking that far ahead, though of course we try to assess trends. The importance of connectivity is already huge, but it is set to grow even further. Devices will be connected by orders of magnitude more than today– such as in the context of the further rise of the Internet of Things. Of course, making this secure unlocks further business opportunities for us.

The amount of data we consume is sure to keep multiplying. This has both commercial and infrastructure significance for us – it translates into further investment. And as AI improves, deepfakes (fraudulent AI-generated visual or audiovisual content – ed.) will increase in intensity, as will AI defences against them. This will also be tied up with educating users more. I think another trend will be a complete shift away from counting megabytes and gigabytes, towards simply stating whether or not a device is connected – securely – to the internet. Even today, connectivity is more or less an essential commodity in our lives. And that necessity will only intensify.

Will privacy protection, and by extension net neutrality, eventually be eroded or suppressed? Perhaps in the name of greater protection not from spoofing, but from the deepfakes we have mentioned? Is neutrality a relic that will not survive the new era, or is it a freedom that must be preserved even at the cost of the problems its existence creates?

This is a debate that’s been going on for a while now, with both sides presenting logical arguments. I like to use the analogy of a newly built motorway. In this case, net neutrality means that it is none of your concern, as an operator, whether a bicycle, a racing car or an overloaded lorry is going down the motorway. Yet that doesn’t strike me as illogical. A certain level of control is called for.

Another thing to consider is that every customer has different needs of that motorway. There are people who are happy to drive along at 130 km/h, but then there are people who don’t want, but sometimes need, to be able to drive 500 km/h. You can’t make such a motorway happen, of course, but we can. We know how to set up virtual private networks, virtually split up the spectrum, and assign its various parts, with different quality levels, to different customers according to what they need. This approach is not directly in conflict with net neutrality, but it will result in the fragmentation of service quality across the network depending on customer needs.

I think it’s right that we can’t see the content of data streams. There are certain drawbacks, of course, but we just have to accept it, even if you as our customer may at some point want someone or something to protect you, to guarantee content safety, to scan it. You might even be willing to give up your virtual freedom for greater security. I believe there are and will be tools offering this.

Source: e15
Author: Marek Schwarzmann

Share on social networks

Share on social networks


Copy link