IBM last month released their 2024 CEO survey which collated perspectives from a wide range of business leaders on their key concerns and challenges, creating a picture of the primary trends and recommendations for how to respond.  Overwhelmingly the sentiment reported is one of perceived disruptions, driven by technologies with GenAI at the top of the list.

To meet this, 67% of CEOs reported that they would have to accept significant risk in changing their business in order to stay competitive. The report quotes Nobuhiro Tsunoda, Chair of Ernst & Young Tax in Japan, as saying “If someone else destroys our old business model, we will be ruined. But if we destroy our old business model, we will survive.”

So how does this play into the telecoms world? By any measure it’s already working through significant changes. Stephen Rose stepped up this year from leading IBM’s global telecoms business to working across industries, so was in a great position to discuss the lessons and impact with 6GWorld.

Similar But Different

“All industries that we spoke to have got the same sorts of pressures,” Rose began. These include:

• Challenges around filling key technology roles.
• Reducing the cost of goods sold or the cost of operating expenditure.
• Much more surgical deployment of CAPEX.
• Organisational change as a result of implementing AI.
• Change readiness – overcoming the human resistance to change.

“The last one is about CEOs really worrying about whether they’re putting enough time and resources on long term innovations that can give them sustainable competitive advantage,” Rose explained. While leaders realise the need to make long-term fundamental changes, it’s not so straightforward.

“Their boards are putting them under pressure for short term performance, especially with things like the cost of capital. If you look at some of the telcos, their cost of capital has put billions of dollars on their debt just because interest rates went from 2 to 5% in a single year when they renewed those loans.”

Strikingly, while these concerns are common across many industries, Rose believes the telecoms sector may well have an advantage… but not in terms of technology.

“There are a ton of industry forums where collective learning appreciates, and there are forums that can actually then take what pioneers in the industry are doing out to a broader collective. We’re pretty well organised for that in the telco sector. There are other industries that are nothing like as organised for industry bodies level or influencing,” he commented.

The drawbacks we face? Regulatory oversight, which encourages a cautious approach to making changes.

“And, of course, we’ve been used to running vertically integrated stacks from the OEMs,” Rose noted.

“I don’t know a firm in the world that hasn’t shifted to a horizontal technology strategy, whether they’re conscious about doing it or not.”

This shift is playing out in the move towards network APIs and exploring the use of AI within the telecoms sector, but the changes involved are going to be fundamental not only for technology but for business and operational processes. With the cornerstone vendors of the industry used to a very different approach, this is going to be challenging. Perhaps no wonder that one of the highest priorities CEOs in the survey had was “business model innovation”.

GenAI Growing Pains

Business model innovation has traditionally been an Achilles’ heel for the industry, largely because there has been a focus on cost optimisation, whether overall or per bit, which has driven technology innovation. Sales approaches that enable CSPs to do more with their network, or reduce the spending needed to meet demands, have unsurprisingly been influential in times of dramatically rising data usage – the business case is very easy to make. However, Rose pointed out the downside.

“If you incentivise your CEO and the rest of your organisation around the TCO [Total Cost of Ownership] model too much, then you will ignore the growth imperative,” he explained.

“The industry has been trying to solve that for some time now, little bit by little bit, and I think we are getting much more interested in the growth imperative. But for me, generative AI is a growth imperative.”

That might seem like a bold statement, but Rose proposed three directions to think about this growth.

“It can grow the bottom line margin just by making people more efficient and a greater output per unit measure of time,” Rose explained. “The problem is that we still look at it as a cost-cutting initiative. I think we can be much more creative around it.”

The second opportunity lies in the ability to encourage customers to have greater experiences and greater willingness to pay. For example, connecting network data and customer information could enable proactive messaging about expected problems and the actions being taken to remedy them, making customers feel actively valued.

“And that’s where you see things like customer success or customer satisfaction come up. The same applies to enterprise customers too,” Rose noted.  

“And then I look at the sales organisations. You can get AI to work for you in so many different ways – pulling customer insight and data out of emails and just populating databases, creating different ideas to generate offers to your customers.”

All these opportunities, of course, lie within the domain of existing business models. The telecoms environment lends itself to other business models for leveraging the growth of GenAI, including supporting capabilities at the edge. Perhaps here is a viable business case, as well as use case, for MEC?

Rose commented, “A lot of the initial large language models were anything between a few billion parameters to a trillion parameters. And the problem is that when you want to use generative AI in the business setting one of the things you’ve got to be worried about is the cost.”

For many problems, especially in an enterprise setting, large language models aren’t necessary and are hugely over-engineered. Instead, the issue is domain-specific and may depend on very particular datasets.

“As you go towards the intelligent edge, you would see that the language models get smaller and smaller and smaller, and that’s because they’re doing two things. One is they’re using proprietary data, and as a result of that, you can obviously then do more interesting and different things. The second thing is, you dimension the model specific to the problem that you’re actually solving and then, of course, your cost of tuning, operating, maintaining and storing the model is fantastically smaller.”

A Matter of Control

While there is clearly a great enthusiasm for AI in enterprise, there is still considerable risk involved. Stories of AI systems providing unintended and problematic outcomes are not hard to find. Especially when a CEO is gambling a company on the outcome, doesn’t that provide a heavy counterweight to the desire to innovate? In an environment where the need to control the direction of the company is more significant than ever, depending on something that can be unpredictable seems contradictory.

While Rose didn’t disagree, “There are a number of things that you can do about that risk,” he noted.

“The approach is “We’ll put in the control mechanisms, the policies, and the governance to enable everybody to do this, and then we feel more secure.”

Having said that, it’s not an easy lift… and this is where many companies are having problems, as they realise that the implementation of AI in their business isn’t just a technology change but has a ripple effect across the business.

“Governance of these models is really, really problematic,” Rose noted. “People take a base model and then they obviously create new models or derivatives from that. And, of course, you need to know what those models are, where they’re being used, who’s got access to them and where the data came from.”

However, “the more that you actually do that, the greater your chance of democratising the technology across the organisation because you feel that you’re sufficiently in control.”

Another element is to actively test different AI models with your data. Different models will behave better or worse for particular situations, so having the opportunity to select from different ones, whether developed in-house or elsewhere, gives a great deal more opportunity to find the models that will perform as desired. Which leads to Rose’s third recommendation for control mechanisms:

“You need make sure that you understand which models can do what and how well they’re performing. If you have your own performance measurement system, you can say “Actually, we’re swapping from Model A to Model B right now, because it suits our purpose better.” And you can build policies around that.”

All this is very well, but people have to know how to do all this and skills are in short supply. This is where IBM’s collaboration with the GSMA might stand people in good stead.

“We’ve created a programme with them, the Watson X Challenge. It creates an ecosystem or a co-innovation framework which encourages members to participate with individual use cases. We will then support them through the proof of value or the proof of concept around that, and then once we’ve proven it and they’re willing to implement it we put that idea back into the ecosystem so everybody gets to raise their insight and foresight and knowledge around how to implement AI,” Rose explained.

“We’re trying to raise the whole industry by 10% rather than the individual customers all over the place.”

Image courtesy of Microsoft AI image generator