Artificial intelligence isn’t entirely new — although its manifestation as generative AI certainly is — and the telecom industry has been pursuing it in one way or another for some time. So why is it difficult to fully realize AI for telecommunications providers’ businesses?
During a panel discussion at the recent Telco AI Forum virtual event, industry experts offered up a number of perspectives on this, particularly through the lens of challenges for implementing telco AI across operators’ businesses.
Here are five of the main challenges that they identified for implementing telco AI.
–The scale and pace of data generation. From the carrier perspective, said Ankush Saikia, senior manager of network strategy and architecture at Three UK, the scale and speed of incoming data is a massive hurdle. “Three is a very data-driven network. So although we are the fourth largest operator in the country, we almost carry 30% of the nation’s data. Which means we have the challenge of the handling of the data—because at the end of the day, to generate any type of analytics … to improve the customer experience in the network, you need to digest the metadata, and many carrier networks get generating terabytes of metadata,” he explained. “Handling of the data at scale and at pace is a major challenge,” Saikia continued, adding that people in operations or engineering need to see the data in near-real-time rather than be alerted when customers start reporting problems. One of the key challenges there, he said, is getting “good, quality data” so that data scientists don’t have to spend most of their time cleansing the data in order for it to be useful.
–The cost associated with processing large amounts of data to implement AI. “When you talk about handling that kind of data at scale, that brings the second challenge, around cost,” Saikia said. “As you handle more and more data, you’re incorporating more and more cost. And telecom, like many other industries, is very much cost-sensitive at the moment. So building the business case [for AI] is a challenge — so we need to show efficiency in handling this kind of data and bring value out of that.”
–Lack of clarity on AI’s business value. Fatih Nar, distinguished architect at Red Hat, reminded the audience that ultimately, telcos are businesses with the goal of making money, either via generating new revenues or lowering costs. When it comes to AI implementation, this foundational truth has to shape implementation. “Everything has to tie into, what are you going to do with AI in the name of business? There has to be a really solid, clear definition of it and what we are seeing in the market is, we are lacking a lot of it yet,” he said. While organizations are jumping straight to wanting to use AI, they first have to answer the questions of why, and how it ties into value for the business, as well as the how of implementation in terms of data governance, ethics, which model to use and what data will feed into it, for what business case, and put time and effort into calculating the total cost of ownerships as well as potential return on investment. There is a great deal of engineering that goes into the use of AI, but it has to be tightly tied to a telco’s business strategy. “Most of the AI projects are failing because of these: no clear use case, no clear ownership, no clear dependencies and needs, and no clear success story,” he said.
-The human and cultural element of telecom companies. “Telecom companies are traditionally made of up of people who are network experts. They are not developers. They are not debuggers. So you don’t find that kind of skill set in-house readily to bring about an AI kind of capability,” said Saikia, adding that it is therefore important for telcos to strike partnerships with AI tech companies. Three UK, he noted, took this approach and struck a partnership with Microsoft in 2023 to leverage the tech company’s Azure Operator Insights to implement AIOps and collect, organize, and process its large datasets. Jorri Kronjee, VP of engineering for KORE Wireless, pointed out that there are various skill sets that are needed in the stages of AI implementation. “Building a model and rolling it out is one thing, but verifying its performance and making sure that it does what you intend it to, is another skill altogether,” he added — and for telecom companies, data experts who understand both AI and telecom are particularly difficult to find.
–The state of telco data. Many telcos are finding that their data, while plentiful, isn’t in a condition that can easily and quickly utilize AI, especially generative AI. (Read more about this conundrum in Industry 4.0.) That means there is a fairly high (and expensive) effort that telcos need to make in order to put together the hardware and storage necessary to support AI, and make sure their data is in a format that is usable by the desired AI model in order to train or fine-tune it. “A lot of telcos may have a lot of data, but if you don’t have it labeled, if you don’t have it formatted in a way that you can feed it into a model, it is no use,” said Kronjee. “And that’s also what all of your expenses will go into, in creating those models.”
Stephen Douglas, head of market strategy for test and assurance company Spirent Communications, summed it up neatly. “To be honest, we’ve had AI/machine learning in telco for years,” said Douglas, explaining that AI/ML has been part of security gateways and firewalls for a decade and in recent years, in radios to improve energy efficiency. “It’s not absolutely new, and there have been demonstrable and proven benefits from it,” he said. The difference and new challenge, Douglas continued, is that now the industry wants to scale out its use of AI to realize even bigger benefits. In the midst of telecom’s journey through 5G and toward more disaggregated, software-based cloud-native networks, AI is seen as part of the industry’s ambition to one day achieve fully automated networks — a step that could be taken today toward such a future, even if zero-touch is actually many years away.
Douglas said: “We’re stuck in this sort of trap a little bit at the moment, because we’d like to get to Point B faster than we realistically are able to get there,” because of all the factors that were already mentioned — and because in Spirent’s experience, he added, telcos often start out looking to use AI and realizing that they actually need automation instead. But overall, Douglas said, “I think that it really goes does go back to, what is the business value you are trying to achieve, and what are the KPIs I need to get to that business case?”
Watch the full session and more on-demand at the Telco AI Forum website.
