What’s in the way of telco AI adoption?
The telecom industry is anxious to figure out artificial intelligence, particularly generative AI, and how it can be applied to their businesses and either create new revenues or reduce costs. The will is there, but the way to do so — with a return on investment — isn’t quite as clear at this early juncture.
Reece Hayden, principal analyst with ABI Research, is part of ABI research’s strategic technologies team and leads the Artificial intelligence and machine learning team. “From a telco perspective, the challenges around AI implementation are enormous,” he said at the recent Telco AI Forum virtual event. Why? Because telecom providers are operationally complex with huge data silos, huge footprints, lots of legacy infrastructure and systems and don’t necessarily have the necessary talent or capital to organize and integrate all of those things into a coherent and holistic foundation for AI, with an available and consistent data set on which to train foundational models or otherwise fine-tune AI for specific use cases. On the flip side, however, there are plenty of potential opportunities for telcos to apply AI, Hayden said. Some of the use cases that ABI has identified include customer or internal-facing chatbots, workforce scheduling, summarization of internal documentation, regulatory monitoring and summarization, and fraud detection, eventually escalating to automated capacity planning and automated customer ticket handling.
So what are the barriers to telco AI adoption? He identified two major factors hindering AI adoption: Is the technology ready for the way the telco wants to use it? And is the telco ready for the use of AI? In many cases, at this point, neither one of those is true.
“Generative AI is not there yet — it’s not ready for high-risk use cases,” Hayden said. He went on to add: “A lot of the conversations I have within the industry at the moment are around generative AI, generative AI — how can we implement it, how can we make money from it, how can we improve upon it? But when you take a step back and recognize that generative AI is one of many different options in terms of deployment, I think it’s important to recognize that it’s not always right. And generative AI has a huge cost profile, it’s a probabilistic model, so it lacks a fair degree of accuracy. It has limited reliability in certain use cases. The data availability [required] to train generative AI models is much higher. Time to value is much longer. When we take all of those in and look at the available telco AI use cases, we realize that generative AI is not always the best fit. And it’s not always going to deliver that great ROI that you’re looking for,” he concluded.
Hayden pointed out that if, say, generative AI can provide 80% accuracy, telcos should expect it to fail 20% of the time at an assigned task. So while there are uses for gen AI, building an AI strategy means matching both the business’ readiness, the technology’s readiness and the telco’s willingness to take on the risks when any given AI models gets things wrong.
“We’re still in a very early phase of adoption, where there are high ROI use cases such as customer chatbots, which have been implemented effectively,” Hayden said. “Those use cases are, yes, impactful, but they still have a huge element of human interaction.” Most AI use cases at the moment are more about augmenting human employees and making them more effective, and thereby reducing costs, than about creating new value or revenue. “When you add human oversight into that generative AI model, it can become a very powerful tool,” Hayden said.
The second factor hindering telco AI adoption is the telco’s readiness — and that primarily means data readiness. “One key area that implementation will hinge on within telcos and other enterprise verticals is definitely the data strategy, and it’s realistically the foundation for AI/ML from the base up,” said Hayden. “You’re only going to get as good a model, and as good an ROI and outcome, as the data you train the models on, fine-tune the models on.”
Telcos face especially daunting data challenges around data reliability, availability and regulation, he added: They have a massive amount of data coming in, often unstructured and sometimes suspect; data silos across their different business units; and shifting data regulations and required regionalization and privacy laws, which can translate to data mismatches across the telco’s business. Hayden pointed to five data areas within any business, but particularly telcos, which need to be addressed in order to put in place a successful AI strategy: OSS/BSS data; public and synthetic data (such as data used for testing); data standardization and sharing; centralized data governance and anonymous data.
While the challenges associated with getting telco data in order and using it as a basis for AI are substantial, Hayden remained optimistic that they will ultimately be worth it. But, he also added that a successful AI strategy won’t look at gen AI alone.
“AI implementation is not just generative AI, and it can’t be about generative AI. It’s about blending different AI models depending on a range of factors,” he concluded.
Listen to this session and more on-demand here.
Click here for more on how telco AI supports network automation and 5G monetization.
