McKinsey: Agentic AI can unlock new monetisation opportunities for telcos

• Casa26 keynote highlights revenue opportunity from agentic AI
• Telcos can capitalise on sticky infrastructure assets, reckons McKinsey’s Ivan Ostojic
• Network capabilities such as analytics and fraud detection could be money-makers in AI age
• But telcos need to avoid pilot purgatory and focus their AI efforts, he stressed

For years, telecom operators have watched other companies capture value from services built on top of their networks, but the rise of agentic AI could open up new monetisation opportunities for the industry.

That was one of the messages from Ivan Ostojic, a partner at global management consulting firm McKinsey, during his presentation at the recent Casa26 event in Amsterdam, where he argued that AI is reshaping the economics of the communications stack.

During his keynote at the event, organised by the CPaaS Acceleration Alliance (CPaaSAA), the McKinsey man argued there is growing pressure on the API reseller model, which he claimed is “structurally deteriorating”. Companies that package network connectivity into APIs have benefited from making it easier for developers to build communications services, but Ostojic believes that position is becoming harder to defend.

Agentic AI could reduce the developer lock-in that has traditionally helped API providers retain customers, and that trend could ultimately put pressure on some communications platform as a service (CPaaS) providers. 

It also highlights the value of the infrastructure that underpins such services, he added. Carrier-grade network infrastructure is not necessarily a high-margin business, Ostojic said, but it has one important advantage: it is difficult to displace.

“That position is defendable when it comes to strategy, but it's low margin, and it's getting more and more squeezed. But once it's provisioned, it's sticky. People don't change this very quickly, and it's not so easy to displace [physical infrastructure] because there is a limited number of companies that build the networks around the world.”

Value in the AI stack

But Ostojic's presentation was not just about the impact of current trends on the CPaaS sector – he also addressed the much bigger shift taking place as AI moves from copilots and assistants towards agents capable of handling entire workflows. McKinsey sees potentially significant productivity gains as agents learn to work together, creating a substantial new pool of enterprise technology spending, he explained.

The opportunity for operators, then, may not simply be to sell more connectivity to AI companies: Rather, it is to make the network itself a more valuable part of the AI stack, by leveraging the capabilities that are most difficult for other companies to replicate. Ostojic highlighted identity, live network signals, fraud detection and analytics generated during communications as examples of assets that can be built into AI services.

He explained: “The network-only assets are really defendable things. If you can have real-time signals on fraud, real-time analytics during the interaction – whether it's call or messaging – [and] if you can have network assets and infrastructure, that's much harder to destroy than the reseller API economics.”

That potentially gives operators a different route into the AI opportunity. Rather than simply trying to become another software or CPaaS company, they can use the network as the foundation for services that combine connectivity with data, intelligence and compute.

Ostojic warned that API providers with no strong network connectivity assets are particularly exposed in the AI era. “That archetype is going to be squeezed completely because the economics are really eroding. They're squeezed from both sides, and when agentic AI comes in, there will be even more squeeze because you won't have so much developer lock-in. On one hand, they don't have a software premium to protect the margin. On the other hand, they also don't own any of the network assets.”

For operators, that makes the strategic question less about whether to become an AI company and more about where their existing assets can give them an advantage.

“You need to choose your posture, your archetype. You can't be everything – really specialise and go deep, whether you're going up the stack or you're owning the network. That’s critical if you want to have a source of sustained competitive advantage” in the AI economy.

Escaping pilot purgatory

Ostojic doesn’t think telcos are starting from as far behind as they sometimes assume. McKinsey's analysis of the market, in terms of capabilities and AI maturity, puts telcos behind technology companies, but he said the gap is smaller than the industry might think. 

He pointed to Deutsche Telekom and T-Mobile US's use of AI in customer support, including a reported 10% improvement in first-contact resolution and more than 8,000 agents upskilled to use AI.

“Every year we run a tech and AI maturity index where we're benchmarking maturity. As you would expect, the tech companies are leading, but actually, telcos are not much behind,” he said. He noted that the telecom sector often criticises itself for being too slow, “but there is a good momentum in this industry.”

The bigger problem, argued Ostojic, is what happens after the initial pilot stage of AI adoption. His advice to operators is to “escape pilot purgatory”: According to figures released last year by MIT, less than one in ten AI pilots reach production. 

“Adoption doesn't mean advantage,” Ostojic said. “The game is changing now. The real question is – how can players in this industry attack this budget? I think it's better to focus on a couple of use cases that will transform domains in your companies” and have a tangible impact with those targeted efforts, rather than “trying to do everything everywhere, throwing a copilot to people and hoping it will work. We need to escape this purgatory and find a few vertical use cases to scale.”

For telcos, then, AI may not be about finding a way to leave the network behind but making the network more useful to the AI economy.

- James Pearce, Editor, TelecomTV