We still can’t really pin down the value of AI overall, so it’s hardly a surprise that telcos can’t pin it down either. What’s a bit surprising to me is that a lot of the talk about AI savings ignores the fact that AI is a technology that can be applied to enhance automation, and that telcos have been focusing on automating for decades. How many opex apples are still within reach? That’s the real question. Utility, the ability of a technology to do something useful, isn’t the same thing as making a business case. Light Reading has an interesting story on this topic, so let’s take a look at it and compare it with what I’ve heard and seen.
One issue is just what “opex” includes. The story quotes a consultancy estimate that telcos could save 30% from their opex budget, which my data says is roughly 30 cents on every revenue dollar. This would then equate to nine cents’ gain in profit. However, only a rough (depending on the operator) half of this relates to actual network operations; the other portion is the non-capital expenses associated with any business. That takes us down to 4.5 cents for bottom-line impact.
Then there’s the issue of those various-height apples. My data says that the “process opex”, the stuff that’s directly impacted by any sort of automation, has already been cut by 62% through past initiatives. Think of how often you hear a human when you call a telco, or see one when you need a problem resolved. The truth is that telcos have actually done pretty well with reducing headcount with automation, and headcount is the biggest component of process opex. And energy, which is the second-largest component? When has AI been known to cut power consumption? In any event, 62% of total apples have been picked. That means that only $1.70 or so remains on the table. The story quotes a telco as offering a forecast savings of ten million euros on a cost base of 500 million euros, which is 1.8%, surprisingly close to my model value. The same telco source, though, suggests that savings of ten percent could be achieved by the end of the decade. We’ll see.
The story also quotes Omdia, a LR property, saying that opex has been roughly steady as a percent of revenue for years, and that’s largely true because automation cut the majority of accessible human-cost apples before 2019. The question is whether AI could somehow to more, whether AI offers superior automation to other techniques, and whether the superiority could be monetized and exceed the cost of AI overall.
Not all telcos fit the happy model. In the US, for example, operators tell me that their labor cost has grown for a decade now, which says that while automation did in fact cut the labor content of telco costs, the cuts occurred early and the remaining labor has gotten more expensive. A part of this, a big part according to what telcos tell me, is that any form of automation builds out operations staff at the expense of outside craft staff and customer support staff. Ops people are more expensive. Business management compensation has also exploded, and if automation was a factor in this, could we reasonably expect automation via AI to shrink what automation overall has driven up?
Telcos have the same problem as any enterprise vertical in making a case for AI, because workers are the core of any targets for cost reduction via automation, and workers are workers wherever you find them. The classic AI concept, the “copilot” concept, has largely proved to save workers a little time, but in most cases this hasn’t cut cost because the worker is still needed, and there’s nothing else that can be slotted into that saved time to improve profits. One executive said “I don’t dispute that [a copilot tool] has saved maybe half and hour a day for each of a thousand workers, which is 500 hours a day overall. I do dispute that would mean we could cut sixty of those thousand workers. We didn’t cut any.”
One positive note here is the fact that none of the data I’ve presented, or that I’ve seen others present, is inclusive of a major AI agent deployment. AI agents, based on enterprise feedback I get, can offer a way of improving operations and perhaps even reducing staff. At the very least, they could improve handling of things like telecom service problems, which could reduce churn. This, to me, suggests that if we want to find AI opportunities to reduce opex, we need to look beyond the human-cost target areas that have been the focus of telecom automation projects for decades. AI agents are becoming the focus of more startups and even the AI giants; Google’s recent enterprise initiative is focused on agent technology. Telco adventures into AI agents are very limited, more so than enterprises, and many telcos even think an AI agent is a support chatbot. This terminology issue is understandable (our tech world is very careless with terms, often deliberately to “wash” something with a popular trend), but it’s a bit disheartening.
Another potential, emphasis on the qualifier, positive note is that enterprises have reported that even the generative AI online services can benefit a chunk of their knowledge-worker cadre, and this group is generally seen as being able to make valuable use of time saved. Better results, recovered time, and beneficial ways of using that time could in theory be a way for telcos to exploit AI. Why only “potential”? Because unlike enterprises, telcos don’t have a history of creating those beneficial uses for incremental time recovered, nor do they respond as well to opportunities to use tech to make better decisions. For this approach to succeed, telcos would have to start thinking and behaving like demand-driven verticals. How likely is that?
The biggest potential, though, would come if telcos could exploit AI to sell services to enterprises, a topic I mentioned earlier this week. In fact, telco AI services could either justify initial AI commitments that could lower the cost of internal AI exploitation to reduce costs, or the latter could help prep for the former. I think the symbiosis between telco AI services and telco AI internal AI applications is clear; what’s not so clear is whether that symbiosis can be exploited.
Some telcos tell me they have AI projects internally, and the comments they offer demonstrate that they’re less successful with AI as a cost management tool than other verticals, meaning than enterprises. Some also say they’ve launched AI services, but these seem to be very low-level, meaning things like GPUaaS, which have limited profit potential for the telco at best. None of those who say they offer these services say they’ve been particularly successful.
I think the story is right, though. I think that telcos have already automated all the things that presented obvious inefficiencies, and I think that while there is a role that AI could still play in reducing opex, it’s not going to make a good business case if it has to fund the whole of a telco AI deployment.
