What, if anything, can we say about the value of AI to network operators, to telcos in particular? Is it transformational, justified, or simply applicable? Are there any missions that fall clearly into the “gain” category? I’ve dug through almost 200 comments made by operators on the topic, and this is what I’ve found.
First, perhaps surprisingly, none of those who commented were prepared to say that AI was a game-changer. Can it make things better? Yes, in many missions. Will it make a difference in telcos’ long battle against profit pressure? Few think it would, and none thought it would resolve those challenges. Might there be future missions where AI could shine, change things more radically? Some think so, more think it’s too soon to say.
Let me say at this point that all this seems to run strongly against what we all hear constantly about AI. OK, that’s true, but it’s also true that both enterprise and telco comments I get on AI have been noticeably more cautious than the broader media publicity has carried. Some of what I’ve heard even notes this; one telco assistant CFO said “I don’t know who these people are talking to. Nobody I talk to, apparently.” At any rate, I can only say what I hear and you’ll have to judge for yourself.
To dig deeper, we need to look at missions for AI by category. The comments I got relate almost totally to four application areas, which I’ll list in order of mention. First, impacts of AI on operations and opex. Second, impacts on capital equipment and capex. Third, on customer acquisition and retention. Forth, on energy efficiency, sustainability, and resource optimization. We’ll explore them one by one.
Operations missions top almost everyone’s list. Netops, for operators, is increasingly complex, with more different pieces that interact in more ways, creating a web of interdependencies that make fault isolation difficult and raise the risk of “breaking something big when trying to fix something little.” This is, not surprisingly, the only AI application that seems to consistently pay back for those who adopt it. However, there’s a lot of groping around needed to get it right.
The big problem, say operators in comments, is that there’s a management and market conception that operations AI is some giant tool that autonomously runs the network. Senior management often believes this is possible, and some AI promoters play to that. Operators say no such thing exists, and they’d likely not adopt it at this point even if it did. Networks are an interlocking grid of technologies and vendors, and netops has always been the same. They’d like to see AIOps work that way, a set of technology- or area-based tools that were then somehow aggregated into a master viewpoint. They also see these tools as a kind of “manufacturer’s rep” expert sitting at the elbow of the network operations professional, offering diagnostic help, recommending actions, and warning of consequences. Might, over time, something more autonomous be accepted, even sought out? Sure, most say, but this isn’t that time. Still, this particular mission gets a rating of justified from almost all those who commented.
Savings on capex comes up mostly in relation to mobile operators and optimization of tower locations, spectrum, beamforming, etc. Less than a third of operators say this has a clear capex benefit though; most really relate the benefit to “optimizing” RAN costs and maximizing user quality of experience to avoid dissatisfaction and churn. They point out that, with mobile networking, periods of high use tend to be created by a broadly impactful condition, like a traffic jam, that requires urgent communication, even just to tell someone you’d be late. Too many cases where the network overloads locally because of this, and you lose a customer.
AI in 5G RAN, though, may be one of the most oversold missions as far as demonstrable ROI is concerned. This is the space where most operators tend to say that there is an application for AI, but even considering things like churn, it’s difficult to find a capex justification for it, and it’s never transformational.
Regarding customer acquisition and retention, we can already see the potential value in the comment on mobile capex AI above. Operators most often mention the need to frame network behavior to respond to major events as they are happening, not when people start to complain. In most cases, they say, you could predict vehicular traffic-driven issues because you see the movement of vehicles through cells, the density per cell, conditions in the power grid, weather events, and so forth. Could AI predict, correlate, and even establish a set of operating states that it could invoke (with human consent or override) based on analysis of the then-and-now compared with records of past conditions? It could, they believe, but right now it doesn’t do that.
There’s also a lot of interest in using AI to analyze marketing campaigns and the demography of those who elect to change their operators. What works best to acquire customers, what offers operators the best return on any premium cost they provide, and what kind of prospect returns the best over time? Some work is being done here, but not as much as you’d think. So far, most believe this mission is justified, but it’s still early.
With regard to overall energy efficiency and sustainability, it’s a mixed bag. About a third of comments suggest that those who have used AI in these applications have gotten favorable results with few failures, they also note that the impact hasn’t been substantial, at least not yet. “This is really something you have to look at as a long-term planning activity,” one said. “We haven’t run the program long enough to broadly impact power and cooling decisions because things already in place can’t be redone to make them more efficient. That’s inefficient.”
Operators are also mixed on the value of the “social capital gains” they would achieve from this application of AI. They point out the strong political division on the issue, and the fact that a change in government can quickly reverse the perception of sustainable operations. That’s focused them more on a direct power/cooling benefit, which as I’ve noted above is easiest to achieve when planning new deployments, and so can’t be fully effective immediately. “We think a lot about ‘first cost’ and so we think about ‘last benefit’ too.” They want to minimize both, and having the former and latter together means having ROI back-loaded to the extreme, at best. Rate this one as an application of AI that may, over time, prove justified.
The point here is that while telcos are still optimistic about AI having value, they’re far from demonstrating the value is transformational, and they’re not totally sure that any of the AI missions will always make a business case under current financial guidelines. In this regard, they’re not as “in” on the AI topic as enterprises, who generally believe it could be transformational even though few have really even proved it to justify the investment. Telcos are more uncertain, and none of them defined the kind of AI-centric projects that are starting to pop up at the enterprise level. In some ways, that’s surprising given that telecom is a vertical, where enterprises reside in many verticals, and concentration of interest usually makes it easier to focus on technology solutions. In other ways, given that regulations have often barred collaboration among telcos as collusion, it’s perhaps less surprising. Whether lack of experimentation will hamper further telco AI progress remains to be seen.