Given that AI is listed by enterprises as the hottest current technology development, and that I believe the same is true on the consumer side, it’s fair to ask whether the network operators are as committed to it. Do they see themselves as users of AI, providers of AI, or both, and if the last choice is the one they make, are they providing AI internally or as a service to customers? Let’s look at what they say, based on 74 comments they’ve offered me so far this year, and compare this with the survey referenced HERE.
Not surprisingly, all 74 said they would be AI users, but the specifics varied both across the 74 and even within some (38) of those where I got multiple planner/manager/executive comments. Everyone said customer support, meaning a replacement for older-style chatbots or a supplement to customer service reps, was an AI priority. Only 23 said they’d be using it for network operations and management, and in 21 of those cases there was disagreement within company sources. A group of 20 said they would use AI in ways similar to other verticals, meaning assisting in writing tasks, and 14 said software development was an AI goal (again, with internal disagreement on the point).
In all these cases, most operators who expected to be AI users in a given area were already trying it out. The greatest interest in the operations and management application came from smaller telcos, but this group also had the largest variation in specifics. They fall into two apparent groups, one whose AI commitment is associated with an AI offering from their primary network equipment vendor, and the other whose is not, and their views are rather different.
Smaller operators tended to go first to their primary vendor for AI ops/management, usually in response to difficulties associated with recruiting and maintaining skilled operations personnel. Other larger operators, and all those who had a more open vendor commitment, tend to look to AI in general terms. That group has not reported much specific progress, only a few trials that have so far led to minimal commitments. Thus, it’s hard to say much about the ops/management applications as general AI targets at this point.
Of the 74 operators, 48 indicated they were exploring offering AI services. This group was made up largely of operators in Asia, middle-east, and Africa (to a lesser extent) who also served an active business customer base. Interestingly, these geographies represent the places where both consumers and businesses express the highest levels of strategic confidence in network operators as a group. That suggests that AI services from operators would be viewed favorably.
While North American and EU operators didn’t make up a majority of those interested in providing AI services, they all showed some interest in AI as a service. I got mixed views from this group on whether they’d rely on public cloud AI hosting or even services, or host AI themselves. Of the 22 operators, 8 wanted to resell public cloud AI service, 7 wanted to host their own AI on a public cloud, and 3 wanted to self-host. Four favored a combo approach. In all these operators there’s a mixed view among operator personnel on this topic. There’s an interesting correlation here with a Nokia Bell Labs chart that was posted on LinkedIn (you need to sign in to view this so I’ve posted it here) that represents the potential ROI of various telco AI applications. Generally, applications’ ROI potential maps to operator interest. There’s an associated white paper on this worth reading, though it doesn’t contain or explicitly amplify the chart.
The operators who saw a self-hosting option, or hybrid involving it, were uniformly in the group of operators who were looking at a general AI tool for ops/management. To me, this suggests that operators are reluctant to embrace the “carrier cloud” notion of self-hosting if the only justification is offering AI services. Given that they also seem reluctant to commit based on offering cloud computing services, that’s logical.
What I draw from all of this is that network operators are uncertain about just what they should be doing with, or about, AI. Generative AI, simply because it’s easily approached, has not only dominated prospective use, it’s slanted operator mindsets to the point where they seem to be thinking that it’s the only kind of AI out there. Applications of AI, then, become applications of generative AI except where some other influence, like that of a network vendor, intercede. You can see direct evidence of this in the fact that, of the 59 of 74 operators who offered assessments of the problems AI adoption faced for operators, 41 said that the lack of a clear productization was top. Network-vendor AI makes sense because it’s a product. Generative AI is a tool, a capability. Other forms of AI, such as machine learning and inference engines, are mostly foreign concepts. The most common answer to the question “What can AI do” is, for operators, the same as it is for most enterprises; “answer questions”. That, of course, is far too vague an answer.
Enterprises are seeing more productization of AI, including both the generative form and other forms. IBM’s own success with AI seem, based on enterprise comments, to be directly related to the fact that they take a problem-solution path to an AI strategy, which presents AI in a mission context that makes even a general tool seem like a product. Even for enterprises, though, any form of AI productization is slow to develop, and for operators any development of it at all seems invisible to the operators themselves. IBM’s initiatives there have yet to bear fruit.
Some of my IBM friends, going back even a couple decades, believe that a problem-solution strategy is still too limiting. Even generalizing it to “problem/opportunity” begs the question of whether any prospective AI user could recognize problems or opportunities to solve or exploit. One describes this as the “Phantom of productivity problem.”
What is the justification for any project? ROI, or return on investment. That’s the gain in profit that results from spending, but what generates it? To most, if pressed, the answer would be “productivity”. Technology is generally justified by making labor more efficient. In past periods where a new tech paradigm has emerged to boost IT spending growth faster than GDP growth, that’s happened by applying something specific. But AI isn’t specific, at least to most enterprises and operators. Specific technology advances that drove past cycles of IT growth essentially replaced humans, which is after all the same thing as saying “improved productivity”. How do you deal with a technology that is said to become human?
There’s also a matter of simple math at work. If we go back to the Nokia Bell Labs chart above, we note two things. First “productivity” isn’t one of the words behind the “ROI” acronym. If a worker is twice as productive, that’s a return or benefit only if there’s more work to be done, something else the worker could do, a reduction in total staff could be made, or some other financial gain generated. Second, almost every company will tell you that a positive ROI isn’t enough to justify a project. There’s an ROI threshold set by the CFO that must be met. For operators, it’s often in the 20% to 25% range, and where AI can’t generate that much ROI, there’s no project justification.
Does it help if AI lets us write a better email or letter, or even a better report, or write it faster? Could it be that “productizing” AI, for operators and even for enterprises, means aligning AI with a very readily quantifiable benefit? I wonder if all the operators’ AI targets are vague or lack conviction because they’re not linked to quantifiable benefits? That may be a good lesson for AI overall, because the lack of that linkage is arguably what was wrong with 5G.
