I took some time from my time off last week to do some modeling on the way AI could impact telcos. There are, according to my model, some verticals that have extraordinary potential to gain from AI adoption. Is telecom one of them? Let’s see what I found, starting first with some foundation points.
First, let me say that when I talk about the value of AI, I’m talking about the general case of artificial intelligence and machine learning (AI/ML), not just about the popular example of large language models (LLMs) and generative AI. I’m also talking about the “assistant” model of application, the autonomous model, and the application component model, where AI elements are used as a part of an application workflow. It’s important to understand that there is no single AI type or application model that’s useful, nor is the utility of any consistent across all these verticals.
Overall, telecom revenue dollars break down into capex, opex, and profit. Typically, capex makes up 21% of the applied revenues, opex makes up 68%, and profit 11%, which is quite different from a couple decades ago. Not only that, roughly 52% of opex is related to network operations (what I’ve called “process opex”) today, when in the past that was roughly a third of total opex.
Success, according to telcos themselves, is achieved through growth in average revenue per user (ARPU), increasing share of the total addressable market (TAM), and managing churn to avoid losing your customers to someone else. Thus, their measure of success really relates to revenue gains. However, all telcos recognize that there is a need to control or (better yet) reduce opex in general and process opex in particular.
OK, these are the foundation points we need to consider. AI works for telcos if it can meet the goals these points dictate. Can it? We can look at that from two perspectives; what telcos themselves say, and what my model says.
Right now, telcos report having a focus on the LLM/generative, AI-assistant, approach. There is no vertical in which I find that enterprises report great AI success with that approach, nor do telcos report it. The model says that telco value in AI-based customer support could be great, but only 8% of telcos report significant gains with AI in that mission. From what I see, the problem is one of implementation. To succeed with a support chatbot, enterprise results say you need a combination of a support-centric foundation model, a strong training strategy based on your own technology, and a means of keeping support data in sync with service changes and experience. The number of telcos who say they have all of this is below the level of statistical significance.
The failure to create optimum AI-based customer support has a major impact on the revenue side of telco aspirations, since problems with support rank in the top two causes of churn across the whole telco community. Even where a competitor has an appealing free-phone premium option, customers are less than half as likely to respond to such offers if their support experience is strong. In fact, my model says that a customer portal driven by AI, if fully optimized, is such a benefit that it would be justified to develop non-problem missions for the portal to ensure customers have an opportunity to get familiar with it. Again, this is not currently being exploited, according to the telco comments I get.
Process opex, meaning the activity related to network operations, planning, etc., is equivalent to business operations in the enterprise space. In general, the verticals who report the largest gains from AI are those who have the most complex operations activities, usually related to a large “plant in service”, fleet of vehicles, and workers with a high unit value of labor. Telcos qualify on all these grounds, yet the data I get from them falls significantly short of other enterprise verticals in claims of AI success. My model says telcos should lead in this AI mission, and they do not.
The barriers here, I think, are complicated to understand. First, the telco vertical reports more difficulty in acquiring AI expertise than other verticals with similar opportunity profiles. Second, telcos don’t have vendors pushing AI adoption to them as fervently as those other verticals do. Third, telcos have unusually long depreciation cycles compared to similar verticals, which means that they can’t implement major changes in technology as easily because the write-down of residual value of the old stuff makes a business case harder to make.
The vendor issue needs some explanation, I think. Enterprises typically have a mixture of vendors, but in almost all cases there is one vendor (almost always the primary IT vendor, meaning server or platform software) who has decisive strategic influence. This vendor is in a position to drive project processes, and for most enterprises the strategic leader sees a benefit to promoting AI. For telcos, there is almost always one dominant vendor, a network vendor. This vendor often doesn’t see a specific benefit to pushing an AI project, and may in fact think that such a project might increase another vendor’s influence. I think this accounts for the difference in how telcos’ vendors see AI, and whether they’re an active force for AI adoption.
The three issues I cited above are almost surely related to the telco fixation on generative AI assistant models, which first tend to emphasize the productivity of lower-value-of-labor workers and second, don’t take advantage of the ability of AI to enhance current operations process workflows. Where similar enterprise verticals are significant users of specialized-model AI or ML, telcos are behind the curve there.
On the average, my model suggests that telcos could cut process opex by 37% with proper use of AI, and cut churn by 45%. AI could enhance marketing to raise revenues for mobile services by 18%, and improve the profitability of business network services by 25%. My model also says that AI augmentation to home and business IoT could give telcos a revenue boost of about 13% overall. No telco has offered comment on any of these potential moves, which doesn’t prove they’ve not looked at them, but is suggestive.
Some telco comments I’ve gotten suggest that there’s no “AI culture” at telcos, and that also seems to be true if we judge the culture from the bias of senior management/executive comments on AI and its potential. In all, while over 80% of enterprise CxOs are at least claiming to be AI proponents, only a little more than half of telco CxOs made positive spontaneous remarks on AI. But that’s not out of line with some of the enterprise verticals, so it may be that like actual AI opportunity, AI attitudes are a function of the profile of each company and its markets.
Then there’s the “anthropic principle” angle. The principle says that there’s an inherent observer bias, that we see things and interpret the basic rules that govern those things according to what we see, so our own views set our perceptions of the value of stuff like AI. Telco executives, perhaps more than most executives, live in a very structured world and have tuned their thinking to that structure. What fits is what has fit, not what should, and with AI we’re surely in a period when it’s the “should” that matters.
