If you want to raise profit per bit as a network operator, you have to raise revenues, lower costs, or both. For decades, operators enjoyed significant growth in mobile services and even good growth in wireline broadband, which gave them a cushion in terms of profit, but even during that period, profit per bit was falling. Cost management has been a big part of operator planning because of this, and it’s expected to be a big part of 2024 plans too. The question is how they’ll manage costs.
All 88 operators I interact with have offered comments in where they stand with cost management and planning. All of them say that the expect to target both capex and opex in 2024 and also in 2025, but there are pretty radical differences in their planning beyond those basic points. Some of the differences reflect differences in their market areas, some result from differences in where they are with respect to capital plans and modernization, and some from age of current equipment. It’s all worth looking at.
Only 18 of the operators believe they will be able to improve profit per bit in 2024, largely because the initiatives they have planned won’t be effective fast enough to turn things around. This is why they’re doing something fairly rare in my experience, which is to talk about a two-year program more than about the next year alone. Those 18 operators have a common condition; they have been undertaking a major 5G rollout that’s now largely complete, and so their capital spending is going to drop without requiring any special measures at all. What’s really interesting is that only 9 operators that saw 5G budget relief didn’t see a likely improvement in profit per bit, and that was because this group planned to increase capital spending in other areas in 2024, using some of the 5G largess to replace aging gear or to increase capacity. This group expects capex to fall in 2025, and in fact 75 of the 88 operators said their capital spending for 2024 and 2025 would be “at least slightly lower”.
One tech area that’s a potential beneficiary of this capex pressure is the open-model network area. The fall planning cycle that’s now almost complete for network operators has raised the profile of open-network projects to the point where 49 operators say they plan to review the use of open-model (white box) technology in the next two years. Slightly less than half this group say that they don’t expect to see much beyond investigation and perhaps a lab trial in 2024, but the remainder actually believe they’ll pilot some open technology in their networks in the coming year.
According to the operators who are looking (or say they’re looking) at white boxes, the big problem they face is the scope of deployment. Most operators say that 5G should have been an opportunity to introduce more open technology into their networks, since a major upgrade always improves the opportunity to penetrate a significant piece of infrastructure. Only 5 said that they’d done that, and the others now believe it was probably a mistake to miss the chance. As a result, they now face the problem of having, on the average, only about a quarter of their network subject to modernization through the introduction of white boxes, and most say that won’t be enough to reap full benefits. That’s a big reason why 2024 and 2025 are being lumped together in planning.
There are three areas of focus with open technology. One is in the RAN, for those operators who still face 5G NR upgrades. One is in the core, where some operators have capacity increases planned that would enable them to consider open technology, and one is metro, where hosting and aggregation combine to create an opportunity to build a new model of infrastructure. RAN and metro have the most support among operators with open-network aspirations; core modernization seems to focus on 5G Core, which is mostly really a metro and control overlay technology.
None of the operators think that capex management in 2024/2025 will be sufficient to offset revenue growth challenges. Their problem is on the opex side, and how to continue a decades-long effort to reduce costs. While nobody likes to say this, opex reduction is really about headcount reduction, something operators have been working on since the 2000’s. That’s the problem; low apples have been plucked.
According to all the operators, their opex initiatives have had the greatest impact in two areas—craft personnel working in the NOC or outside plant, and customer service personnel. In the former group, self-install practices have expanded as operators shift to a model that gets broadband media into the home or accessible to the home, with the customer taking it the rest of the way. FWA is the extreme case of this; you deploy an RF node and let the customer install everything in a spot where the RF can reach. In the latter group, the big hope is for AI.
Having people answer support calls is expensive. Offshoring the effort has some benefit because of lower labor costs, but operators admit that it’s difficult to get offshore personnel who can speak the local language well enough to be understood, and that problems in comprehension result in abandoned calls, resentful customers, and churn. Chat approaches are considered highly valuable, but the problem is that most support problems relate to wireline services and if those services are down, it’s necessary to use mobile service to create the chat session. Support apps are considered the best approach here, but operators are slow in exploiting the technology and often their apps are primitive and inefficient. In any event, chats often serve only as a way of siphoning off a portion of calls, and the next level usually then requires more-skilled (and more expensive) personnel.
AI integration with chats has been a mixed blessing in opex terms, according to the 39 operators who say they’ve done it. The problem is, in the words of one operator, “AI can help us understand and respond to meaningful customer queries, but most of our queries really aren’t meaningful.” The user rarely has an idea of what question to ask, and so what is really necessary is an AI agent that can see into the service and deduce things rather than rely on customer statements.
Operators, and most others who have used chatbots or AI in call center applications, believe that their biggest problem (and the reason users don’t ask meaningful questions) is that we’ve grown dependent on technologies we don’t understand. I don’t have data on the broad consumer market, but it’s interesting to me that of 214 technologists I’ve chatted with over the last year, 188 said they’ve made support calls to service providers or vendors themselves. Even technical people can’t fully self-support with regard to the technologies they use. And, of course, this increases the number of support interactions and the likely depth of knowledge required to properly address them.
Which means, likely, that our demand for support is advancing faster than our ability to fulfill it. Could AI fill that gap? Obviously not what we usually call generative AI, because the typical models are trained on broad-market data rather than on product, service, or company data. Specialized LLM technology, trained on up-to-date company information, could probably offer proper support, but none of the operators I’ve talked with said they had a program to do that.
Stepping past support-based missions, the other area where AI might be helpful is in what I’ve called “process opex”, meaning the opex related to maintaining network services and infrastructure. For process opex, the focus is less on LLM than on machine learning, at least for the moment, and operators are very interested in being able to use AI to improve service and network operations. There are barriers to this, however.
The biggest problem operators cite is the “organizational divide” between OSS/BSS and NMS technologies. The great majority of operators put OSS/BSS under one organization (the “CIO” is the traditional lead) and NMS under another (“operations” or the COO). This leads to a classic internal dispute; does a “service” get created by network behavior, or is the “network” simply molded into and judged by service requirements? So far, the majority of operators with interest in AI for process automation have been on the NMS side, and vendors there have been more active. Juniper’s growth in AI products, for example, has been market-leading, and Nokia has recently introduced the idea of using digital twin technology to model systems of devices, which could increase the value of AI agents by increasing what they know about device states and device relationships.
Operators are a bit behind enterprises in terms of AI commitments, reflecting perhaps the organizational separation, the relatively long-lived and varied infrastructure elements, and the inertia of telco organizations overall. Perhaps this is a reason why only 53 of the 88 operators believed they could achieve “significant” opex reduction in 2024, and why they again had higher hopes in 2025.
Cost management may be critical for operators, but none of them think it’s a magic bullet in the battle for improved profit per bit, particularly in 2024. That means that budgets will remain tight next year, and that the health of the vendors in the space may depend on how well they do prepping for a better result in 2025.
