My recent blogs on telecom opex provoked some interesting interactions with telcos, and in particular with a group of 22 who engaged me in a bit of a dialog. They were particularly focused on the point that there was an opportunity for AI to address the “business” process of telcos, not the “network” process. As this group pointed out, there’s also an overlapping set of points between the two, and that this set of issues (cost sources) might be fertile ground for AI or non-AI interplay.
One point this group had in common (and what led me to pull their views out for analysis) is that telcos are highly focused on churn. If you can reduce (or, ideally eliminate) churn, you could raise your revenue. If you could do that with mechanisms less costly than those employed today, you could lower your cost. Double whammy? Sure sounds good.
If you look back into the history of mobile telephony, you’d see that the early drivers of churn were related to the service and how the telco supported customers. “Can you hear me now?” reflected the fact that there were plenty of holes in coverage. Poor connection quality or dropped connections were common, and so were problems with customer support. If these continued to drive churn these days, then you might expect that improved services would in itself reduce churn. Twelve of the group said that they believed “management” continued to believe these old churn drivers dominate. They didn’t agree.
Nine of this group said they were aware of research done by their company into what causes a customer to drop their current provider for a new one. All agreed that their research showed that service problems or customer support problems were well down on the list. The top two items, according to every one, were “to get a better/newer phone” and “to lower costs”.
You can see the problem here. Giving customers a top-end new phone regularly is surely a significant cost. Discounts are as well. The question is how you avoid them and still retain customers. None of the group believed that it would be possible to fully eliminate either or both these churn drivers, in no small part because the phone vendors work hard to promote their new devices. None think you have to fully eliminate these factors, though. It would be helpful to simply reduce them, particularly because of another thing the group reports. That other thing is that both these prime churn drivers are diminishing over time, not because of something telcos are doing, but for natural market reasons of their own.
Phone upgrades are perpetual, but increasingly not compelling. Most customers don’t demand a new phone each year, but wait until there’s some meaningful feature to drive a change. One such feature used to be regular updates, but Google has promoted Android updates and security updates for multiple years, and Apple has had that feature for years. Camera upgrades remain a potential driver of interest in a new device, but the group indicated that in the last two years or so, there has been a slowing of meaningful camera improvement. AI features are now the big driver of interest in a new device, they say.
Cost is almost always an issue, but cost-driven churn comes along only when some operator is offering a lower cost, and obviously this could create a race to the bottom and universal unprofitability. What has been happening, though, is that reductions in overall cost have gone to mechanisms (discounts, family plans, MVNO relationships) that create lower prices, not to improving profits.
The question, this group says, that AI or any opex strategy aimed at business efficiency, has to answer is how it can help manage churn. The group had a number of suggestions.
The first, in first place by a narrow majority of 13 of 22, was to elevate the service and customer support position, not so much to immediately and directly impact churn, as to make “good service” into a specific benefit for the operator to use to hold onto subscribers.
This requires operators build a better service and support model. That would mean analyzing the state of the network in an ongoing way, and recognizing network conditions that could impact service as soon as symptoms appeared, perfect missions for AI agents. It would likely involve things like changing cell beamforming to accommodate movement, particularly regular things like commuting or the start/end of sporting events and other mass activity. It also means proactive linking of a complaint from a customer to known conditions. The group also thinks that longer-term analysis of trends in traffic, movement patterns, and complaints could be used to help position future cells, and even guide marketing initiatives. One called this “hyper-local marketing”, which would aim ads at specific service benefits in the primary use area of each customer.
They also think that customer support, especially that offered in chatbot-like form, should be more personalized. Address the customer by name. Know the likely problem in advance, and cite a specific response that’s already playing out, suggest a time of likely mitigation, and anything that might be helpful in the meantime. Even, some said, to the extent of offering an alternate route for a driver to take to get better service. A few even thought that offering routing advice based on traffic alone would be well-received, and traffic conditions can often be inferred by summarizing phone locations. “You have to look to every customer like a person who cares,” one commented. Clearly AI could help achieve this.
Another suggestion was to think about designing new service features to augment phone features. For example, AI agents could be hosted by the telco and invoked from a phone, via a telco-provided app that would also serve as a support conduit. Even an AI agent to process photos or videos could be added as a service feature, and since these could be largely or completely independent of phone features per se, they might provide a telco with valuable feature differentiation. The notion of targeting new service features to what appears to create a smarter smartphone is, I think, an interesting twist.
The last point operators made was that there were actually two fundamental problems with opex-targeting as a means of sustaining profit. The well-known one, of course, is the “cost management vanishes to a point” problem; you can’t cut costs below zero. The less-known one is that cost management only promotes price-based service differentiation, which is in itself undesirable. Some sort of visible feature, even in customer support, is preferable to simply lowering costs, though if that feature can help do that, so much the better. Sounds like a good rule to me.
