One clear trend in wireless is the increasing level of network complexity and the corresponding rise in operating expenses. Networks support three or four technologies (e.g. GSM, 3G, LTE, Wi-Fi). A proliferation of frequency bands ranging between 600 MHz and 3500 MHz lead to different service performance. There are many different types of base stations such as macro cells and small cells. More networks are planned in the mix such as 5G, LAA and MuLTEfire. A nightmare scenario is developing for network operators who have to maintain these networks and meet target performance level. Operating expenses are increasing at a time when revenues are stalled. This depressed EBITDA margins into 20% range for some service providers from a level that ranged between 40-50% only a few years ago! Could Artificial Intelligence (AI) be the answer to conundrum?
TEMs are concerned
Ripples of this challenge are reverberating on the vendor side. TEMs are asking how they can help the service providers better manage the operation of the networks. One answer is through more technology: Artificial Intelligence. Huawei, for example, presented at their recent Analyst Summit a well-laid out vision for implementing Artificial Intelligence to aid service providers better manage their networks. But what intrigued me most is that Self-Organizing Networks (SON) was the answer to reducing network opex, yet it has failed to gain traction among operators. So will Artificial Intelligence succeed where SON failed?
Looking back, a few factors delayed SON implementation:
1. Operators were not ready for SON
SON includes a number of features ranging from the basic to the very complex. Operators managed to implement many basic SON features internally without relying on vendors’ SON solutions. Where it came to advanced features, operators backed out as they were not ready to implement these features: they simply had more basic issues to tackle.
2. Operators mistrusted SON
This is a bit of a overstatement! More precisely, operators mistrusted closed-loop SON operations, or the automation of SON processes. In closed-loop mode, SON algorithms implement fixes automatically. Operators could not trust turning the operation of their networks to computers and wanted to maintain the human involvement in decision making. This is a case where technology is ahead of people!
3. Operators did not want to pay for SON
With basic features implemented internally, operators could not justify the expense associated with more advanced technology features. The super-sized networks based on small cells did not materialize. The business case for SON had little justification.
4. Vendors failed to deliver on promises
Some advanced SON features simply did not work! Trials by operators uncovered many shortcomings of different SON solutions. It was the case of overpromise and under-delivery. A few key operator requirements were not available such as multi-vendor support.
Will AI win where SON failed?
First, AI is not a substitute to SON, but another technology that can be used to improve SON implementation. Think of it as SON 2.0. But more importantly, AI transcends SON to a host of application that enable operators to not only reduce opex but also to capitalize on missed revenue opportunities.
To be successful, AI applications need to work in the first place. Also, they need to have a demonstrable value proposition and deliver on a viable and proven business case.
Operators are interested in services that tie in improved opportunities for revenue growth, especially in targeting specific customer segments on the fly. These AI applications will receive attention.
To lower the barriers to entry for AI applications, vendors should consider new business models where AI applications are available as a service instead of a product. Such practices already exist, so learning from the experience is important. Additional approaches that consider the strategic objectives of MNOs exist!