Artificial intelligence is in yet another hype cycle – the fourth since the 1960’s. What is different this time from past hype cycles is that it is driven by data from the bottom up as opposed a top-down approach in previous hype cycles. The pervasiveness of connectivity and the Internet make AI applicable in many industries, including telecom. At the year’s Mobile World Congress, AI was a key theme. But there was little to show that has not been demonstrated in prior shows. So what are the prospects and challenges for the implementation of AI and how could AI applications in telecom evolve in the future?
A hotshot view!
Before leaving a few of my thoughts – the following video shows the view on AI in telecom from the experts:
The AI Opportunity
Essentially, the development of AI follow two tracks:
- Optimization applications: I think of these as ‘defensive’ applications where an operator aims to improve service performance and maintain its customer-base. We can sub-segment this category into two segments:
- Network operations: this leverages tera bytes of data available from thousands of nodes in a network, including base stations, routers and switches, and gateways. We have seen a few such applications, leading with Nokia Predictive Care introduced in 2015. Huawei introduced its Atlas AI platform at MWC18. A number of operators such have experimented with AI on specific use cases such as handover optimization.
- Marketing operations: applications of AI to the marketing activities of the mobile network operator, for example, to identify clients about to churn, or to make new offers and pricing models.
- Monetization applications: These applications allow network operators to target new markets and generate new revenue streams. There has been very little such applications conceived to date. But I believe this is where the business case for AI potentially lies.
AI Challenges
There are a few challenges to implementing AI in the telco space. two key ones are:
- Expertise: AI is a multi-disciplinary field; AI talent is rare and expensive. AI practitioners have a wide range of options for employment opportunities – it can be hard for a network operator to compete with a package offered by a Google or Facebook.
- Business case: it is not cheap to implement AI, so how can one justify it? Particularly, it is difficult to quantify the direct financial benefits of network optimization applications, so how to make the business case for AI? In some applications of AI in eCommerce a 0.05% improvement in sales leads to millions in additional revenues for the likes of Amazon and others. Such small improvements are noise in network optimization applications.
Inflated expectations
Will AI result in mass layoffs as machines take over network operations? Networks are getting more complex. AI helps to engineers to maintain the network more effectively. But this is does not translate to mass job losses. Service providers are hesitant to have machines control their networks. Closed-loop SON is a good example of that. We need to be realistic on what AI provides. While we reached levels of accuracy on an order exceeding 98%, for some applications this is not good enough!
A disclosure!
The Xona team has worked on three AI axes: corporate strategy, technical services and investment strategy. We also offer an AI workshop to companies looking to map their forward path. If you are interested in what we do, contact me at this link.