How telcos are using AI to make smarter capex decision

Telco once had high margins, with users happy to pay premiums for on-the-go calls, text, and data. But, as infrastructure costs rise with each new generation (operators are predicted to invest $1.5 trillion worldwide in capex between 2023 and 2030) and, as competition drives down customer price expectations, margins have become squeezed.

Keeping the industry profitable long-term will require telcos to routinely deploy and upgrade infrastructure that delivers customers the service quality they demand, whilst keeping the costs of doing so as low as possible.

How to deploy physical infrastructure effectively

Telcos face similar challenges to any infrastructure industry. They need to deploy big, expensive physical infrastructure – cell towers, fibre cables etc. – that deliver their service across the country. They need to deploy enough to meet existing demand, and be ready to meet future demand. But every bit of capacity that goes unused is wasted money.

At this scale, it is impossible to match supply to demand perfectly, just as it is impossible for power plants to generate exactly the amount of electricity needed by consumers. But for too long, network operators (like power plants) have spent too much money creating too much capacity.

For a long time, there was no other way. Because demand fluctuated, telcos needed to ensure there was always more capacity than was really needed, in order to ensure that a demand would never outstrip supply. That meant spending more on CAPEX than was necessary. In this business, the lights – or capacity – must be kept on.

Now, thanks to AI, we can be much more precise with predicting customer demand and expectations, and therefore much more precise with capex. We can cut down ‘just in case’ capacity and ‘better safe than sorry’ early upgrades. Perhaps more importantly, we can be highly targeted in how we roll out new infrastructure – such as 5G – to ensure it aligns with demand, and starts delivering value as soon as possible. As in many infrastructure industries, digital technologies are enabling better decisions about the physical world.

Using AI to know where to invest

The heart of the problem is that operators have tended to do capacity planning predictions based only on network capacity metrics. This is a blunt instrument. AI can be more nuanced.

Data & AI can look at multiple sources of data to make more accurate predictions. It can look at capacity and extrapolate the future, but it can also do this in a wider context.

For example, it can consider customer complaints to gain a deeper understanding of why customers complain, correlate that with network metrics to really understand how customer experience is impacted, predict where and when capacity issues – which impact customer experience – are bubbling up, and therefore where investment is needed quickly to retain these customers. But equally, it can tell you where people are perfectly happy with a service that looks substandard on paper (eg. some people want 100Mb/s of data and 1ms of latency all the time, but not everyone).

It can also look beyond network metrics to other departments. For example, analyzing marketing plans and historical conversion rates will help predict where future customers will come from. External data, eg. on sales of connected devices, from smartphones to cars to industrial machinery, can offer valuable predictors of future demand for connectivity.

Acting on AI predictions for smarter capex spending

Why is this useful? Taken together, these AI insights can build a nuanced and highly predictive picture of how demand is changing, which can inform where, when, and how to make physical infrastructure investments.

This allows prioritization of new infrastructure to where/when there is demand, and money to be made. It also reduces unnecessary costs; operators tend to deploy conservatively, but by pushing back investments until they are really needed – as shown by the data – operators can save money and extend the overall useful life of their infrastructure.

Equally, AI can show where demand is low and could be met with less capacity, allowing operators to shut down some cells to save money, with the added benefit of reducing energy and carbon emissions.

AI can also help operators become more bullish about network thresholds. Conventional wisdom says a network should ensure the load from users does not exceed the predefined level of capacity. However, we have shown that, by carefully analyzing traffic and customer expectations, networks can be safely pushed well beyond predefined levels without compromising customer experience. That means networks can do more with what they have, and reduce costly new deployments.

Finally, such data can benefit beyond capacity planning. For example, a data demand analysis will also identify useful insights, such as dense urban areas with few customers. That can alert marketing teams to where they should focus campaigns.

A framework for smart capex

Doing this is complex, but at a very top level, we can think about it as a three step framework.

The first is to gather the data and build bespoke AI models to correlate network KPIs and customer experience.

The objective is to identify network KPIs that indicate poor customer experience, due to capacity issues. To this end, various network metrics are compared with Net Promoter Score (NPS) surveys and customer complaint calls. Machine learning models are employed to pinpoint the most relevant metrics. These key metrics are then correlated with different capacity loads to determine the thresholds that impact customer experience, thereby identifying the capacity levels that will cause problems for customers.

Next, it is essential to predict when these capacity issues will occur. AI models are used to forecast network growth for up to three years. These algorithms are fully automated, and their performance can be monitored using dedicated dashboards. With this forecast and the previous data, it is possible to anticipate when problems will arise.

Finally, to prioritize and solve simultaneous issues (eg. the same capacity problems in different sectors), a financial algorithm is employed. This algorithm considers factors, like user data usage, connection time in different sectors, and overall activity across various sectors over several weeks. By aggregating this data, we are able to build a financial model, based on customer payment trends and customer experience impacts, to understand which sites are most valuable. This helps us to prioritize our capex investments, according to ROI.

Conclusion

As one might imagine, a project that combines telco network planning with bespoke AI and disparate data sources comes with technical and organizational challenges. Work must be done to understand data and minimize noise going into the models, and to build these complex models themselves. Buy in from multiple teams – including engineering, IT, finance, and marketing – who are willing to share their data and engage with resulting insights, is critical to success.

Even then, we are modeling complex systems, and the answers will not be perfect. But, done right, predictions will be much closer to the actual future network demands. Over time, that will help telcos routinely make much smarter decisions to get the most value-per-dollar from their infrastructure investments.

TelcoInsights is a series of posts about the latest trends and opportunities in the telecommunications industry – powered by a community of global industry experts and thought leaders.