How BT Group leveraged AI to switch off 3G

One of BT Group’s most significant energy-saving initiatives is the retirement of its legacy 3G network. The operator switched off the last 3G site in Belfast in February 2024 and has already reported reducing its draw on the grid, saving millions of pounds in energy costs. The multi-year project also shows how the operator has deployed artificial intelligence (AI) to reduce energy consumption. The telco’s own AI tools enabled some sites to be switched off early and ensured customers did not lose service throughout the process. 

BT Group’s 3G network was more than 20 years old, had a total of 18,000 sites and accounted for 35% of its total mobile network power consumption, despite delivering less than 2% of the data. But the network could not be shut down all at once without disrupting mobile services for customers. The operator needed a way to determine which sites could be safely switched off and in what order. 

Building an AI engine

BT Group developed a Recommendation Engine that used AI techniques to identify 3G sites that could be turned off earlier than others to thin out the network rather than remove coverage. 

Coverage data from the networks team was ingested by the platform and processed by an algorithm. Based on the output, BT Group could see the sites that had 100% overlapping 3G coverage with other sites and prioritize further sites in order of the least lost coverage. 

“Prior to the Recommendation Engine we developed, identifying coverage changes was a human visual check between two coverage maps — one with the site [included] and one with the site removed. Our networks team could complete about 20 per person per day,” said Leo Cohen, Principal of Data Management for Networks at BT Group.

After the Recommendation Engine was built, he said the team could produce a list of “all 18,000 3G sites prioritized by lost coverage in a matter of a couple of hours.”

The Engine enabled BT Group to switch off approximately 4,000 3G sites early, resulting in energy and cost savings.

After two years of planning and a trial in Warrington in July 2023, BT Group started the full switch-off process in January 2024 with its 3G network supplier Ericsson. AI was instrumental during this phase as well. 

Simon Stormer, Principal for Simpler Network Transformation at BT Group, said Ericsson used automation “to complete the lockdown of sites” and this meant “significantly” more sites could be turned off each night. The supplier activated more than 1,800 lockdowns per night remotely, compared to 50 if the process was done manually. 

Automation was also used to support “bulk updates to our tools and systems,” he said. 

Data analytics keep customers in service

BT Group also developed data analytics products that combined data from its networks and consumer units to identify customers who could lose service after the switch-off. 

“This allowed our colleagues in our consumer units to proactively contact and move customers onto 4G,” said Cohen.

By creating dashboards in the Qlik Sense analytics platform and leveraging customer lists, BT Group could not only see who still used 3G services but also have insight into why they needed the service and had not upgraded to 4G. For example, they might not have a compatible device, SIM, or service plan; or they could live in an area where coverage is poor, he explained. 

With this knowledge, BT Group’s consumer teams could tailor support to help those customers migrate to 4G. For example, they advised customers to switch on 4G on their phones; migrated thousands to 4G SIMs; provided free 4G devices to vulnerable customers; and updated customer-facing websites with up-to-date information. Potentially affected customers were also flagged in the operator’s service platform, he said. 

Challenges with data

While AI helped to make the 3G switch-off efficient, there were challenges as well. The project’s timing was somewhat problematic because it coincided with various ongoing data platform cloud migrations.  

The team had to ingest data from Google Cloud Platform sources as well as from legacy platforms that were due to be migrated, which created “technical debt”, said Cohen.

The team also had to stop and implement “tokenization” part of the way through the process. 

“The challenges were significant, and we helped the central teams identify issues with volume capability in the process and even helped develop new ways to tokenise large data volumes that were 10-times more cost and time effective,” said Cohen.

Results

The total power savings from the 3G switch-off is 46 Gigawatt hours (GWh) to date, which the operator said is equivalent to 1.5 billion smartphone charges. In the nine months to 31 December 2024, the telco reported energy usage in its networks decreased 3%.

The in-house developed Recommendation Engine enabled BT to switch off around 4,000 sites early and bring forward 10 GWh of power savings and “several million pounds” in cost savings. 

“It simply would not have been possible to do early switch-off in any significant volume [without the Recommendation Engine],” said Stormer.

The Recommendation Engine also alerted BT Group to improve the quality of its coverage data, which was reported back to the networks team who could use the information to improve customer experience on the network. 

“By switching off some customer impacting sites early, we were able to refine our modelling to manage the remainder of the switch-off and be exceedingly cautious about the impact on customers,” said Stormer.

After shuttering the 3G network, BT is re-using the spectrum to bolster 4G and 5G, increasing the capacity and performance of these services.