Sri Lanka Telecom has set itself the goal of becoming an AI-native operator, and it is already halfway through retraining its workforce of more than 3,500 people to get there.

"It has become a strategic priority for us to build an AI-native telco, although it is a very ambitious target," Ishari Siriwardane, SLT's deputy chief transformation officer, told Light Reading on the sidelines of FutureNet 2026. "Right now we have about 3,500 plus staff which we plan to train. So 50% of this is already trained and we are targeting to complete the training by the end of the year."

The company's timeline for the AI push began roughly four years ago, not with algorithms but with what it calls a digital maturity assessment. That exercise spawned more than 20 initiatives under a digital transformation program and a dedicated division, now housed within the IT group. SLT has since formed a Center of Excellence team for AI and appointed a head of data and AI.

What has AI already changed at SLT?

Customer service is furthest along. Chatbots and assistants now handle front-line customer interactions, while AI agents are being built at the back end so staff no longer have to log into multiple systems when resolving a complaint.

"Now we have AI agents being built at the back end so that they can fetch this information quite quickly and then support the customer," Siriwardane said. "So the cycle time that we take to resolve an issue for the customer has improved."

Credit control is next in line. SLT plans personalized collection actions informed by AI, using signals such as customer behavior, payment history and the triggers that prompt a customer to pay.

On the network side, AI supports fault identification and resolution, particularly in the wireless network. But Siriwardane was blunt about its limits: the capability is "not gone up to the autonomous level," and SLT's network remains "very fragmented and heterogeneous." The current objective is scale — making AI "the default setting in the organization so that everybody uses it."

Why is the transformation hard for a 169-year-old telco?

Few operators carry as much history as SLT. It started as a government department in 1858, went through privatization with Japan's NTT as a strategic investor, and now trades as a blue-chip name on the Colombo Stock Exchange. The Sri Lankan government still holds 52%, and the company competes with Dialog Axiata and Hutch Sri Lanka.

Three constraints shape the AI program:

• Ageing workforce. The average SLT employee is more than 45 years old. "Changing the mindset of the people, getting them to adapt to AI and also sustain that change has become a challenge," Siriwardane said, citing the need to move away from "this fear factor that AI is going to take my job."
• Legacy infrastructure. "Running AI on top of the legacy infrastructure is quite challenging," she said. Modernization requires capital when margins are thin. "We don't want to burn our fingers as well by spending a lot of capex."
• State ownership. "We are also bound by government processes and procedures," Siriwardane said, pointing to red tape that comes with 52% government ownership.

How does SLT plan to monetize 5G?

SLT launched 5G last year and, like other service providers, is struggling to monetize it. Its answer is a new commercial angle: bundling AI tokens with 5G connectivity.

"We are looking at new products and solutions where we can bundle AI tokens with 5G connectivity," said Siriwardane.

The operator has deliberately avoided a pan-country rollout in Sri Lanka, an island of 22 million people. "It [5G monetization] is a challenge for us as well," she said. "Since we are a small country, we don't want to deploy 5G all across the country." 4G remains the mainstream network, with 5G reserved for strategic areas of high business and economic activity, such as airports and stadiums.

What about tooling costs?

SLT has provided Microsoft Copilot licenses to its entire staff, with different versions assigned depending on employee profiles. At the same time, it is evaluating cheaper open source models for internal use as token costs rise.

"We are cost conscious," Siriwardane said. "We want to make sure that it doesn't hit our balance sheet."

The company expects to complete training for its remaining staff by the end of the year, with the scale-up of AI from isolated use cases to a default operating setting the next test of whether a 169-year-old operator can absorb an AI-native identity.