AI influencing growth and cost, says Bharti Airtel CTO Randeep Sekhon

Bharti Airtel CTO Randeep Sekhon explains how AI is boosting network efficiency, cutting costs, improving customer experience and preventing real-time fraud across Airtel's operations

Integrating artificial intelligence (AI) into networks has driven down costs and raised efficiency for Bharti Airtel, says Chief Technology Officer Randeep Sekhon, in an in-person interaction with Gulveen Aulakh. Its implementation is positively impacting capital expenditure, operating expenditure, and business growth, and enabling the telco to stop KYC-related frauds. Edited excerpts: 

How much has been the implementation of AI on the network side? 

Anything we do on the network, including AI, influences experience, growth, or costs. We started this energy project four years ago. We used to do a rule-based (at a certain time) shutdown of capacity at cell level. We’ve now started real-time capacity management with AI, and savings on operational expenditure are up two to two and a half times without impacting user experience.  

 

How does it help business growth? 

On capital expenditure, the first is capacity planning. With the new modelling, you can forecast capacity accurately, or else you end up having some places where your capacity is over-utilised, and some places where it is under-utilised. The second is the rural acceleration programme, where earlier putting a new physical site would be based on a human survey, but now you can create an accurate model, using multiple sources, including non-telecom data sources — like traffic, blue and white goods sold, institutes, etc — to quantify which site will give better returns on investment. The third is 5G site deployment. Three years ago the criterion was the number of 5G devices and high data usage. Today, we have fine-tuned that model by adding other parameters to see where the 5G uptake will be better, and since 5G is fuelling FWA (fixed wireless access) growth, fuelling users who are converting, capital expenditure is better, which also helps growth.

 

As users we’re using AI. As a company you’re propelling AI usage (with Perplexity AI partnership). Will that not segue into more capacity being used or even capacity drainage? 

We have to see the capacity. There are two kinds of AI usage. One is a search replacement, which is not consuming any more data than the search used. Two, use cases where the UI/UX will change. For example with AI glasses where you’ll speak into the AI agent and it will tell you what you’re looking at. Tomorrow, a use case of payments is bound to come up.  

 

Does the adoption of AI also translate into spectrum management and spectrum purchase, now with the auctions coming up?

 

Lots of parts about spectrum are strategic. You would like to have spectrum. So you do take spectrum which you think I don’t need today but (you will need) in the next three to five years, because we get it for 20 years and it is made available today.

 

Is the transition from non-standalone architecture to standalone being speeded up? 

There’s no need to speed up all this. But we have moved our FWA (fixed wireless access) customers to standalone. We are, first of all, fully standalone-ready across India. That’s why all FWA customers in India are coming to this in a gradual manner. On mobile we are not moving. We are in a phase of trying what we can do. But we do not see any big gain. Tomorrow, if my mobile user goes to standalone, we will operate both standalone and non-standalone in dual mode.

 

Is there a push to get a 2G customer transition to 4G or 5G? Can AI help in any way?

 

If a person wants to use only voice, he should have the freedom of choosing his mode of connectivity and there’s nothing we will get back if we were to shut down 2G. You only lose a customer who’s giving money. So why do that? AI is only a way of making things efficient.

 

How is AI being used for preventing the next wave of spam or fraud? 

Spurious links are getting created. To counter that, you have to have a system in which every new unknown link is tested in a sandbox, marked safe or unsafe, and what is unsafe is killed. If you go back to the standard DNS blacklist, it will probably create that exposure for longer. AI allows you to predict real-time, and it is coming on the behest of a very higher processing capability of very high unstructured data.