Truecaller’s India engineers built its Smart SMS feature

Truecaller has been a great app to figure out who is calling you, if the number is not in your contacts list, and to search for numbers. Two years ago, the Insights Team at Truecaller’s Bengaluru facility, felt they could make messaging also smarter. They observed that most messages nowadays are incoming messages, and are usually either transactional (related to payments, etc) or promotional.

“We thought we could help users stay on top of their messages by filtering out spam, and keeping them up to date with useful messages such as bill transactions, one-time passwords, bank account updates, bill reminders etc,” says John Joseph, who leads the Insights Team.

The feature they built did exactly that. Spam is filtered. All the transactions go into a single place. The feature, launched last month, also highlights the most important part of the transaction – the OTP, for instance, is shown prominently as a notification. “If you get a notification from Truecaller, you can be sure it is for an important message,” Joseph says.

However, executing Smart SMS, as it is called, involved many challenges. Joseph says knowing what is a ‘relevant’ and ‘important’ SMS for millions of users across geographies required a lot of research and feedback. A promotional message may be an irritant for most, but important for someone looking for deals. What is a relevant SMS for a user in India could be very different from that for users in Europe or America or Africa. In Africa, people are into mobile betting, so those messages are important for them; in Sweden, delivery messages are important.

Customisation for different regions and people required a lot of feedback. That was a huge challenge. Feedback was enabled on the app itself, which asks the user to tag a message as relevant or not. All the data was analysed using machine learning, and then the message models were designed using AI.

“Knowing what is spam for whom is a constant work in progress. Also, whitelisting of brands and the numbers they call from needs human effort, as it needs to be personally verified,” Joseph says.

Another big challenge was to do the data processing offline, on the user’s handset, rather than any remote server. This was needed to assure users of data privacy, since many messages contained important financial information. And it had to be done across hundreds of different makes of phones. “Every line of code had to be super-optimised to work on every device, including old and basic Android phones,” Joseph says.

FacebookTwitterLinkedin