Late to the personalisation journey, The Hindu feels lucky in a way that they were able to bypass a lot of the time and effort it takes to build recommendation engines.

Chief Product Officer Pundi Sriram spoke to attendees of INMA’s recent Product and Tech Town Hall about issues unique to news in India and how it’s never too late to begin the personalisation journey.

The Hindu has been at it for about a year and a half. They recognised they had millions of users going on their Web sites and apps, and kept asking themselves if they were giving those users the right content and enough of it.

This is when they started paying closer attention to personalisation.

What the data showed

Across all of its formats, The Hindu creates 20,000 stories a month. When they started analysing this, they found the sheer amount of content they were creating felt overwhelming to users. They also found 95% of user attention was going to 33% of their stories. Another stat they saw: One in two users would go to their homepage and leave without reading anything.

“We need people to actually see this content,” Sriram said. “That’s a problem we thought we could solve with personalisation.”

They started creating premium content that takes a lot of time and effort — a move they wouldn’t have made without having a subscription-based business model. They next had to figure out how they were going to get people to actually see this new content they were working so hard on.

The team keeps in mind a balance between business objectives and user needs with everything they do with personalisation. From a business perspective, they were looking to increase user engagement. From a user need point of view, they wanted readers to find content they were personally interested in, that they wanted to read at that very moment, and in a format that is convenient to them.

“Sometimes it might be a short format, a summary format of video, a podcast, or read aloud version of an article, so all of these are parts of things we’re working on and thinking of doing today,” Sriram said.

The “old school” approach to personalisation of taking tagged and classified content, giving it context and relevance, and creating user segments didn’t work for them. It took a lot of effort for them to try to tag all their content and it’s something they haven’t been traditionally good at. Also, depth of content was limited beyond core news topics. So anything specialised might only have three to four stories in a given month.

“That’s really not enough to be able to produce a personalised, segmented approach for presenting content to users,” Sriram said.

Another issue is in India there’s a high reliance on Google, so a lot of their users are anonymous and collecting data over the years wasn’t a strong point.

How technology is helping

“In today’s world of AI, ML, and GenAI, we can actually go from user segments to individual users quite easily. The way we are looking at it is every single user comes into contact with us at various points of interaction,” Sriram said.

For every point of interaction, they’re creating new points of interaction with generative AI-driven content. It could be through the home page, a push notification, or an e-mail news digest for example.

“The good news that’s happened for us because we were late to start, we’ve kind of been able to leapfrog some of the hard, heavy lifting work that others had to do to build their own personalised recommendation engines,” Sriram said.

Now there’s an app for all of it. The Hindu experimented with a few options and found a few different ones to serve different purposes.

“If you’re in your early days of your journey in personalisation, what you should be thinking about is how to not spend a lot of time and effort into building it but use what’s available readily,” Sriram said.

The downside here is that since they use a third-party, there’s only so much tweaking and fine-tuning they can do. With a home-built product, nearly everything can be personalised to the company that built it and uses it.

Right now, The Hindu is using Artificial Intelligence and machine learning technology, teams are starting to look more to generative AI options. 

“I suspect once we have that, there will be much more contextual accuracy as far as our recommendations are concerned,” Sriram said.

The results

One of the big successes The Hindu has seen in their personalisation journey is with personalised push notifications. Before personalising, their push notifications were automated and random.

The Hindu created a personalised recommendation engine that uses a combination of behavioural and collaborative algorithms, sending notifications at specific times of day they know users are more likely to open them. 

“Since we did that, we’ve seen a dramatic increase in click-through rates on notifications,” Sriram said.

In fact, with app users, the click-through rate increased eight times over.

The Hindu also personalised its article page and saw both session duration and CTR increase. Before personalising, they had very low engagement rates and users weren’t reading on to other recommended stories. Now, a balanced algorithm is used.

“The next story that you read is likely to be on a similar topic to what you’re reading currently, and what we noticed here was a 10% increase in session duration,” Sriram said.

This again is a result with the mobile app, but even more impressive is the 200% increase in click-throughs on their Web site.

The Hindu has also been able to create new points of interaction to enable discovery. They created a trending feed in the app that optimises stories that have been gaining traction with readers and are of particular interest to the user. 

“This is something that has driven us about a 20% increase in pageviews in our app,” Sriram said. 

Even The Hindu was surprised to see some of the successes. Monthly active users increased 2.5x and pageviews increased almost 4x over a period of eight to nine months. 

“One of the key objectives that we set out to achieve here was to get more stories seen by readers,” Sriram said. “We’ve seen more known users coming into our system, registering, and us providing better quality recommendations to these known users.”

Instead of one-third of the content being seen like before, more than half of it is being seen by 95% of known users.

“Best of all, our subscription teams are happier,” Sriram said. “Forty percent of subscriptions now happen in our app and probably a bigger share of revenue from subscriptions also come from our app.”

What’s next?

Besides the GenAI-based recommendations, The Hindu is looking at using the same personalisation algorithm to curate news digests that go out as e-mails as well as WhatsApp. They’re also working on a new, short summary style format in the form of article summary teasers for their publication Business Line.

Advice Sriram has for other media companies looking to start their personalisation journey?

“Something we discovered through this process, we did have a version where we asked users to pick and choose what they wanted to do,” Sriram said. “One thing we realised is that’s not something that works because users will tell us what they think they want to read, but then they go and read something completely different. It threw all our systems out of whack, and it took us a few months to work our way back from that to where we are today.”

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .