In any industry or line of business, if the user experience is anything but excellent, customers are likely to switch to a competitor.
For several Indian businesses, this was a reality they were keen to avoid, prompting a swift re-evaluation of their data strategy and infrastructure.
Take media giant Zee, for example, which runs over 100 TV channels and an over-the-top (OTT) platform with more than 20 million monthly active users across 190 countries worldwide.
According to Kishor Krishnamurty, then-CTO of Zee, the company’s Android and iOS apps were rated 3 stars and 2.8 stars respectively around the time he joined. Meanwhile, the media giant’s net promoter score (NPS) was negative.
“As Zee transforms its business to digital, it faces competition from new players because it’s an attention economy out there. We’re up against giants like Google, Facebook, Apple, TikTok, and Snapchat, so it’s a whole different ball game in the digital world. Consequently, media companies like ours worldwide are trying to figure out what can be disaggregated, and the OTT cost structure itself is very expensive compared to linear,” he said during Confluent’s Kafka Summit in Bangalore last May.
Cost reduction
To make the company’s digital platforms more agile, Krishnamurty and his team deployed Confluent on Zee’s heartbeat ecosystem. A heartbeat is a request fired at regular intervals during video playback, while the heartbeat ecosystem governs platform features such as resume watching, recommendation system, and ads preferences.
“When we needed to sample more heartbeats per minute, our costs started to rise immediately. So, we built a system using Kafka with KSQL and ScyllaDB, which significantly reduced the cost structure, making it no longer linear,” the then-CTO explained.
The ultimate goal, Krishnamurty revealed, was to move from a monolithic system to a more event-driven, distributed microservices architecture, a transition Zee is well ahead on.
“Our cost structure, particularly our daily burn from an OpEx perspective, is now 50% of what it used to be. So, when we examined other parts of the ecosystem, like subscriptions and payments — basically the entire order management — we moved those to a Kafka event-driven system,” he added.
Today, the company’s ZEE5 app is rated 4.1 on the Google Play Store and 4.7 on Apple’s App Store. Likewise, Zee’s NPS has also improved to 50 points positive, according to the CTO.
In addition to its heartbeat ecosystem, Zee also deployed Kafka on its CMS platform.
“When new content gets published, or if you want to update the title or description of existing content, or add a new subtitle or language, this entire communication happens over Kafka,” Krishnamurty said.
Zee’s video publishing system also leverages Kafka: “All the content that comes to us from production houses gets uploaded, and then we transcode that into various bitrates and codecs to support various devices. The entire orchestration of those processes happens using Kafka.”
Faster response
In the case of cloud-native digital lending start-up Kissht, the objective was clear: to provide consumer lending for the masses in India — 400 million people, to be exact.
“For new customers, the fastest loan approval in India back in 2015, when we started, was three or four days. So, we set out with the vision of reducing that time to under 10 minutes,” said Karan Mehta, the company’s CTO.
During the company’s first couple of years, many of its processes were still done manually. However, over the last seven years, Kissht has transitioned to a fully digital operation.
Although Kissht’s approval process was already averaging 10 minutes per new loan, the challenge arose when competitors began operating within a similar timeframe. Initially, Kissht’s system was built with a monolithic architecture, but it eventually transitioned to an event-driven model.
“In our system, the moment a customer accepts a loan, around 15 or 16 processes are triggered simultaneously. One process communicates with the management system that a loan has been accepted, another sends a welcome letter to the customer, and another updates governmental records,” Mehta explained.
Previously, Kissht operated with around 30,000 to 40,000 lines of code. After adopting Confluent’s event-driven platform, they now work with just 5,000 lines of code.
“This reduction meant less code to maintain, and as we kept adding more features to this event-driven system, it created less overhead for newer developers to contribute. That really changed the way we think about writing code,” the CTO continued.
Mehta added that while there are simpler tools available, the organisation has committed to using Kafka, with plans to invest in it for the next 10 years. Kissht aims to avoid using a solution that might need to be changed in a few years due to scaling limitations.
Better agility
YES Securities (India) Ltd. began their journey with Confluent six years ago and has since embraced a more modern operational approach, leaving behind their traditional methods.
The company uses Kafka in two key areas: first, by putting data onto Kafka streams and sorting it into different buckets, allowing various teams to access the data they need when they need it.
“The customer needs to be notified in the evenings, and the brokers also need to be kept in the loop. Additionally, we have a risk management area that assesses the risk before each trade. This requires continuous monitoring of market data to determine if the risk has increased or decreased based on new values. With Kafka as our central messaging system, managing these tasks becomes much easier,” remarked Kinjal Shah, CTO of YES Securities.
Secondly, Kafka powers the company’s new mobile app and web dashboard, enabling them to track customer behaviour and trading activity.
“We want to track how many customers are visiting our platform, how many are completing trades, and how many are passive versus active investors. Our goal is to analyse all this data so that by next year, we can provide near real-time analytics to our business,” Shah concluded.
