How One of India’s Largest FinTech Players Builds AI Models 10x Faster

“DataRobot is an extreme developer productivity multiplier. We went from five days to less than four hours for each model created.”

Building the Future of AI in Financial Services

From collecting payments online to enabling the point of sale to accepting foreign currencies, millions of business owners across India and Southeast Asia rely on Razorpay for smooth customer transactions. Founded in 2014, the financial technology company has revolutionized digital payments in India and is now valued at $7.5 billion as it further expands its footprint in the region. 

In order to deliver the most impactful solutions to their customers at the fastest possible pace, Razorpay turns to AI to make sense of the company’s rapidly expanding dataset. 

Their small-but-mighty data science team leads these initiatives and needed to find new ways to maximize the value of AI without increasing headcount. They also needed an AI platform that would enable more team members to contribute to AI programs, whether they were experienced data scientists or business users.

“We’re more interested in empowering the existing workforce rather than hiring people with specific skill sets,” said Pranjal Yadav, Head of AI/ML at Razorpay. “We need to prepare people for the next generation of engineering.”

Noted for its ease-of-use for both data scientists and business users by industry analysts, DataRobot fit the needs of Razorpay’s ambitious team. 

Scaling AI Collaboration for Data Scientists, Engineers, and Business Users

For the ability to empower engineers and other users without data science backgrounds, the company chose the DataRobot AI Platform.

Beyond this core requirement, DataRobot provided multiple other key needs, including the ability to deploy quickly and critical bias mitigation and fairness capabilities. 

“DataRobot is a natural fit for us,” Yadav said. “It fits our key parameters and incorporates significant support capabilities.” 

Building AI Infrastructure in Days with DataRobot and AWS

Just as essential, DataRobot easily integrates into the company’s broader technology ecosystem, most critically Amazon Web Services. The combination of DataRobot and AWS enabled a fast start and gives Razorpay the infrastructure and confidence to tackle its generative and predictive AI initiatives.

“Close collaboration between DataRobot and AWS is critical to our success,” Yadav said. “Acquiring DataRobot via the AWS marketplace, we were up and running within a day or two. If this had taken a week, as it often does with new services, we would have experienced a service outage.”

“We are thrilled to collaborate with DataRobot and its customers,” added Subodh Kumar, Senior Manager of Technology Partnerships at AWS. “Together, we can deliver more robust, scalable, and trustworthy generative and predictive AI solutions to customers like Razorpay, empowering them to solve business challenges and make critical decisions quickly and efficiently and derive maximum value from their AI investments.” 

“Extreme Developer Productivity” Delivers Models 10x Faster

Ease-of-use with DataRobot means that data scientists and non-data science users can build and evaluate models quickly. They experiment in ways they never could before, even on complex challenges like fraud detection.

“It’s very simple, yet beautiful, for us to click and see models being trained,” Yadav said. “In that layer, we’ve found some really meaningful insights about our own data.”

In the past, models targeting the fraud problem demanded four to five days per model. With about eight models dedicated to the challenge, that added up to about a month’s worth of time. Not so, anymore.

“DataRobot is an extreme developer productivity multiplier,” Yadav said. “We went from five days to less than four hours to create each model.”

A Major Business Problem… Solved in 12 Hours

Razorpay’s AI use cases span insurance, payments, risk, fraud, and customer retention. In one of its toughest business challenges, the company insures merchants against fraudulent or undeliverable orders. Given the logistics of fulfilling orders, costs can quickly add up over thousands of transactions.

As the problem intensified, the Product Head brought the team together for a focused effort on solving it in a “war room” setting. They would spend 12 hours specifically troubleshooting fraudulent orders. With DataRobot’s custom capabilities, a small team designed a blueprint within a couple of minutes and completed a model by midnight that day.

“The next day we went live and blocked all malicious orders without affecting any single real order. It’s pretty magical when your ideas become reality that fast,” Yadav said. “In data science, timelines are usually weeks and months. Now, maybe we can turn around a solution in 12 hours.”

Once models are in production, the team monitors them within DataRobot. 

“There’s a lot of value added, especially for drift detection,” Yadav said. “That’s insanely valuable for us.”

In the highly regulated fintech industry, Razorpay also appreciates DataRobot’s built-in bias mitigation and fairness capabilities.

Sharpening Competitive Advantage with AI in Financial Services

At each step, the DataRobot team has been there to ensure that Razorpay users achieve maximum business value with the platform.

“Kudos to the DataRobot team,” Yadav said. “In terms of support and enablement, they’ve been excellent – especially working within the constraints of a fintech domain. The team is always willing to understand the problem, take a stab at it, come with proposals and timelines, and long- and short-term solutions.”

DataRobot also connected Razorpay with other fintech customers to exchange information and transfer knowledge.

That support will be essential as the company extends DataRobot within the organization – a step that will sharpen its competitive advantage.

“Our competitors are probably 10 times bigger than us in terms of team size. With the time we save with DataRobot, we now have the opportunity to get ahead of them. It’s a really good co-pilot for experimentation, idea exploration, and creating a working solution in no time.”