Jabong is an India-based online marketplace, connecting sellers and buyers across the country. It’s a one-stop shop with a widespread user base, and offers users the option to shop via mobile apps or a website.
An e-commerce case study
How Jabong uses machine learning to proactively block fraud rings
Saved each month
Better data on linked users
Told from the perspective ofVarun Sabu Verghese,Risk Team Manager
“We had a fraud solution before Sift Science, and had simple rules to block bad users, but customers were still creating fake accounts with false information. We needed Sift Science’s machine learning to prevent returning malicious customers, not just react to past fraud.”
A leading marketplace in a crowded market
A leading Indian e-commerce marketplace, Jabong offers fashions for men, women, and children from retailers and warehouses around the country. Whether a shopper wants local styles or top international brands, Jabong is committed to creating an amazing experience for its customers with fast and easy checkout, a variety of payment methods, and around-the-clock customer service.
Jabong currently delivers to almost all Indian post codes. Their business model is B2C, connecting retailers with shoppers via their inclusive and innovative marketplace platform. With hundreds of thousands of shoppers visiting Jabong monthly via their website and native apps, Jabong is setting itself apart in an increasingly crowded market by offering regular promo codes and sales, 2,000+ fashion brands, and over 200,000+ styles.
Spotting the sneaky users
Jabong is a lean operation, despite having a sizeable chunk of market share. With approximately 1% of the overall company working on fraud, scaling their efficiency as the business continues to grow is a challenge. When combined, the many conveniences offered to shoppers can create a unique vortex of fraud for the team to handle.
While the most common type of fraud that Jabong battles is fake orders using “guest checkout” with no account or user information attached, fraudsters have employed a variety of other creative behaviors to scam the company. Among them are:
Return fraud, where a malicious shopper returns a different item than what was ordered
Prepaid fraud, where a stolen payment method is used to place an order
Voucher misuse, where promo codes or coupons are abused
These various fraud and abuse types eventually led to the creation of a rules-based system to alert the Jabong fraud team in real time. The rules system required the building of specific parameters and structures for escalating issues to review teams. Nonetheless, as chargebacks on prepaid orders hovered at 0.4-0.5% – and blocked users kept returning – Jabong knew that they had to get ahead of those fraudsters circumventing the rules system and returning to the marketplace.
Machine learning and preventative analytics
In order to get operational costs down while continuing to maintain an excellent customer experience, Jabong decided to try out Sift Science, a machine learning-based solution that could identify returning bad accounts. With a highly tailored and hands-on integration process led by Varun Sabu Verghese, Jabong’s Risk Team Manager, they were able to begin learning on live data to identify fake orders.
Within two months of becoming fully integrated with Sift Science, Varun and his team successfully trained the global model and began to utilize the network connection view to spot rings of fraud.
“Sift Science’s user interface is very intuitive, and we’re able to use the console to investigate and track repeating fraudsters and identify fake orders by pattern recognition.”
Keeping a pulse on fraudsters
For Varun and his team, the greatest benefit of using Sift Science’s machine learning solution is the ability to use a battery of signals to identify and monitor returning users, regardless of whether they use guest checkout or try to mask their identities in other ways. Specifically, device fingerprinting keeps an eye on known fraudulent individuals, while the connected users visualization allows Jabong to spot linked users and determine their probability of fraud.
While Jabong has saved $15,000 in monthly revenue with Sift Science, the ability to stay ahead of bad users is priceless. Next for Varun is putting automation in place through Sift Science, allowing his team to spend less time on what little fraud they have and instead to ensure that the platform is risk-free in the future.
“Sift Science’s data visualizations are very helpful in identifying large sets of fraudulent users. Address patterns have become much easier to detect and we can focus on specifics instead of worrying about a broader set of fraud.”