Overview

Buying and selling more effectively

Carousell is a Singapore-based peer-to-peer marketplace, where users can buy and sell effectively and easily, and sellers and buyers can connect privately. The company has both an app and a desktop site. In addition to Singapore, Carousell operates in Malaysia, Hong Kong, Taiwan, Philippines, Indonesia, and Australia, encompassing 250 million listings and 70 million transactions across all markets.

Challenge

Fake listings and the need for proactive prevention

As Carousell began to scale, they started to see fraudsters posting fake and spammy product listings for products that either arrived to the buyer not as described or never got delivered to the buyer at all. Carousell didn’t have a way of proactively preventing these listings and relied on user flags to spot and remove them. This meant that these listings not only posed a threat to good users until they were eventually removed but threatened to sully the reputation of the platform, as well.

Repeat fraudsters were also finding ways to get back onto the platform even after Carousell deleted their accounts, and continued to post abusive, fake listings with their new accounts. Carousell limits the number of accounts a user may have to a maximum of two, but fraudsters were creating multiple accounts and Carousell was finding it difficult to keep track of them all. Carousell was using a rules-based fraud solution, but it was time-consuming to have to jump in and change rules every time fraudsters changed their tactics.

As the number of users grew and Carousell expanded into new markets (each of which had their own unique fraud trends), it was becoming difficult to scale. They needed a way to stop fraud before it made it onto the platform, and keep fraudulent users away for good.

Solution

Automation and machine learning become game changers

When Carousell’s rules-based fraud vendor suddenly got acquired, they needed a new solution, fast. They considered a number of fraud prevention vendors that offered rules-based and machine learning solutions but chose Sift Content Integrity because they were impressed by the intuitive, user-friendly Sift Console and the ease of Workflows.

After integration, it took only a couple of weeks for Carousell to start seeing results. Using Workflows, they were able to automatically block users once they exceeded a certain Sift Score (risk score based on behavioral attributes). They also automated account remittance for users who had more than one account, many of whom were using those accounts to commit fraud. Additionally, Carousell implemented a Workflow that would hide flagged listings while they awaited manual review, which helped keep potentially fraudulent listings out of sight and away from good users.

Carousell also benefitted from the personal assistance of Sift’s support engineers, who provided Carousell with personalized, human interactions when they needed them. No chatbots, no copy-and-paste replies.

Tan Su Lin
“The information from the global network really helped us to scale. Previously, with a rules-based system, we were constantly playing a game of catch up, whereas with the machine learning model we could just let it learn and continue to catch fraud on its own.”

Results

Free to focus on growth, not fraud

Since adopting Sift, Carousell is detecting 23% more fraudulent users and 10.26% more fraudulent listings, has achieved 2x ROI, and is saving over $350,000 a year. They’re successfully keeping fraudsters off-platform, and preventing them from finding ways to come back thanks to the accuracy of their model. Rather than reacting to fraudulent listings only after they’ve been flagged by users, Carousell is preventing them from being posted in the first place.

As a result, Carousell has been able to scale without dividing their time with fraud management and doesn’t need to devote a human team to manual review. Workflows are automating decisions and handling the bulk of the fraud prevention, leaving Carousell’s fraud teams to focus on helping the company continue to grow into new markets and provide the best experiences for their buyers and sellers.

Tan Su Lin
“It’s been incredibly helpful having Sift identify risk signals and risky users, rather than having a human do manual review every day. We’ve been able to scale in ways that we couldn’t with a rules-based solution and create a secure environment for our users.”