Layer 1 copy

An e-commerce case study

How Mercari fights all fraud and expands its business with one platform

Founded in Japan in 2013, Mercari is a mobile peer-to-peer e-commerce marketplace that has been experiencing massive growth in the United States. As they expanded into the new market, Mercari needed a preventative fraud solution that could holistically evaluate buyer and seller behavior to ensure trust in the marketplace and reduce chargebacks.

60%

Drop in chargebacks

Increased automation in 2 weeks

60%

Drop in fraud rates

Told from the perspective of

Puneet Shah,Senior Product Manager
“Payment fraud was the problem we needed to deal with immediately. Invite fraud was the next challenge. With Sift Science, we have a one-stop shop to fight these multiple types of fraud.”

Overview

A real-time shopping community

Mercari is a shopping app that encompasses the largest community of buyers and sellers, allowing easy e-commerce to happen anywhere at any time. This Japan-based marketplace enables real-time mobile commerce for anyone who wants to buy or sell anything. The platform connects with an easy shipping label creator for sellers, and is cross-category in its ability to empower business across many verticals. With over 10 million installs of the app in the U.S. alone, Mercari is a peer-to-peer marketplace that straddles 3 counties: Japan, the United States, and the United Kingdom.

Challenge

Growing across the world

As an agile startup working across geographies and oceans, Mercari is committed to an excellent customer experience while maintaining a sustainable business. Their U.S. efforts encountered many new challenges that the Japan venture had not previously seen. From promotional or referral fraud – with fake invitations to new users taking advantage of Mercari credit – to payment fraud and spam listings, Mercari found that the U.S. marketplace industry was open to various types of bad and costly behavior. In order to maintain the integrity of their marketplace, Mercari required considerable manual review hours to vet accounts and products, enlisting a lean team across San Francisco, Portland, and Japan.

Puneet Shah joined Mercari as a Senior Product Manager tasked with creating a seamless experience for customers while reducing the impact of these fraudulent users. With less than 3% of the overall company dedicated to fraud, Puneet and his team needed a scalable solution that could keep up with their growing customer base while predicting behavioral trends that they hadn’t had to deal with before.

“The rules-based systems couldn’t keep up with Mercari’s volume, and needed constant updating. We needed something better.”

Solution

Machine learning for smarter decisions

With their entry into the U.S. market – and the subsequent growth in malicious sign-ups – Mercari initially turned to more old-fashioned, rules-based solutions. Unfortunately, these outdated technologies couldn’t scale with Mercari’s astronomic growth, and the stagnant rules required constant updating, flagged too many false positives, and negatively impacted their chargeback rate. Puneet and his team decided that they needed a solution that could learn in real time and predict fraud, without relying on in-house fraud experts to write and maintain rules. Because Mercari’s fraud issue was wholly new and the marketplace industry presents challenges where the business takes on various roles and risks, the defaults suggested by other vendors didn’t align with Mercari’s unique needs.

Because the team at Mercari wasn't familiar with machine learning, a new tool required buy in, an extensive trial, and thoughtful analytics before its findings could be confidently acted upon. When Puneet decided to fully integrate and trial the Sift Science system, he decided not to backfill the machine learning model with historical data and instead let the algorithms learn on live data before assessing accuracy. The Mercari team found Sift Science’s API documentation to be clean and organized – especially when compared with other vendors – and the engineers had the solution integrated within a matter of days. Puneet planned on giving the models one month to crunch the data and begin adjusting to Mercari’s needs. Throughout the learning phase, Puneet was committed to sending good quality data via the API and Javascript snippet, to ensure that the learnings would be as actionable as possible.

“Our analysts were immediately drawn to the Sift Science Console – having a robust console to go to and investigate in was an advantage for us.”

Results

Scoring high with machine learning

Very quickly, Puneet’s team saw highly accurate learnings and analysts were reviewing cases in the Sift Science Console. Two weeks into integration, Mercari began using the Queues feature; Sift Science was turning out such accurate scores that Puneet’s team was able to confidently auto-ban high risk orders within one month of integration.

The efficacy of Sift Science has helped Puneet and the company’s team of analysts to become fraud fighting experts. Within 3 months of using Sift Science, Mercari’s chargeback and fraud rates dropped by 60%. Most importantly, however, is Mercari’s ability to focus its energies on growth and the optimal experience for its customers. Puneet and his team have a solution in place that scales with their business, ensuring a sustainable and trustworthy platform for buyers and sellers alike.

“Sift Science has really thought through the marketplace aspect and understands that we work with multiple relationships. Getting started with a dedicated marketplace integration guide gave us a sense of confidence.”

A network of trust

Online businesses face a variety of fraud and abuse threats. With over 6,000 sites on our platform, we collect, analyze, and learn from 100s of millions of events every day. Let’s fight fraud together.