Shopee India: Building the Fraud and Risk Function from Scratch

Peeyush joined Shopee India as part of the founding team when the platform had roughly ten members and no established entity. One of the functions he was handed wholesale was fraud and risk, a critical gap for any e-commerce platform scaling at speed.

He started by building the team from scratch: hiring, defining the operating rules, and implementing them. The initial setup was retrospective, meaning agents reviewed the prior 24 hours of activity and took action on flagged accounts. As order volumes climbed, the limitations of that model became clear.

Moving to Real-Time Detection

Peeyush led the evaluation of whether the function was ready to shift to real-time fraud detection. The question was not just technical. Two things had to be true: the agent team had to be large enough to handle a four-hour response window, and their decision-making had to be accurate enough to trust at speed.

He designed a sandbox environment to test both. Rules were deployed in real time, agents were assigned cases live, and the team tracked both throughput and decision quality over two to three weeks.

Phased Rollout

Rather than flipping the entire system at once, Peeyush made the call to move rule by rule. The most critical rules went real-time first; lower-risk rules stayed retrospective until the team demonstrated it could handle the load. Each rule was stabilized before the next was migrated.

This approach let the function scale its capability without exposing the platform to the risk of a full cutover before the team was ready.