How we boosted acceptance rate and lowered fraud for crypto exchange Ari10

Ari10 (formerly Bitcan) is a crypto exchange that allows for instant cryptocurrency transactions of up to USD 13,500. The most frequently chosen payment form in this exchange is the so-called pay by link. Unfortunately, the simplicity and convenience of pay by link do not always go hand in hand with security. Therefore, the biggest challenge Ari10 faced was increasing the security of transactions, especially in this area of payments.

We proposed a proprietary script embedded on the client’s website, which has been extracting over 5,000 attributes about each Ari10 customer -fingerprinting their device and checking their behavior to spot any repetitive patterns and abnormalities that may imply fraud.

We closely observed and analyzed the traffic to fuel the first ML model with proper data and to train it. That allowed for not only fighting fraud cases with much better accuracy, but also reducing the number of good users blocked by the initially implemented rigid rules. This example again proved the superiority of ML models over static rules but also showed that their hybrid use has its benefits depending on the situation.

Currently, models are constantly refined and modified by a dedicated Data Scientist, based on the acquired knowledge and close cooperation with the client.