Veem is a fast-growing fintech that makes international money transfers easy and affordable for small and mid-sized businesses. Since customers depend on them to transfer funds securely, Veem has invested heavily in dedicated teams and sophisticated technology to guard against fraudulent transactions. Veem’s robust growth sometimes strained their systems, though.
Seeing a potential opportunity late last year, Suhas Maskar observed that “as Veem was scaling more on the growth path, it was slowly becoming too much to handle marking an account fraud or safe. With my background in data sciences, I had recommended that we address this by automating most of these fraud pattern detection via machine learning”
Suhas got in touch with his counterpart on the risk team, Vishal Joy, to identify what an automated system might look like. Vishal agreed that Veem’s “people, process and procedures provided good coverage against fraud.” And he “realized that the need to swiftly respond to everchanging fraud environments [meant] Data Science models would be a key differentiator.”
The new system would need to quickly tag accounts as high risk, low risk, or in need of manual review – and take appropriate action. It would have to work seamlessly during onboarding with minimal impact to customer experience. And it would need to learn from its mistakes in order to avoid adding work to the already stretched teams’ queues.
With requirements in hand and new software for model development and deployment, Suhas got his team to work, relying on the sophisticated data environment that he’d helped build over the years to provide comprehensive, accurate inputs – and effective models.
The project progressed quickly until the team member that had been building the fraud models left Veem. That’s when Suhas brought in Slick Predict. Suhas remembers that “we were worried about falling behind on our timeline for the rollout of this model. This is where Slick Predict stepped up big time and we were able to hit our targeted rollout.”
Slick Predict streamlined data intake, finalized model training, built out reports, and deployed an automated model retraining process – while incorporating the Risk team’s suggestions. This collaborative approach has been invaluable to Vishal, who notes that “Suhas and Ted have been excellent partners in this journey, where there has been a constant dialogue and exchange of ideas enriching the work every step of the way.”
Within a few weeks’ time, Veem had a fully functional, AI-based system scoring thousands of new accounts and, importantly, only pulling in risk analysts for manual reviews a fraction of the time. Of Slick Predict’s modeling additions, Suhas felt that the “retraining paths that would keep improving the model as we go along […] was the most critical piece of fraud that we are very proud of.”
Not only is the model extremely good at identifying fraud, exceeding its targets for real-world accuracy by 10x with a 0.4% false positive rate, it educates itself with fresh data each week – boosting performance even higher against the baseline.
The automated anti-fraud system has been up and running for nearly a year as of this writing (September, 2022) and the benefits have been enormous. The rate at which accounts are automatically verified without a manual review have increased 300% since implementation. The rise in automated verification – without an uptick in fraud – has meant far less strain on the risk team.
In Vishal’s words, “the fraud onboarding model has had a real impact on reducing the amount of time Risk Ops needs to spend reviewing accounts. This capability has opened up other avenues in our mission to identify fraud.”
If your business needs to transfer funds quickly and safely across the country or around the world, please visit veem.com.
Suhas Maskar – Director of Data and Analytics, Veem
Ted Coxworth – President, Slick Predict
Vishal Joy – Director of Risk & AML, Veem
Sign Up Schedule a demo