Layer 1 copy

Sift’s ability to compare data on its own without needing an investigator to query is so powerful. We get new insights all the time without Dwolla’s team having to think of specific needs.

Ryan Hodge,Financial Intelligence Unit Director


  • A mobile and on-demand payments network

  • Serving customers in the U.S.


  • Preventing bad transactions or risks losing those funds

  • On-demand business requires rapid decision-making


  • Custom fields that tailor Sift Science to Dwolla’s unique needs

  • A quick integration of 2-4 weeks of part-time resources


  • In 2015, fraud dropped from ½ to 2/10 of a basis point

  • No need to hire additional fraud analysts although business continues growing

  • Cut the average amount of time the team spent on fraud


Making cents of payments

Dwolla is a payments network that enables its users to more quickly and frictionlessly make bank transfers via its powerful APIs, while supporting easy bank verifications. By securely connecting with U.S banks and credit unions, Dwolla allows for safe bank transfers at the speed of modern business – which is to say, fast. Founded in 2010, its nimble team currently only serves customers in U.S.


Declining bad users

As anyone in payments can tell you, building trust and protecting your customers are tantamount. Fraudsters would love to make fast work of unsuspecting payments companies and take advantage of the transactions and user data that flow through such sites daily. Dwolla’s challenge is even more pressing, since its service works on demand, offering a quick alternative to staid brick-and-mortar alternatives. On average, it takes 3-60 days for unauthorized bank activity to hit Dwolla’s records as a chargeback, so preventing fraud before it happens is crucial to building a sustainable network.

Sift has given us breathing room as we continue to grow. We don’t need to hire additional analysts – we have fewer alerts and can remove the clunky rules.

For Dwolla and its banking partners, fraud comes in various forms, the most damaging being transactional fraud. Monitoring fraud from stolen accounts, scammy payments, or fake bank information is a fulI-time job for their team of experienced analysts. While the problem was still small enough, Dwolla utilized an in-house rules-based system, managed by their Financial Intelligence Unit, who measures fraud as fraud losses to volume. For the business, fraud is cyclical, and as the up-cycle in their annual business approached and their fraud rate hit 0.5 of a basis point, the team sought a more scalable solution.


Machine learning to save money

Sift Science’s machine learning-based solution and real-time updates offered a flexible tool for Dwolla’s team, and the promise of a quick integration was immensely appealing. To test Sift Science’s effectiveness, the team began with a 2-day, low-key integration using just custom data fields based on Dwolla-specific information, instead of leveraging the global model as well. Sift Science delivered such useful results based on these purely custom fields that Dwolla’s Financial Intelligence Unit Director, Ryan Hodge, decided to commit to a full integration. He felt confident that a deeper implementation would yield even more accurate and actionable insights for Dwolla.

Machine learning has cut our fraud management time by very quickly recognizing the worst of the worst accounts and identifying those for the investigator. Without the Sift Science solution, multi-factor and cross-account intelligence would be so much harder.

The full integration took 2-4 weeks of part-time engineering resources, including many additional custom fields, but the Financial Intelligence Unit saw immediate success. Within days, Sift Science was able to shorten the average amount of time the team spent on fraud by efficiently identifying the definitively bad users to analysts. Ryan plugged Sift Science directly into its existing in-house fraud management system used to queue suspicious transactions via webhook. This seamless merging allowed the team to easily spot and investigate – via the web console – any users that were flagged by Sift Science’s scoring system.


Prevention pays off

Within the first week of being fully integrated, the Financial Intelligent Unit saw accurate, actionable results. By labeling and training its machine learning system, Dwolla rapidly improved its model, and its insights have only grown since then. The flexibility of layering Sift Science with its existing business rules means increased efficiency without losing time constantly updating its hard-and-fast rules. Dwolla's fraud rate was halved by using Sift Science, dropping from 0.5 of a basis point to 2/10 of a basis point.

Sift quickly gives us concrete rulings on when it sees fraud, so we don’t have to manually investigate.

For Ryan’s team, the biggest net-gain with using Sift Science is the solution’s ability to cross-link users, highlighting those accounts with shared attributes. The Sift Score also offers an efficient, at-a-glance way to measure suspiciousness and allows the fraud analysts to quickly assess riskiness without digging through too much data.