Reducing risk and streamlining fintech identity verification with AI name screening
Fintech company KOHO provides Canadians with a better way to manage their personal finances through their full service spending and savings account, which issues prepaid reloadable cards that empower Canadians to quickly receive funds, save, easily pay bills, and so much more. Users can receive their payroll direct deposit or load money on their card via a phone app. KOHO’s business depends on ensuring the right customer receives the money in a timely fashion, which means payee names must be verified against KOHO customer account lists.
Tens of thousands of times a day, the KOHO system must match names in real time for e-transfers coming through Interac (a funds transfer gateway). Interac enables a KOHO account to be loaded in less than a minute by SMS or email. In addition, payroll direct deposits occur in batch mode via KOHO’s banking partner, People’s Group.
The problem is, sometimes the name on the KOHO account differs slightly from the payee name. While the KOHO account holder might be “Rebecca Hockenbury,” the name from the partner bank could be one of these:
- R Hawkinberry
- Rebecca C Haukenbury
- Mme. Rebecca Hockenbury
- Mlle. Rebecca Hockenbury
- Hockenbury, Becky
- R Hockenbury
- Rebecca H
- Hockenbury R
- Mme. Hockenbury Rebecca
The legacy KOHO system used a customized open-source library whose code was complex and hard to maintain, and results were full of errors. Matches that should have been found were missed, and names that were not a good match were labeled matches. Consequently, the matching process required a lot of manual review, which slowed down payments.
“Every time that we reject one of these payments because of name matching issues, a person on the KOHO risk team is manually reviewing the identity of the person,” Yan Matagne, Senior Manager in Payments at KOHO, said. “Imagine you’ve loaded $5,000 to your account. If the funds don’t show up immediately, that’s a scary feeling. A customer might walk away if that was their first transfer. Our goal was to speed up and automate this review for a better customer experience.”
On a daily basis, over 900 pre-authorized payments required manual review. To bring this number down, KOHO started to look for a way to automate name matching.
KOHO’s top requirements for name matching were:
- Cloud-based deployment
- Accelerated name matching in real time and batch mode
After a broad search and numerous proofs of concept,, KOHO chose RosetteⓇ by BasisTech.
Rosette was available as a full-featured text analytics platform in the cloud or on-premises, and it was possible to license just the name matching feature instead of being tied to a monolithic data mining platform.
“Rosette was the clear winner for us because it increased our identity verification accuracy by 20% right out of the box, without any configuration,” Matagne said. “The better performance in accuracy and speed, and the simplicity of an API call to the cloud make Rosette worth it.”
Rosette returns a match score from 0 to 100% for every pair of names, which allows KOHO to set a match threshold — such as 60% or 70% — to qualify two names as a match. Names that don’t find a matching account that meet the threshold will be routed for human intervention.
“Rosette has been running 24/7 for a year now, without any trouble or errors in our system,” Matagne said. “Rosette returns match scores in under 2 or 3 milliseconds. Globally across all our requests, we are getting an answer in less than 200 milliseconds. In real time, that is pretty good, to be honest. We want customers to get their money as fast as possible or in less than a couple of minutes, so response time for calls to Rosette were really important.”
Matagne added that although KOHO only needs to match names in the Latin script at present, Rosette’s scalability and ability to match names in different languages and scripts were further reasons to go with BasisTech.
Impact: Reduced manual labor and increased accuracy
As a result of integrating Rosette into KOHO’s business processes, the percentage of automated identification matching has increased significantly and reduced manual labor review by 50%. Pre-authorized payments requiring review dropped from an average of 930 per month to about 430. Before Rosette, 10% of direct deposits were delayed and reviewed manually. With Rosette, that dropped to only 5%.
“We reduced 25% of the manual labor of our risk team, one full-time equivalent seven days a week,” Matagne said. “And with the accuracy of our new name matching system, we were able to create new features like authorizing our customers to get third-party transfers from out-of- network.”