Scaling use cases, increasing productivity
Applying Intelligent Document Processing (IDP) to Japanese financial documents
75%
reduction in manual workload
50 sec
avg. manual validation time per doc
27.5%
time saved per document
Master Trust Bank of Japan – who are they?
Master Trust Bank of Japan (MTBJ), part of the Mitsubishi Financial Group, Inc., is one of the largest trust banks in Japan specializing in asset servicing. MTBJ has 40 years of experience as a trustee and administers 2630 funds, with assets under custody totaling 571 trillion yen (over 4.35 trillion US dollars).
Under the corporate motto “Be professional”, MTBJ aspires to the highest financial global standards and strives for the most responsive and efficient service possible.
The starting point: “pure-play time saving”
MTBJ handles an enormous amount of sensitive financial documentation, from Trade Instructions, and dividend notices to tax returns. Accuracy is essential due to the regulatory nature of the documentation and compliance requirements.
MTBJ had a total of 40 staff members to manually enter data for further processing. Each document could take up to 1.5 hours to process depending on the complexity. Furthermore, language limitations led to the bulk of the workload falling on the shoulders of a few key staff members.
The initial drive for the project was “pure-play time saving” in other words improving operational efficiency by speeding up complex data capture. However, there was a lot of variability in the document layouts, with some documents several pages in length, which made it difficult for conventional OCR solutions to extract data.
MTBJ spent 18 months searching for an automated solution that could “read” complex Japanese documents and conducted a trial with a total of 15 vendors. MTBJ eventually selected Rossum as it was impressed with Rossum’s superior AI, user-friendly interface and ability to handle a multitude of different document types.
Training Rossum to “read” Japanese
The initial step of the project was to train the Rossum Intelligent Document Processing (IDP) platform to understand Japanese, a complex language that uses three different alphabets: Hiragana, Katakana, and Kanji.
Rossum combined an open-source Japanese OCR module with our own proprietary OCR technology to get the best performance, and injected this compatibility layer into our main AI engine. The advantage to this was that the AI engine could leverage its current knowledge and apply it to Japanese documents, accelerating the time to value as it was not starting from scratch. Once the compatibility layer was built it took only 1-2 days to train the AI. Time to value was a matter of seven weeks from project kick-off, during which time we also localized the user interface (UI) to Japanese.
We also used sophisticated parsing logic to extract Japanese date formats including the year eras distinct to Japan. Rossum is one of the few IDP platforms to support this, thanks to the modular approach of our architecture which allows us the flexibility to augment the platform with new AI capabilities.
Scaling to new use cases
As part of the project we integrated Rossum with their proprietary RPA system to enable the end-to-end flow of data communication with their other systems of record which led to an expansion in use cases.
Over time the MTBJ team increased the number of use cases to over 90 different financial document types used by 10-15 different departments. Rossum’s cloud-based architecture was easily able to scale with MTBJ’s needs while Rossum’s Services team continuously monitored the AI in order to maintain the accuracy of the models.
To date, Rossum has scaled to handle 100,000 documents per year for MTBJ. It now takes less than 1.5 minutes for Rossum to extract the data from documents while full processing by MTBJ’s internal RPA system now takes less than 10 minutes, a huge leap forward compared to the 1.5 hours per document previously.
Outcomes
MTBJ has experienced a huge improvement in productivity as its staff has been freed up from data entry work to focus on more value driving activities. The Rossum IDP platform has reduced the manual workload by 75% and improved the time spent verifying and validating documents. Staff now spend only 50 seconds on average validating each document which is a time saving of 27.5% per document.
- 75% reduction in manual workload
- 50 sec avg. manual validation time per doc
- 27.5% time saved per document
“With Rossum, we see impact early on: from reducing overhead costs to increasing the speed of commercial transactions and significantly reducing the risk of exposure. The solution has a positive influence on both internal users and our clients.”
Ryo Kawaguchi, Manager at Master Trust Bank of Japan
The future
Rossum has scaled to over 10 international departments across Japan and the UK. Rossum and MTBJ are now working together to explore a number of new use cases across the Group including Unique Transaction Identifiers to comply with new Japanese financial regulations.
With Rossum’s help, MTBJ is leading the way to “Be professional” by harnessing intelligent document processing to improve business relationships as well as processes and provide the most efficient service possible to its customers.