Urban Innovation Japan is Japan’s first program for collaborative development between local government and start-ups. What does that look like in a real life setting?
In Kobe City, staff visually check and process receipt data for 2.5 million cases a year. These claims often include errors and checking for errors takes a considerable amount of human resources.
After organizing and analyzing the work performed by the Kobe City staff, Monstar Robo created a tool and conducted a demonstration experiment for the purpose of making the check work shorter and more efficient.
Recently, Kobe implemented the RPA* tool created by Monster Lab, Inc., called Monstar Robo, and between July 2, 2018 to October 31, we experimented with the effects of the change. Our modeling found the receipt checking work to be reduced by up to 459 hours a year.
In cooperation with Monster Lab, we verified automation of error check work in medical treatment months outside the period covered by the grant. Including:
We plan to further automate operations by incorporating the AI engine of FlyData Co., Ltd., which is another UIJ-adopted company, and aim for full-scale introduction next year.
RPA is tool that can give instructions to software-type robots and automate mouse and keyboard operations performed by humans on personal computers. It is being introduced as a tool to automate and streamline routine tasks.
In another of several Urban Innovation Japan projects adopted in the past year, Kobe City recently introduced AI technology from FlyData, a Silicon Valley-based company, in order to streamline checks on receipts. Proof-of-concept and effect verification was carried out from July 2, 2009 to January 31, 2018. We then reported the result of this proof-of-concept, an advanced approach which introduced AI to the receipt check work done by staff as a rule of thumb.
By learning the tendencies of past number input mistakes, we created an AI model that infers the right numbers from incorrect beneficiary numbers. The effect of contributing to the efficiency of business was verified by outputting the inferred results as a list of correct candidate responses and using it to supplement check work of the actual responses.
It’s been proven that introducing check results into the field can streamline check work. By increasing the data available for inference by AI, we expect to continue to improve estimation accuracy. Eventually, we expect to develop this technology to include other checking operations.