Business solution examples

CASE 1

AI Invoice Reading Robot

Reduce costs by automatically reading bills with AI.
Dramatically streamline operations through data linkage to ERP and accounting systems.

Case overview

We want to automatically read invoices with different formats. We want to reduce the cost of the administration department, but how can RPA do this in cases where a large degree of accuracy is required in the work?

Background / Challenges

  • In the current situation, several people are needed to record invoices.
  • We want to automatically read invoices with AI.
  • Since each company's invoice format is different, we want to deal with each as automatically as possible.
  • We want to import the result of each invoice reading into the expense system.

Results / Outlook

The AI invoice reading system's accuracy has reached a high level of 99.6% in English and 99.5% in Japanese. (※However, this can be affected by the image quality of the invoice)

Our solution

Development of an AI invoice reading system

Invoice formats vary widely among companies, which is one reason that OCR, a converter of handwritten and printed characters to digital character codes, does not work well in this application. By analyzing the format of the invoice and setting it as a template, it is possible to read repeat clients' data faster and more accurately.

  • Import invoices with scanner and make them into files
  • Analyze the format of the invoice and make it ino a template
  • Make the contents of the invoice into a text data
  • Humans check the contents of bills and data
  • Create the data files and link them to the accounting systems
  • AI's learning ability improves reading ability
How to separate sentences in invoices automatically (auto segmentation)
How to separate sentences in invoices automatically (auto segmentation)

With invoices, characters are not arranged in order, but rather the billing source and order details are laid out in columns or in a table. Auto segmentation separates image areas and character areas, analyzes how these sentences are arranged, and decides the order of character recognition automatically.

Automatically convert to text, then a human checks the contents before database linking happens
Automatically convert to text, then a human checks the contents before database linking happens

The data, which has been quickly converted into text, boasts accuracy numbers of 99.6% for English and 99.5% for Japanese with AI alone, but when the final step of a human check is added in, those numbers improve even further. With this done, complete and usable data can be linked with ERP or the accounting system database.