Japan may be a high-tech country, but it still loves paper and old-school administrative processes. Manually-processed application forms, invoices and financial documents all waste valuable worker hours. But what if deep learning could be used to automate these and other forms of unstructured data?

That’s exactly what one Tokyo startup is doing. Cinnamon, a company dedicated to “eliminating repetitive tasks” in the workplace, has developed Flax Scanner, a product it describes as a “cognitive document reader.” It scans digital or printed sources including PDFs, Word files, handwritten notes and faxes, extracting key points from contracts, invoices and other documents with an accuracy of over 99%.

Flax Scanner works not only with documents with uniform layouts of data, such as driver’s licenses, but unstructured ones such as invoices. The software can also extract data from complex documents such as contracts.

Removing repetitive tasks from the world

“After I had my children, I began to wonder what kind of work they would do, and whether they will still have to do uninspiring, repetitive tasks like we do today,” Miku Hirano, co-founder and CEO of Cinnamon, told attendees of Rakuten Technology Conference 2018 at Rakuten Crimson House in Tokyo. “Our mission is to remove all the repetitive tasks from the world and to make people happy.”

During her presentation, Hirano discussed how she studied AI and complex networks at the University of Tokyo and sold her first company, Naked Technology, which was devoted to smartphone middleware, to Japanese social network Mixi. She described how Cinnamon, which was founded in Singapore, nearly went bankrupt after several failed ideas, but went with AI-powered administrative automation and has since raised $17 million in debt and equity financing. One of its advisors is prominent Japanese scientist Hiroaki Kitano, president & CEO, director of research of Sony Computer Science Laboratories. Staffed by over 100 people, Cinnamon is now headquartered in Tokyo and has offices in Silicon Valley, Vietnam and Taiwan.

“Japanese work too much and there is so much mendokusai (troublesome) repetitive work for office people,” Hirano said in an interview on the sidelines of the Rakuten Techonology conference in 2018. “If AI can do this work instead of people, they can focus on creative things and spend more time with family and enjoying hobbies. That’s our mission.”

“Japanese work too much and there is so much mendokusai (troublesome) repetitive work for office people,” Hirano said in an interview on the sidelines of Rakuten Technology Conference 2018. “If AI can do this work instead of people, they can focus on creative things and spend more time with family and enjoying hobbies. That’s our mission.”

Empowerment through automation

“Our definition of machine learning engineers is people who are able to program deep learning from scratch,” Hirano said. “That’s what you need if you really want to increase productivity for people, because using general AI alone does not give you sufficient accuracy to realize actual cost reductions.”

The scope of businesses that can benefit from office automation tools is enormous. Hirano said growth in enterprise AI has greater potential than the breakneck expansion of consumer smartphones seen from 2010 to 2014.

42% of finance activities can be fully automated in the future, and Japan ranks high on a list of countries in terms of automation potential, Hirano said, citing a study by McKinsey Global Institute.

“Japanese work too much and there is so much mendokusai (troublesome) repetitive work for office people,” Hirano said in an interview on the sidelines of the conference. “If AI can do this work instead of people, they can focus on creative things and spend more time with family and enjoying hobbies. That’s our mission.”

“We want to increase our product lineup, even though human work cannot be entirely replaced by automation,” Hirano said. “The very basis of my motivation is to bring humans to the next level, the future.”

“We want to increase our product lineup, even though human work cannot be entirely replaced by automation,” Hirano said. “The very basis of my motivation is to bring humans to the next level, the future.”

Changing how businesses work in Japan and abroad

Cinnamon has been developing other products besides Flax Scanner. It has developed a natural language-based chatbot called Scuro Bot, a recommendation engine called Lapis Engine that can suggest products such as real-estate listings by utilizing vectorization of user and product information, and a speech-recognition platform called Rossa Voice that can recognize unique keywords such as product names.

The company has big plans to change how businesses work in Japan and abroad. It wants to grow its team of 70 machine learning engineers to 500 by 2022, and Hirano made an appeal to conference attendees to apply for open positions at her company.

“We want to increase our product lineup, even though human work cannot be entirely replaced by automation,” Hirano said. “The very basis of my motivation is to bring humans to the next level, the future.”


For more information about the other speakers and events at Rakuten Technology Conference 2018, visit here.