Moonoia
Meet Docbrain, The AI That Rethinks Data Extraction

The creation of conversational chatbots, self-driving cars, and manufacturing robots is proof enough that AI and ML are technological marvels. Apart from these commonly known examples, not a lot of concepts from these technologies find their way into production at the corporate level. Part of the reason is that ML and deep learning are complex and require specialized skills that are scarcely available.

In the document management space, AI and ML have a huge potential through deep neural network (DNN) models to extract information from document images including complex, unstructured data such as cursive writing—something that is impossible to achieve through old-style OCR solutions. Traditional OCR technologies can only read limited handwritten, deteriorated or unconstrained documents, leaving too much of the content to human operators to decipher. Manual validation again makes up for a time consuming and cost intensive job, does not guarantee the accuracy levels required today and is prone to many other errors.

During the past few years, neural network technology has revolutionized image and natural language processing, data extraction (capture), and document classification. Through a continuous deep learning process, layers of artificial neurons are being trained on millions of sample images to detect patterns and recognize assets that even knowledge workers cannot. Automating high-value, document-centric processes is crucial for gaining competitive advantage in this day and age, freeing up more time and resources to focus on core activities while ensuring that business decisions are based on the correct data. Moreover, human resources can undertake higher-value tasks while becoming more flexible.

Venturing into the unchartered intersection between data capture, DNN technology, and Platform-as-a-Service model, Moonoia a software company based in Brussles, Belgium created docBrain—an industrialized, multi-purpose, document-centric AI platform. The platform allows clients to train their neural network models while still offering ready-to-use, generic models, and capabilities such as language-agnostic cursive handwriting recognition, image quality enhancement or document classification. “We like to call this functionality ‘conceive, consume, and everything in between,’” says Geert Truyen, CEO of Moonoia. With several partner companies operating in Europe, the U.S. and the Middle East, docBrain is a foundation for redesigning entire business processes. It not only solves document challenges but also helps implement a future-proof AI strategy that drives sustainable, high-impact business value.

"docBrain is the only document-centric platform that allows you to train your own AI and create your own solutions while still offering powerful, off-the-shelf capabilities you can deploy right away"

From BPO Document Services to Proven AI Software

Commencing its journey as an agile scanning bureau for organizations struggling with paper-intensive processes in an uncertain, post-recession European business climate in 2011, Moonoia has tremendously evolved over the years. The company began building bespoke software solutions internally to solve the difficulty of processing an ever-increasing amount of unstructured, difficult to read and highly sensitive type of content: medical documents. Years of expertise in the BPO sector and extended R&D capabilities made it possible for Moonoia to develop proprietary software that met the industry's highest, newest standards, recognizing and classifying handwritten documents as accurately and efficiently as possible. Managing electronic documents for some of Belgium's largest health insurance companies has offered Moonoia the opportunity to develop and continuously improve its deep learning algorithms. What was initially created as a point solution to solve Moonoia’s own challenges of data velocity and veracity, docBrain has now taken the shape of a state-of-the-art document-centric AI platform. Today, the platform addresses the needs of both the end-user and reseller/partner companies in search of advanced, secure, scalable, cloud-based, and tailor-made document analysis and recognition solutions.

docBrain’s advanced capabilities typically improve existing technologies rather than replacing them altogether, helping organizations to supercharge business processes and fill efficiency gaps.

“When it comes to pinpointing what makes docBrain unique, I like to paraphrase Google’s Cassie Kozyrkov who said that, instead of figuring out how a microwave oven works and what components are needed, companies should just focus on the actual cooking at scale. docBrain does precisely that. It shields the complexities of machine learning to bring together data science, solution engineering and DevOps for end-to-end document-centric productive purpose,” adds Truyen.

Moonoia’s DNN technology optimizes images and accelerates document-centric processes of its clients with efficiency levels surpassing every other technology in the market. docBrain offers state-of-the-art data extraction and automated classification abilities to decipher, extract data even from handwritten, unstructured, and deteriorated documents. Moreover, docBrain can also read cursive Arabic. “docBrain’s neural networks result from all previous efforts of Moonoia, its partners and end-customers challenges and the open source community. Together, they bring very powerful machine learning capabilities”, continues Truyen.

Conceive, Consume, and Everything in Between

Geared towards solving complex document data extraction challenges, docBrain allows custom training and deployment of deep neural network models while offering an end-to-end platform and machine learning workflow, from concept/experimenting to production deployment. Further, the platform boasts of complete data security and privacy by design, ensuring data is protected at all times. Importantly, docBrain can be used by any user even without any machine learning infrastructure knowledge. Clients having specific data extraction problems can easily overcome their challenges using docBrain, because it is created to bring specific solutions. The platform delivers accuracy levels equal to or surpassing human levels for the banking, insurance, medical, and other document rich industries. Moonoia delivers its business model through partners who are aligned with the specific know-how of their customers and deploy the platform to the end-users.

The docBrain AI platform has two significant parts to it—the conceive part, and the consume part—put at the disposal of the clients. With conceive, Truyen explains that clients can create specific point solutions for problems based on neural network models that they train themselves, provided they have the necessary datasets (documents) to experiment with. It is called “conceive” because the resulting models are basically a client's own creation. “Consume”, if we were to simplify, is where all trained models become usable solutions. It is a customer production environment enabling quick deployment—in a matter of days. docBrain is the platform that brings data science, solution engineering and DevOps together for end-to-end document-centric productive purpose.

Moonoia's data trained algorithms have integration with many large and small companies around the globe including EY, CSG, Shoot &Prove, CGI, IN-RGY, InfoFort and Contextor, leading to the distribution and implementation of its docBrain point solution in their platforms. For SquareOne Technologies among the partners, docBrain is a platform (PaaS) that enables them to build tailored customer solutions. For end-users, the platform is a powerful technology at the core of a solution (SaaS) that they use for their own benefit. For both categories, docBrain helps redesign entire business processes without the need to invest in additional infrastructure, data science skills or AI expertise. The uniqueness of of docBrain lies in the fact that it is not just identifying and extracting the text in the embedded documents, but doing the necessary verification and subsequently feeding it into an update. This results in straight-through processing with absolute minimal need for manual validation.

docBrain as Part of an Ecosystem

While docBrain is Moonoia’s main offering, KAPTAIN is an AI-powered mobile data capture and automated document processing solution. KAPTAIN is a combination of three solutions that, together, create massive potential.

Shoot & Prove mobile solutions allow individual users to delegate the capture of information on smartphones and to generate and exchange electronic originals of documents with legal probative force. Next, docBrain is the technology powering the data extraction from the mobile images. Finally, Contextor is an RPA application responsible for automating the processes and integrating the resulting data into the company’s backend systems.

Achieving New Milestones in Cognitive Capture

Moonoia has assisted numerous companies spanning various industry verticals over the course of time. In an instance, the company has undertaken an award-winning project for one of Belgium’s largest health insurance funds for processing handwritten medical claims. Health insurance funds in Belgium known as “mutualities” were undergoing a digital transformation with a deadline. The challenge was dual: the federal government demanded that most paper-based medical claim forms must be electronic by 2019 and all these private companies were receiving less government funds. One of the health insurance funds, The Partena mutualities have been scanning documents for decades, notably outsourcing their business processes to Moonoia since 2009. Moonoia proposed the introduction of neural network algorithms to automate the data capture and validation stages of the process, and after a successful proof of concept, the solution was deployed. Moonoia used neural network technology to improve what was an error-prone, slow and costly end-to-end processing of handwritten, complex and sensitive medical documents up to the point where 80 percent of the operations were automated. The accuracy levels were above the statistically proven history of human reading levels for these documents. “Better monitoring and reporting brought stability, predictability and process intelligence across departments and branches for Partena,” extols Truyen. In 2017, this success story was awarded the “Project of the Year” prize by the Document Manager magazine, a prestigious industry publication based in the UK. Getting recognized among hundreds of established players so early on was a tremendous boost for the company.

Partners of Moonoia materialized successful projects for their customers in 2018. Some examples:

For AXA Insurance, as part of a KYC (Know Your Customer) process, docBrain was used to capture the personal data from tens of thousands of national identity cards. Trained neural network models read the input image coming from different touchpoint sources, determined where exactly the ID card was within the image, extracted the content and also checked if the information on recto corresponded to the information on verso. Everything was done automatically, resulting in straight-through processing with minimal remaining manual validation.

For a US-based provider of payment processing services, IN-RGY (one of Moonoia’s partners in the US) used multiple trained neural network models to turn a purely manual process into an automated invoice processing solution including: splitting the files into packages, identifying document types, extracting data from all document types, automatically validating data based on customer business rules. docBrain reduced costs and processing time significantly while increasing the data quality and process efficiency and decreasing payment time.

Adding more to its achievements, Moonoia was shortlisted in the Data News “CIO of the Year” competition 2017 and was included in the top ten of the best use cases at the NVIDIA GTC Europe Conference 2018. With such outstanding testaments, Moonoia is constantly improving and expanding docBrain’s functionalities and models, and aims to become an established leader in the cognitive caption sector. “To do this we will improve the docBrain product to meet and even outmatch customer needs in terms of user experience, trialability, accuracy, and end-to-end automation,” says Truyen. By offering language-agnostic recognition at the highest possible accuracy, the company aims to deliver straight-through processing, continuous quality monitoring and the possibility to perform unique and complex document inquiries. “As there is no one-size-fits-all when it comes to cognitive extraction, we believe that docBrain, with its ‘conceive-consume’ capabilities, will impact the market in a significant way,” concludes the CEO.