„AI is a solution that saves money, scales automatically and makes robust decisions.“

Artificial intelligence has long established itself as a useful support in many industries. The many annoying bookkeeping tasks are easier to handle nowadays with the help of AI. The Berlin company CANDIS has made it its business to automate invoice releases with AI. Thanks to the software, companies not only save time, but also reduce errors. We spoke with CEO Christian Ritosek and Senior Machine Learning Engineer Deepak Mishra about the original idea behind CANDIS, the advantages of AI and Berlin as an ideal ecosystem for machine learning and AI start-ups.

Hello Mr. Ritosek, hello Mr. Mishra! Mr. Ritosek, how did you come to founding CANDIS and how has your team developed over the years?

Christian Ritosek: I founded CANDIS together with my entrepreneur friend Christopher Becker. It was always a thorn in our side that many business processes, such as in marketing or sales, are automated, but accounting and especially invoicing processes are still largely manual. The idea of CANDIS was born. We participated in the largest hackathon in Germany and built the prototype. With this prototype we won the hackathon and that is how the first companies became aware of us. We have developed CANDIS further and the team has grown from originally four people to now 100 employees.

Deepka Mishra​​​​​​​: I came to know about CANDIS through one of CANDIS colleagues via Linkedin in 2018 October and the idea behind it made me extremely interested and intrigued. Over the last two years, we have enhanced our team from two to four members. The machine learning team (called as Team Beyonce) has been a special task force team for CANDIS under engineering department, focussing both on research and practical implementation of algorithms, that is designed to precisely tackle modern problems of accounting.

CANDIS offers a software that automates invoice approvals. Mr. Mishra, how does the artificial intelligence behind it work?

DM: The Artificial Intelligence (AI) recognizes the pattern of approver/requester cycle for a particular organization and matches it with the contact issued by clients, formulates and learns those mappings to suggest and automate the approval process. It is a dynamic algorithm that learns, on the fly, with every correction and change in the workflow designed by users.

Mr. Mishra, how is the development team behind the artificial intelligence?

DM: The team essentially is a R&D team that focuses both on the research frontier by attending conferences, paper publications, applying patents and on implementing the novel concepts and algorithms as a software solution. We have researchers, interns and senior machine learning scientists that work with developers from the Engineering and Business Intelligence team to reach the desired goals.

What advantages does the AI have in the case of your product compared to classical accounting?

DM: Classical accounting involves a lot of repetitive and tedious work. AI comes as a savior here by automating the work that is tedious and time consuming. AI automatically manages the process of gathering, sorting and visualising pertinent data in a way that helps the business run more efficiently. This frees up staff to do more productive tasks and gives them more time to drive the business forward.

It also understands patterns and can dynamically learn from the user behaviour on the fly that scales the product as a real time automation. As one business grows so does the automation. There is financial fraud being done by hackers using AI and can be easily decoded and notified by building robust fraud detection AI modules to combat that. Minor accounting mistakes that can be overlooked by humans can be notified by AI. 

AI is a solution that saves money, scales automatically, makes robust decisions that are usually overlooked. It is important to understand that AI based solutions seem underachieving in the short run but have tremendous potential in the long run because the prediction & confidence improves with data that it feeds on. 

Your solutions contribute to revolutionizing the financial sector. How will technology and especially AI develop in the future? How can people and AI "hand in hand" work together in the future?

DM: The research of AI in the accounting sector is at an infant stage. It has all the potential to grow with the active open source conferences, research work by Institutions, Universities, Start-ups in the financial sector . The ease of experimenting AI ideas using cloud platforms for training/testing has literally brought resources into the hands of data scientists. Imperative and predictive power of AI will only scale up with the advancement in hardware in the next decade, not to forget the impact of progress in quantum computing.  

AI solutions can assist clients with information gathering, data crunching, routine tedious work, and physical labor, thereby freeing them for higher-level tasks that require leadership, creative thinking, judgment, and other human skills. It is difficult to predict with certainty how AI will shape up in next decades but we data scientist community have responsibility to design artificially intelligent systems using codes of conduct that ensure an automated system is able to respond to situations in an ethical way.

When it comes to finance, data security is one of the highest priorities. How do you ensure data security for your customers?

DM: Data security is achieved by different teams in the engineering department. The whole cluster of our customers data is only accessed by a few and with a virtual private secure network (VPN) with mobile authentication keys. Most of the crucial sensitive customer data is indexed and only the anonymised indexing is used for experimentation. For building up generic machine learning solutions, we have agreements in place according to GDPR laws with customers for using their data for training. For customers that do not provide that agreement, we have custom models deployed for them. 

Data emulation techniques are used to mimic patterns of customer data with a fake set of data that resembles the original and provides the solution we intend for. 

You have recently raised fresh capital, congratulations!  What changes are you planning thanks to the financial injection? 

DM: As an AI team, our focus would be to make the machine learning team run as an independent product team that delivers end to end solutions based on user/customer research interviews.

We would be hiring professionals to achieve the same. We communicate indirectly that the effective standard as a product for information extraction is CANDIS. We want to be torchbearers in this domain. 

So far, most companies in the field of artificial intelligence have been founded and established in Berlin. What advantages do you see in the location?

DM: Berlin is an ever growing start up center in Europe. The ecosystem surrounding ML/AI based startups is tremendous. It is exciting to present novel ideas and theories through meetups and conferences on a regular basis. Companies are also open to discuss, publish the crux of the research on medium for example, so that peers can make those approaches robust and scalable. Putting the novel research paper in the community also helps others to pick this up and enhance the algorithms built with new ideas and take this to unseen future forefronts which in turn benefits everyone in the long run. It is easier to find interns and have research collaboration with expert PHD candidates like we had for an IBB project (Investitionsbank Berlin (IBB)) in collaborating with TU Berlin while getting access to their research labs.