How to save Time in Reporting through Natural Language Generation
Reading Time: 4 minutes
Hello Stephan, please introduce yourself to our readers.
Hello Saim, it’s nice to meet you here in Berlin. I would be glad to introduce myself. I am now 46 years old and have been working for Creditreform Rating in Neuss since 2003. Since 2013 I have been an authorized signatory at Creditreform Rating, and since 2015, Managing Director of our 100% subsidiary CRA Services Ltd. in Bulgaria.
Creditreform Rating is an EU registered rating agency that provides ratings of companies, banks, Pfandbriefe, structured financings, and countries. I’m responsible for the credit services in our company, and we offer services around credit and risk management. Our goal is to make credit processes smarter, faster, and more cost-efficient. In recent years, our business segment has changed, and the digitization of processes has become extremely important.
Hasn’t Creditreform been digitized since its founding?
Creditreform was founded on March 9, 1979. At that time, digitization was not quite as advanced. As far as I know, the digitization process at Creditreform began in the 1980s and has continued to progress ever further. Creditreform Rating started operations in the year 2000. As part of the Creditreform Group, it was clear to us early on that we had to design our processes and services digitally. From Creditreform Rating’s point of view, that was a definite yes.
What is Creditreform’s new digitization strategy?
Creditreform has created a digital agenda in order to be able to react early on to the demands of the market and our customers. These include trend watching, idea scouting, cooperation with universities, and the establishment of our own lab. Artificial intelligence is an important topic.
However, transparency in the use of AI methods is also required here, as we as a service provider in the environment of financial institutions must meet the regulatory requirements. It must become clear how the supervisors will position themselves on this issue. At Creditreform Rating, we develop solutions in our digital ecosystem to meet specific customer requirements, without losing sight of the big picture.
Since our foundation, we have developed innovative products and established ourselves as an outsourcing company in the financial services sector. An important part of our range of services is the processing of annual financial reports. Here we have developed various solutions in the overall context of granting credits that simplify the procurement, structuring and analysis of financial information.
A customer requirement, for example, is the procurement of financial information. Via our FortDocs submission portal, international financial information (annual financial statements, digital account statements, interlocking information, and other creditworthiness relevant documents) can be transmitted securely, quickly, and easily in digital form. The combined use of FortDocs and the Digital Financial Report (DiFin) enables a complete digital document submission process. Our range of services is complemented by a web content crawler that searches websites worldwide for annual financial statements.
Digital text detection and text analysis via OCR, NLP and DL is the next logical step. Interpreting texts to determine whether they contain information relevant to creditworthiness or the assignment of unstructured balance sheet items to a given taxonomy are just a few examples of applications that we support today.
A very current project is the generation of text balance comments in the voting part of the credit decision process.
What exactly does Creditreform do with Natural Language Generation (NLG)?
The preparation of textual balance sheet comments is a very complex and time-consuming process and depends heavily on the experience of the credit analyst. We want to use NLG to optimize the process of commenting, to significantly reduce the manual effort, and to fully automate the process within the framework of portfolio follow-up disclosure. Currently we see a hybrid approach of human and machine intelligence. The machine prepares balance sheet comments that can then be finalized more efficiently by the credit analyst and, in the best case, reflected back to the machine intelligence.
After our decision to use AX Semantics in October 2018, we divided our NLG project into three learning phases. In the first phase, an experienced analyst from the banking sector prepared balance sheet comments, which we continuously transmitted to the software based on rules. This first learning phase took nine months. Now we are in the second phase, in which we automatically generate balance sheet comments and have them corrected by our analysts. The corrections are then taken back into the rules of regulations.
The first two phases are currently running parallel to each other in order to establish as broad a set of rules as possible. The industry-specific differentiation plays a major role here. The quality of the produced text items is so good that we want to start with the third phase. For this purpose, we are looking for customers who process our pre-produced texts in their processes and who also reflect our correction requests here. We have gained so much experience over the past 12 months that we can process these corrections on short notice on the system side. With each new text product and with each correction, we improve the quality of the texts.
Starting in November 2019, we will use the provided API to retrieve the texts via our own system and to deliver them directly to our customers’ decision-making systems.
It is also exciting that we have identified further use cases for ourselves, which are also being implemented. We will be happy to discuss this at the beginning of the new year.
You work with AX Semantics – what inspires you?
Working with the AX Semantics team is really exciting. I really loved the uncomplicated on-boarding with the integrated training concept. The support is also excellent. Whenever we have had difficulties or questions, there was immediate support. I am looking forward to further cooperation with AX Semantics.