Talanx commits to robotics and Artificial Intelligence: partnership with WorkFusion to accelerate digital processes in claims processing

One of the objectives is to use the so-called “dark” - which means automated - processing without human involvement for even more glass invoices in motor insurance. This is already happening when large companies submit uniformly structured documents to the insurer. Up to now, when documents have been sent in by small businesses and private customers, automated reading of such documents has been a challenge. This problem has now been solved through cooperation with WorkFusion.

“We are already using robots in the insurance business. Thanks to the combination of Robotic Process Automation – RPA for short – and Artificial Intelligence, we are now able to intelligently automate the entire process end-to-end,” explained Dr Christoph Wetzel, Chairman of the Management Board of HDI Versicherung AG. The company is planning rapid upscaling. HDI would therefore like to use the new solution to partially automate a large proportion of the processes in claims as early as the coming year. In addition, application at the international subsidiary companies of the Talanx Group is also being reviewed.

Alongside its growth and profitability targets, the Talanx Group is also focusing strategically on digital transformation. One aim is to rapidly establish and expand competences in the area of Artificial Intelligence (AI), with particular emphasis on optimisation of AI-based business processes.

“AI and robotics go hand in hand for the first time”

Interview with Dr Gunnar Dolling, Head of top.Schaden – Top Claims – at HDI Versicherung AG

Which problem does the new software solve?

Every year, six million documents are sent in to our Claims Department. One example of this relates to service centre bills. These are frequently complicated documents without any uniform structure. While specialist service providers give some support for invoices, many other documents such as letters relating to claims registration and description of damage have to be dealt with manually. We want to accelerate this process for the benefit of customers by now providing in-house automation for many of the documents involved.

How do you envisage this in concrete terms?

It would mean that all the information given in the letter would be available as structured data. The machine then checks whether, for example, a service centre bill is correct. Payment information can be selected, address data are automatically updated and correct cost estimates can be forwarded directly to the service provider. Classification on the basis of keywords would be nowhere near adequate for this. We will automate the process intelligently by using machine learning. This means that the algorithms learn autonomously and constantly improve on the basis of feedback.

How did cooperation with WorkFusion come about?

We carried out a global contest with the aim of finding the best solution. Various providers trained their software for several weeks using anonymised customer datasets. Testing was then carried out. WorkFusion was the winner. They trained a model that immediately tripled the existing productivity in document classification. Furthermore, there was already an accuracy of almost 100 % for the key attributes. For example, the software recognised correctly that an invoice needs to be paid within a period of 30 days – this is information that even real people sometimes miss when they are processing documents. That was a positive surprise for us.

How precisely does Artificial Intelligence operate?

First of all, you have to train the system. This process involves supplying correctly classified documents to the machine. Our clerks manually highlighted individual items of data like addresses and phone numbers in hundreds of documents. The task to be carried out by the software was subsequently assigned to these data. The machine then takes over. It captures the data, considers how the classification has come about and then decides autonomously how it needs to deal with new information. This also works with data that may be structured rather differently from the usual format. The machine learns and continually improves its performance.

Where will you use the software?

Use is initially planned for invoices relating to glass breakage in motor insurance. There was already a successful Proof of Concept here. Colleagues from Indexing Motor, Data Analytics, Processes and Engineering, IT and WorkFusion cooperated intensively together to achieve this and automated the glass process within a few weeks. This meant that documents in the test scenario could be processed three times faster than was the case previously.

What is special about the new solution?

We have intelligent automation of a complete process here. Everything is carried out within one software system, which can make intelligent decisions as an overall system – and not in individual components. IT and robotics are therefore working hand in hand for the first time. This is automation 4.0 – because the AI platform can take end-to-end decisions, also known as Intelligent Process Automation (IPA). We were sceptical at the beginning of the process owing to the complexity of invoices and the claims system that needed to be serviced, but we quickly saw that it really did work.

What are the next steps for the project?

As far as we were concerned, it was important not simply to purchase some software. We are entering into a strategic partnership with WorkFusion. This involves working together with WorkFusion to develop the software further and identify how lots of processes can be automated very quickly. The plan is to keep below four weeks for a process from the idea to production. Various Talanx subsidiaries across the world have already expressed an interest in using the software as well. We are planning rapid upscaling for HDI in Germany and we would like to use the new solution to partially automate a large proportion of the processes in claims as early as the coming year.