Efficient customer communication with artificial intelligence

Ostangler Insurance can process incoming documents much more efficiently than before with the help of artificial intelligence. SEPA mandates and address changes are now 50% without human intervention, which is faster than before. This was made possible by pre-configured AI blueprints and just a few minor adjustments to the standard software.

The Ostangler Brandgilde is one of the oldest independent property insurance companies in Germany. Ostangler Brandgilde's products range from comprehensive cover to more specialized options, and are tailored to the needs and requirements of its members and the modern world. Digitalization, online contracts and modern insurance products are standard for them.

The project at a glance

83% of incoming address changes are detected correctly, 53% of them can be processed in the unassisted.
69% of SEPA mandates are correctly classified, and around 50% can be processed automatically without human intervention.

98% of other correspondence is correctly identified as such.
The target of 50% unassisted processing was quickly reached and reduces staff workload accordingly.

Initial situation

The majority of Ostangler Brandgilde's customer communication is now conducted digitally by email. Employees distribute all incoming correspondence (email and letters) to the appropriate business processes for further processing based on its content. Most of the incoming correspondence consists of customer requests, such as claims notifications or changes to existing contracts, for example changes to bank account details or address.

The effort required for manual processing is very high and takes up a lot of time. Although mail assignment was already performed with digital support, around two employees were needed just to process the incoming correspondence for the processes “change of address” and “SEPA mandates” (scanning, distribution) and to process it using an administration program.

As Ostangler Brandgilde is experiencing significant growth, the company was faced with the decision of either hiring additional staff to handle mail processing or implementing a solution that is scalable. The “change of address” and “SEPA mandate” processes were planned to be handled completely without human intervention. The insurance company, based in Kappeln, also wanted to free up its employees from repetitive tasks.

The goal was to be able to automatically handle at least 50% of the inbox for the two processes SEPA mandate and address change.

What is dark processing?

Dark processing refers to processes in which no human intervention is required to initiate or complete the process. They occur in the dark and are not visible to the normal user

Solution

With SHERLOQ, Ostangler Brandgilde has a flexible and versatile system that was developed cost-effectively for a medium-sized company with a small IT department using artificial intelligence and a combination of free software modules and customized AI models. And this is the SHERLOQ solution for the Ostangler Brandgilde.

Processing in standard dark processing format

Incoming correspondence is converted into a standard format for dark processing and fed into Ostangler Brandgilde's workflows to handle the two processes of “change of address” and “SEPA mandates”.

Where necessary, incoming correspondence is pre-processed using an OCR (optical character recognition) system and assigned to the relevant category of process. Relevant information is then extracted and transferred to downstream processes for further action.

Cloud-ready approach

codecentric pursued a cloud-ready approach to selecting the technology for the project. The solution is optimized for scalability to avoid excessive complexity and delays at peak times. What's more, it will be easy to switch to a cloud provider at a later date.

Processes from A to Z with a workflow engine

The individual process steps – from pre-processing to conversion into a standard format and transfer to internal workflows – are implemented with Camunda as the workflow engine.

Artificial intelligence for classification and extraction

The approach uses artificial intelligence to classify incoming correspondence and extract information. The modules employ current natural-language processing (NLP) approaches, which are taught on the basis of a defined and constantly growing data set. scikit-learn and xgboost were used for machine learning.

In addition, Kubernetes was deployed as the orchestrator and Tesseract as the OCR (optical character recognition) solution.

Result

The SHERLOQ solution meets all of Ostangler Brandgilde's requirements. The pre-configured AI blueprints and only minor adjustments to existing standard software justified its use with even a (currently) low volume of data. This setup made it possible to quickly achieve the goal of 50% automation for incoming correspondence with dark processing for the “change of address” and “SEPA mandate” processes.

Artificial intelligence and machine learning projects usually focus on hard metrics such as the accuracy of a model. From the outset, codecentric AG, working together with Ostangler Brandgilde, focused on making work easier and brought in important stakeholders and employees at an early stage. Dark processing requires a high level of confidence to be established through customer communication. The outcomes therefore clarify the true degree of automation and, with that, the actual added value.

In other words, Ostangler Brandgilde has succeeded in dramatically increasing the efficiency of its incoming correspondence process thanks to SHERLOQ.

We are really happy with the results. The numbers speak for themselves.
Andreas Schmid
⁠Chairman of Ostangler Brandgilde

Do you have more questions about the project?

Would you like to learn more about the project? Are you interested in a similar solution for your company? Then contact me.

Anna Lukas