In order to have an advantage in competition, business processes in insurance companies should be automated as far as possible, i.e. processed in the dark. To control these dark processes, factual data and information are required. If these are not available in a structured, consistent and complete form or are not recorded as part of the incoming mail process, the following situations are commonplace:

  • Imprecisely separated and classified documents "litter" the files and increase the effort until relevant data is found. 
  • Clerks manually type data from documents, for example, to determine whether a business partner already exists. If not, they check whether the data entered in the document is correct: Does the street exist? Is the postal code correct? And so on. This increases the effort.
  • Imprecise checks (partly due to time constraints) lead to duplications in the data stock. These unclean datasets further hinder future automation. 
  • Processing is often focused on a single case and depends to a large extent on the experience of the individual clerk. Duplicate billing, billing errors, potential reductions in claims or anomalies in the processes are sometimes difficult to identify. As a result, there is only a high dependency on specialists - in claims cases sometimes more expenses are incurred than necessary.

Rules and regulations in input management limit further automation

Despite all efforts that policyholders or claimants enter their data in a structured way in portals or use apps, many documents still arrive at insurance companies every day by e-mail, letter and even fax.

Subsequently a long processing chain begins :The letters received must be scanned or digital documents (e-mails and faxes) suitably converted. They are then classified and indexed using automated checking mechanisms. Data is extracted from defined document types and transferred to business processes together with the documents.

Standard systems based on keywords, if-then relationships, and layout are often used for the above functions. An expert administrator explains to a business analyst what the software should recognize. The business analyst translates the business requirements into technical specifications that are programmed as rules. Such sets of rules have often grown over years and are correspondingly complex. They often take into account all subsequent business processes, not just a limited area. As a result, even the smallest changes to automate sub-areas often have a major impact on all other areas. This complexity means that adjustments for further automation are expensive and implementation takes a long time.

As a result, many input management systems do not provide all the specialized data needed but actually present in the documents. The consequence: Automation and dark processing stagnate and fall short of the expectations of the business departments.

Dark processing in insurance with AI: Let machines learn the process!

Artificial intelligence and machine learning can learn the entire processing of incoming documents in a short time with a few examples and apply it to new documents. The starting point is a paradigm shift compared to the previous way of working:

  • Training instead of programming: Instead of programming rules, intelligent systems learn the processing steps from a technical expert or trainer using a few examples. or trainer on the basis of a few examples. This allows fast implementationto be combinedwith high flexibility.
  • Human in the loop: Based on its certainty (confidence), the AI suggests examples in training and eliminates cases in dark processing. This creates a continuous improvement process with minimized effort and the human retains control over the process and its quality.
Dark processing in insurance companies with AI

Intelligent automation in input management

Are you involved in the input management process in an insurance company, or are you in charge of innovation or business development processes? Then you can expect new insights in this free web seminar new insights. We will use practical examples to show you what AI looks like in action.

AI in input management: Achieving more with less effort

Modern AI platforms apply the aforementioned principles and supplement information extraction with machine vision methods, i.e. visual processing of documents:

This allows automatically compensate for distortions, e.g. remove distracting backgrounds from vehicle licenses or other official documents to improve text recognition and extraction. to improve text recognition and extraction. In addition, without human intervention, AI building blocks can documents already according to visual and textual characteristics, to target the further processing process. Thus, critical business processes are prioritized and processed in a targeted manner with a high automation rate. The recognition quality can be demonstrably increased by several percentage points with decreasing effort thanks to automated processes.

With only a few examples of dark processing in insurance with AI. 

Based on patterns, AI systems learn to separate incoming documents automatically and to classify the individual documents automatically. This prevents the files from becoming cluttered in the long term, and the orderly filing structure significantly reduces the amount of research required. 

Also by means of examples, AI systems learn to recognize relevant specialist data on the basis of just a few examples. Like a highlighter used to mark important passages in a document, the system is shown which information is important. No programmer is needed to "translate"; instead, the system learns directly from the subject matter expert. The learning process is so robust that the system learns the changes and adapts accordingly. This "self-healing power" of learning systems means that the specialist data is made available in significantly better quality than with conventional methods.

In the course of processing, special AI modules (predictive analytics) can also learn and make recommendations for action with regard to the content of a document. An exemplary area of application is fraud detection:

  • A workshop bills its working hours to an insurance company more expensively than to private individuals.
  • The system learns typical repair shop billing rates and alerts the clerk that an anomaly exists.
  • This can reduce the hourly rates or billing accordingly and save the insurance company money.

Conclusion: Automation and dark processing in insurance with AI at lower costs at the same time

Through the combined use of several AI or machine learning methods, more quality can be achieved in the extraction of technical data with less effort. All in all, automation and dark processing are sustainably improved in the insurance-related processes, while costs are reduced at the same time.

The introduction and use of such solutions does not require a project lasting several years and only comparatively low investments. With powerful, service-oriented architectures such as the inserve platform, business processes can be automated in just a few weeks.

Questions or comments? We look forward to your feedback on the article and a personal exchange.

We will be happy to answer any questions you may have.

With the online tool Calendly you can choose your desired appointment directly online. Book your 30-minute meeting with our IDP expert Christiane Tetzner today and receive a calendar invitation with access link to your personal team meeting.

Christiane Tetzner

Christiane Tetzner

IDP Expert

mail ctetzner@inserve.de
phone +49 511 936 857 67