Artificial intelligence for insurance companies: The last puzzle piece to a fully automated claims process

Artificial intelligence (AI) offers enormous potential for more efficient claims processes in the insurance industry. Read here to find out how language models help to automatically process adventurous billing documents, where the challenges lie and why people will continue to be needed in the future.

Released:

Category: Professional articles

AI in claims processing

The industry is already relying on apps and digital products and is being supported by text recognition software and chatbots. What else is there to come?

After burglary and theft, water damage is the second most common type of claim in the household insurance. Nevertheless, these cases are not always easy for insurance companies to handle. Initially, it is still easy: many end customers now report the damage themselves via an app. If a smart home sensor is connected, it even reports the damage without human intervention. The insurance company can then automatically check whether it is covered. It is even possible to automatically call a company whose specialist will drive into the wet apartment, calm down the agitated residents and professionally dry the floor and walls. But then it gets complicated.

Because now the insurance company gets the bill. Smaller companies in particular usually don’t have interfaces through which they can transmit the document digitally and in a standardized way. They send paper invoices, often with a carbon copy and written in their own technical language. The document can be scanned and read digitally by the insurance company. So far so good. But in this case, “read” does not mean “understood”.

Ordinary text recognition software? It’s not enough

An example: The company lists “installation of suction hoses”, “installation of condensate dryer” and “emptying condensate bin” as services. Normal text recognition software records these services as individual items, and not always 100 per cent correctly. In technical terms, the unstructured data remains unstructured and therefore incomprehensible for the connected insurance systems. However, not only precise information is required for processing, but also a total amount for the “drying costs” benefit type. So what happens? A clerk pulls out a calculator, adds up the three items by hand, enters them and uses a table to check whether the benefit and total are covered. Only then the payment can be authorised and initiated.

Clearly, this manual processing is time-consuming and resource-intensive, and some specialists could dedicate themselves to something more exciting in the meantime. This is precisely where specialised AI can be very useful for insurance companies – especially in times of rising claims costs. This is because artificial intelligence can take over activities that do not add value. In this way, it helps to reduce processing backlogs, relieve employees, counteract the shortage of skilled labour and increase data quality. In addition, AI makes de facto fewer errors: it never loses concentration.

AI model translates invoices into insurance language

Faktor Zehn has created a use case on streamlining claims processes with the help of AI. Since 2017, we have been cooperating with the Fraunhofer Institute for Industrial Engineering (IAO) as a software manufacturer and implementation partner. The result of this collaboration: a solution that automates the entire claims process. It is the final piece of the puzzle towards 100% shadow processing.

The solution is linked to Faktor-ICS, the claims module of Faktor Zehn. The artificial intelligence used by Faktor Zehn and Fraunhofer IAO goes much further than conventional rule-based AI such as ChatGPT, which works using the probabilities of word combinations. This is because the underlying language model is specially trained for insurance cases – a real AI for insurance companies. It does not hallucinate, but “knows” the context in which it finds information.

In the case of the water damage invoice, this means that individual services – such as hose installation, setting up and emptying the dryers – are structured by the AI. It summarizes them in a meaningful way and translates them into the language of insurers (“drying costs”).

A set of rules verifies whether the service is reimbursable, is based on price limits, recognizes whether the invoice is congruent and whether the values are conclusive.

AI powers seamless claims handling in household, motor, accident, and more

If everything is in order, Faktor-ICS can release the payment, send the customer written information, prepare the final invoice and close the claim. If necessary, this works on a 24/7 basis. Is a human still needed for this process ? Not necessarily, but if you want you can have a person for sample checks, for example. The rules allow the insurance company, for example, to have every seventh case checked by claims handlers or to examine particularly high sums separately – depending on what is specified.

Paper invoices from craftsmen in traditional household claims are just one area where AI can be useful for insurance companies. The same technology also works in many other lines of business, such as accident or motor. Language model-based AI helps wherever unstructured data needs to be processed – regardless of whether it’s receipts for the payment of daily hospital benefits or expert opinions for broken windshields.

The time to decide on this is currently very favorable. The practical maturity of the models comes at a time when the insurance industry is in the midst of transformation, companies are digitizing processes and replacing their legacy systems with lean solutions. When selecting a new system, it is advisable to ensure that it is AI-compatible from the outset.

Irreplaceable: the human being as trainer and controller

A reliable legal framework and continuous training of the language model – these two things are crucial for the success of AI in the insurance industry. The latter creates a new way of working. Only when human experts validate the results the model can evolve profitably and function as it should. This is also a major sticking point: the quality of the AI results stands and falls with the quality of the database.

In terms of regulation, the EU’s AI Act 2023 has given the green light for the use of AI in insurance. The only exception: premiums for life and health insurance may not be calculated by artificial intelligence. This proves once again that AI cannot and will not replace humans. But it will certainly have an even greater influence on work in the insurance industry in the near future.