TrustworthyAI – Insurance AI in practice

Insurance companies have always been open to the digitalisation of their processes. Large amounts of data are generated that need to be processed, managed and evaluated. Any help is welcome here, as all standard processes are already highly automated. Nevertheless, in addition to the processes that have already been digitalised, I can see some where AI will bring a real improvement:

• Sorting and categorisation in input management (letters and emails)
• Support with formulating individual letters and emails
• Support with data analysis
• Automation in contract, claims and invoice verification
• Advice on the further development of products and tariff models

In every company, I see test projects, beacons or laboratories for gathering experience with AI. That's a good thing! I also find the first small applications. Sometimes it's a chatbot on the website, sometimes a test application for sorting e-mails. But I don't see the really big breakthrough. Hardly any company dares to use AI in its core processes. I sense a great deal of reluctance among decision-makers here. But why?

When I talk to decision-makers about it, their questions can be divided into three subject areas:

• 1. Regulation: What is an insurer allowed to do? What laws are there and how stable is the situation? What does a company have to do to use AI in a legally compliant way?
• 2. Employees: How will employees react to AI? Will they accept it or boycott it? What are their fears and desires?
• 3. Customers: What do customers expect? Will they accept AI or withdraw their trust in us?

TrustworthyAI as an approach

I recommend that my clients consider the TrustworthyAI Framework. It was developed by the International Telecommunication Union (ITU) as part of the AI for Good programme and is not software from a specific manufacturer, but rather a kind of toolkit. It covers topics such as AI in highly regulated environments, trust and security requirements, and the needs of employees and customers. This fits very well with the requirements of insurance companies.

The focus is on people

Insurance company employees often see great potential for AI in their field. At the same time, however, they are sceptical of the technology, which seems anonymous and opaque to them.

I can understand that very well. Working with an AI is often fundamentally different than with a classic system. Of course, there is the AI that sorts letters in the background and that you hardly notice, but often the interaction with an AI is more like interacting with colleagues. The AI provides direct support and advice, makes suggestions, helps, inspires, reacts to criticism or corrects the human work. The way the AI is presented can help here. Does the AI have a name? Does it have a voice, a personality? Alexa and Siri would probably not be so successful if they didn't have a name and a robot voice. Perhaps the AI also has an avatar with which it represents itself.

Employees must be able to trust the AI, both in terms of the results and the fact that the AI is not intended to replace them, but to support them. Because exactly the opposite is the case: the use of such tools in core processes requires new roles and skills. Employees who are curious and open to technology can act as multipliers here.

Data as a resource

AI models use data, and that is their basis. In the past, insurers often packed all available data into a data warehouse and hoped that the user departments would pick out the right information. This approach is possible with AI, but it is dangerous. What happens if the AI learns something wrong, categorises the data incorrectly and interprets it wrongly? What if it draws conclusions and learns rules that are prohibited by law?

In this case, employees in the specialist departments can take on the role of data stewards, taking responsibility for the cataloguing and quality of the data. In contrast to traditional applications, AI-supported processes often change more quickly, which also creates an environment of continuous learning and growth for employees.

Understanding results

In addition to the accuracy of the results, an insurer must also ensure that they can be explained. What use is a result that is definitely correct if customers or employees are simply supposed to trust it blindly? This is precisely what creates a sense of insecurity. AI must be able to substantiate its findings and decision bases. Only then can certainty arise. This also applies to regulatory requirements.

Mastering regulatory requirements

Until a few months ago, there was a great deal of uncertainty in the area of regulation. Everyone knew that the EU was working on an AI law, and from time to time drafts leaked out, some of which were over 500 pages long. But there were also repeated rumours that everything could turn out quite differently.

Now we know where we stand. The AI law has been passed and promulgated, and we know the scope and obligations. We are also aware of other relevant regulations such as DORA and VAIT, and TrustworthyAI provides the appropriate framework for these as well.

Assessment Centre for support

It is unrealistic to expect every department and the entire IT department of the insurance company to learn AI and understand and master TrustworthyAI. Here I see a sensible way to bundle AI expertise in a team. This team has a comprehensive overview of all the insurer's AI systems and activities, whether developed in-house or purchased, whether operated in-house or as a service. This team supports the introduction of AI in projects and advises projects on the aspects of TrustworthyAI.

I have had good experiences when such a team is a diverse mix. IT should be represented, as should the business department, and legal and data protection. From a certain level of maturity and for certain projects, it is worthwhile involving customers at an early stage.

Approach change actively

By now it should have become clear that the broad introduction of AI in a company presents a different challenge than replacing an IT system. Training, coaching and communication are essential to sustainably anchor such a change in the company. Employees are needed who approach this change with an open mind and curiosity, but also reflect on it. At the same time, the insurance company needs a tailored communication strategy that takes the customers seriously in their aspects. This can strengthen trust and increase competitiveness.

Conclusion

I see unprecedented competitive pressure on insurance companies in several respects. There are more and more market players, insurtechs, stronger brokers and comparison portals, plus a shortage of skilled workers and rapid technological change.

At the same time, AI is already offering companies that use it intelligently significant competitive advantages over their market competitors. And the development of AI is only just beginning. It can help to close gaps in staffing levels, both today and in the future. In addition, the EU has set the regulatory framework with the AI Act. It is clear what is and is not possible.

I believe that the time for small-scale experiments and test balloons is over. Now is the time to tackle the hot topics, to drill the thick boards, and to do so courageously and energetically. It is time to use AI in the essential processes and to take employees with us on this journey.

Would you like to learn more about exciting topics from the adesso world? Then take a look at our previously published blog posts.