WHEN INSURANCE RELIES ON AI

PROTAGONIST

TRANSCRIPTION

HOST: What makes AI more than just a tool for automation in the insurance industry? How is it transforming claims management?  

CHRISTIAN: So, rule-based automation has been a core focus in the insurance industry for many, many years and specifically in claims management, where it has generated significant deficiencies. But there's two main limitations, you know. 

First, implementations are typically very time-consuming and therefore costly. And as a result, sub-processes with lower volume do not lead to positive business cases. And secondly, rule-based automation requires a solid understanding of historical data to know which decision rules lead to superior economic results. And this data has either not been available or the capacity to draw relevant conclusions out of it was lacking in the past. 

So, AI is now really a game changer as it can address those major challenges of the past. Setting up a self-learning algorithm is much less cost-intense than specifying and implementing complex business rules. And secondly, AI can draw meaningful conclusions out of big amounts of data on its own, if the data is available, of course, and not to forget, if it is of really high quality. 

HOST: Can you walk us through some key AI applications in claims management today? What processes are currently being automated with AI?  

CHRISTIAN: So, we do see innovations in various fields, such as in fraud identification, claims prevention, or assessment of claims amounts, just to name a few. But let me illustrate the fraud example a bit more in detail. Imagine you have a certain amount of criminal energy combined with a professional role at an insurance company that can channel a large number of incoming claims. The second one, tied agents typically do as they serve as kind of a single point of contact to the insurer for their clients. So in case of a claims event, for example, a car accident, they are contacted first and they will organize support such as a garage appointment and they will collect all the relevant claims documentation from their clients. 

Up to certain claims amounts, they are even entitled to decide on their own whether the claim is covered, and the insurance company will pay. This should simplify and shorten the process and intensify the client relationship with the agent. So, in general, it's a good thing, but on the other hand, it doesn't take too much imagination to think of fraud potential resulting from this practice. I see. So, agents do have close relationships with their customers. Some are even personal friends. And the agents could of course help to get uncovered claims paid or even create and let the insurer pay for imaginary claims.  

In one of my recent discussions I had with claims experts, I heard that one insurer has been able to identify a tight agent who has making use of this potential for years by creating fake invoices through duplicating and manipulating marginal aspects such as the invoice number or amounts of a real invoice. 

 Internal claims managers haven't recognized this as the individual electronic invoices were not different from the original one and the claim amounts were always below the threshold for more intense analysis such as an on-site inspection of the damaged car, for example. 

AI with its potential to identify patterns much better than humans was signaling the fraud team that over a longer period, there were so many similar invoices from an individual agent that they started an audit. And when looking deeper into it, they found that in many cases the insured cars had not been damaged, and the repair shops confirmed that they had never issued the invoices submitted by the agent. 

"What we have seen is that in many cases the depth of value creation of the insurer is quite high considering that skills and assets required for a successful AI journey are substantially different from what insurers have required in the past […] what I recommend to insurance managers is a pre-analysis of what is required and an honest assessment of what they can do well and where it is most likely better to partner and get access to external expertise". - Christian Ott, Director of Global Solution Design Insurance at SPS 

 

HOST: You recently worked on a significant white paper about the industrial use of AI in insurance, focusing on success factors and strategies. What are the main challenges that insurance companies face when implementing an AI strategy in claims management, and what obstacles do they encounter?   

CHRISTIAN: There are a couple of common challenges that prevent AI programs from yielding the full benefits possible. It starts usually with the governance framework that an insurer defines for such a program. What we have seen is that in many cases, the depth of value creation of the insurer is quite ..., and I would even say too high, considering that skills and assets required for a successful AI journey are substantially different from what insurers have required in the past. And the track record of successfully demonstrating flexibility and managing change is also limited. 

So, what I recommend to insurance managers responsible for an AI program is a pre-analysis of what is required and an honest assessment of what they can do well and where it is most likely better to partner and get access to external expertise. And secondly, as stated before, any AI implementation depends on having access to the relevant data in the required quality. 

This is quite often the challenge as input management function handling all incoming data and information from external sources is not set up in a scalable and efficient way, being able to provide new data requests required with short lead times and at a competitive cost. 

"With the growing demand for high quality data the need for quality assurance will grow as well. […]  Despite technological improvements, we still see the need for manual quality assurance to guarantee the data quality levels needed for process automation". – Christian Ott, Director of Global Solution Design Insurance at SPS 

HOST: AI tools are revolutionizing customer service automation in extraordinary ways. For example, Clara, the chatbot developed by Helvetia, elevates customer interactions with intuitive and automated conversations. Insurers are also leveraging AI to process images and videos from damage reports, enabling rapid and accurate assessment of damages.  

However, the impact of AI goes beyond automation—it also addresses the challenge of unstructured data. Unstructured data often hinders seamless integration with insurer systems, reducing efficiency. At its Bamberg Input Center, SPS tackles this issue by processing approximately 300 million inputs annually, transforming them into structured data that insurers can easily integrate into their workflows.  How have advancements in AI transformed input management for insurers, and what challenges still need to be addressed to fully optimize these processes?  

CHRISTIAN: So, technologies for classifying unstructured input as well as capturing and verifying relevant content, be it from emails, portal uploads or hard copy mail, have evolved quickly over the past five years. AI is also a key driver of this as well. 

And this journey for sure will continue. So, without constant market screening and experience in testing various technologies, insurers will not be able to feed their AI projects with the data they need, or they will pay substantial premiums for it, putting business cases under pressure and limiting the potential of the AI journey. 

Despite all those technological improvements, we do still see the need for manual quality assurance to guarantee the data quality levels needed to allow for process automation, which at the end means automated claim settlement. With the growing demand for high quality data, the need for quality assurance will grow as well.  

So, input management operations that deliver all those manual tasks from an on-site location in high-cost countries such as Germany, do not leverage the potential from near offshoring; they are paying substantial premiums for the data stream feeding into their AI algorithms as well.  

Outsourcing providers such as SPS, combining scalable platforms, technological expertise and innovation, as well as operational levers like state-of-the-art near and offshore centers, can help transforming the input management function into a powerful enabler of successful AI programs. 

HOST: Changing topics, fraud is a major concern in the insurance industry. How effective is AI in detecting fraudulent claims, and what advancements are on the horizon for fraud prevention?  

CHRISTIAN: As mentioned before AI can identify fraudulent patterns much better than the most experienced human operators. And the main challenge we see in this field is that AI technologies also enable fraudulent practices by using AI to create fake images, descriptions, police reports or invoices. So, without AI that constantly learn and evolves in line with the fraud industry that will improve their techniques with increasing speed as well, claims cost will grow without limits.  

Human experience and operators to handle fraudulent cases will be required rather more than less but without smart technology identifying relevant cases to dig into, they will run out of work quite soon.  

HOST: What roadmap should insurance companies follow to prioritize their investments in AI?  

CHRISTIAN: I cannot give a general advice on what is the best prioritization of use cases. This is really an individual decision, considering the specific client base as well as the technology stack given and historical data available. But as said before, I highly recommend carefully deciding what parts in an AI transformation journey should be sourced from external partners. This can save money in the long run as well as reduce investments needed. Furthermore, AI investments can only be as beneficial as possible when new tools are well connected to the existing legacy IT and external data input. Failures on one of those two preconditions will negatively impact margins from all other investments in AI and those aspects should therefore be prioritized.     

HOST: At what point can a company's structure impact the successful implementation of AI solutions? And how can insurers manage the necessary skill and mindset changes needed to work effectively with AI?  

CHRISTIAN: Like in all other traditional industries, an AI transformation requires a huge skill and mindset change on all levels of an insurance organization. And this is not a short-term project but rather a constant challenge that needs attention.  

In our survey, we learned that successful AI programs address this in three different ways.  

First of all, the organizational setup and structure: So decentralized and interdisciplinary teams, so bringing together data scientists and process owners or domain experts in their daily routines can generate much better outcomes than siloed organizations.  

The second point is qualification: Domain knowledge is really critical for data scientists and as well as AI developers so that they are able to fully understand the use cases they should find adequate solutions for. And on the other hand, the domain experts, for them it's really helpful to understand capabilities and limitations of the technologies used.  

And last one, communication: Naturally, there's fear of negative side effects of change, for example, job losses and other things. So, it is important to emphasize the positive impact such as improved client satisfaction or supporting operational units and let their expert resources really focus on client communication as well as feeding AI models and verifying the output of those. 

"Predictive modeling enables insurance companies to respond proactively to potential losses by identifying risks early and taking action before a loss occurs". - Christian Ott, Director of Global Solution Design Insurance at SPS 

HOST: What are the future trends in AI for claims management?  

CHRISTIAN: Let me give you three examples where I see AI to further transform claims management going forward.  

The first one is Virtual Assistance: The way insurance companies communicate with their customers will evolve. With the integration of generative AI, conversational interactions will become the norm.  

Customers will increasingly interact with AI-powered systems that are able to have natural and context-based conversations. So, this technology will improve the customer experience while increasing the efficiency of the communication processes. Instead of a customer filling out a form or editing a PDF, as it is today, they would speak to an AI-powered chatbot to report the damage. The chatbot would then ask targeted questions, clarifying ambiguities and ensuring that all necessary information is recorded. It can also ensure that the information is complete by asking questions as they are necessary until all the data is complete. 

The second one is AI Integration in IoT, Internet of Things: linking sensor data with AI-supported analytics. Insurance companies can monitor damage in real time basically, and also respond to it early. For example, sensors in buildings or vehicles could detect potential risks such as water damage or technical defects and transmit the information directly to the AI system, which then automatically initiates appropriate countermeasures.  

And the third one, Real-time analysis and predictive modeling. - my favorite one! -. Predictive modeling enables insurance companies to respond proactively to potential losses by identifying risks early and taking actions before a loss occurs. This technology will take loss prevention to a new level and further increase the efficiency of claims processing. The first examples I have seen are in the context of using weather predictions and GPS data for an alerting mechanism. An insurer, for example, has sent vouchers for nearby underground parking to clients driving in an area with a short-term hail forecast.