Mathematics is an exact science that has been used to calculate premiums and risks since the inception of insurance. This was traditionally done by actuaries. However, with the rise of machine learning (ML) and artificial intelligence (AI), it is possible to use advanced mathematical formulas (algorithms) to analyze large amounts of data and discover patterns and trends. This already led to more accurate premium calculations (example the VPI box). The combination of AI and the vast amounts of data that insurers have can also be used to make underwriting and claims handling processes much more efficient and offers opportunities for personalized customer service at scale. All of this means that tomorrow's winning insurers will definitely drastically reduce their cost loading, while the customer service will actually be much better.

The hard truth is that this will put further pressure on the intermediary 's earnings model. After all, the intermediary becomes a relatively expensive distribution channel if both informing, advising and managing can largely be done by algorithms for a fraction of the commission that is currently paid.

Consumers, in turn, can find the information they are looking for increasingly easily on the Internet. For example, you can very simply askChatGPT what are the 20 most important points to consider when insuring a motorhome, for example. There are already several apps on the market that provide tailored financial advice using AI, such as Parthean or Mint. Admittedly not yet suitable for linking your Dutch bank accounts.

Thus, in this rapidly changing environment where AI is having a major impact on the insurance industry, it is critical for intermediary to make the right strategic decisions. On the one hand, by focusing even more on deployment of human attributes and, on the other, by learning to harness the power of AI to support their work.

The power of the advisor

Human qualities such as kindness, empathy, empathy, trust and respect are difficult to replicate by AI. It has been mentioned many times in this magazine, these qualities are and will continue to be essential for building good relationships with customers. Customer comes from the French word "chalant," which means as much as "attention. After all, we all know the word "nonchalant. With personal attention, advisors can create a loyalty factor beyond the purely transactional relationship between client and AI system. Especially if you also provide proactive support so that clients feel valued and well cared for.

Advisors can also specialize, building in-depth knowledge within specific niches or product areas. By staying abreast of the latest developments, advisor can provide valuable insights far beyond what AI can provide by harnessing their intuition.

So far so good, but as an intermediary with sometimes tens of thousands of clients, how can you give everyone that personal attention? Employing more advisors is not scalable and far too expensive. So how do I ensure that the advisor 's scarce time is used as effectively as possible and that time is invested in the right client at the right time? Smart AI tooling can help with this.

The power of AI

AI is a systems technology, like electricity and the internal combustion engine, for example. A systems technology always has a major impact on society, which cannot be foreseen in advance. For intermediaries, this means that they have to (learn to) embrace this new technology anyway in order to make their work more effective and efficient. We distinguish two main forms of AI tooling namely generative AI and specific AI.

Generative AI

Generative AI refers to the use of algorithms capable of autonomously generating new content , such as texts, images or sounds, based on existing data. The best-known example is on the website openai.com: ChatGPT-3 . GPT stands for Generative Pre-trained Transformer. It is a type of neural network architecture used to generate natural-sounding language. Advisors can already make good use of GPT to generate powerful answers to frequently asked questions, generate relevant content for Web sites or, for example, quickly request a summary of lengthy legal texts. Insurance software companies such as Wegroup are experimenting with this to use GPT in customer service as well. In the US, Sixfold.ai has developed a GPT model that can automatically estimate the risks of objects to be insured and provide appropriate coverage recommendations, within the acceptance guidelines provided.

For now, generative AI has some significant drawbacks. For example, manual corrections and quality control are always necessary especially when high-quality and accurate data is required. Also, the content must be validated because it is a black box. It is very difficult to understand the exact reasoning behind the generated content. Finally, generative AI requires large amounts of training data to effectively learn and generate new data. Those huge amounts of data can be found on the Internet, but almost never within a company for its own exclusive application. There are also legal risks, see for example www.arag.nl/nieuws/ chatgpt-legal-risk.

Specific AI

Besides generative AI, we know specific AI, also known as narrow AI. It is aimed at performing a specific task efficiently with a particularly high level of expertise and accuracy. This is something that is obviously very important in insurance land and there

'The intermediary is becoming an expensive distribution channel'

Voor de volgende use cases kan AI-technologieVoor de volgende use cases kan AI-technologie bij intermediairs en volmachten al succesvol ingezet worden;het verhogen van polisdichtheidhet voorkomen van royementactief klantbeheer op schaal inrichtenhet effectiever inzetten van de adviseursefficiëntere acceptatie- of schadeprocessenhet verbeteren van combined ratiohetidentificeren van bleeders en feedershetmonitoren van autoportefeuilleshet verbeteren van (digitale) klantbedieningom wordt deze AI meer en meer succesvol toegepast. Narrow AI is also used in applications such as image and speech recognition, recommendation systems and autonomous vehicles.

This AI requires (only) training data specific to the task for which it is designed. Suppose you want to use AI in this way to predict acceptance (probability) for the benefit of a motor vehicle STP, then you only need historical data of accepted and unaccepted motor vehicle applications. The relevance and representativeness of the training data are critical to the performance of these AI models. We mention some applications that are already successfully deployed at Dutch intermediaries and proxy companies.

'Advisors can make good use of GPT to generate powerful answers to frequently asked questions'

Churn Prediction: Based on historical customer and policy data, machine learning models can be trained to predict the probability that customers will cancel their policies within a certain period of time ( churn ). Algorithms such as logistic regression, decision trees or neural networks are used for this purpose. Advisors can use this information in a very targeted way. Customers with a high churn probability need immediate attention and should be proactively approached for a maintenance call and/or with an incentive ( defend strategy ). Customers with a low churn rate are obviously the loyal customers and with them the relationship can be structurally strengthened ( nurture strategy ). Direct action is not necessary. The model also shows at a diagnostic level why the probability of churn is high or low. This can be used strategically.

CLV Prediction : Customer Lifetime Value (CLV) is an important measure for intermediaries, who want to systematically and future-proof - read sustainably profitable - customer service. Technically, CLV can be calculated using a machine learning model that analyzes the same historical customer policy data. By combining this with advanced algorithms such as regression analysis or survival analysis, the model can predict the future value in euros for each customer taking into account, among other things, customer lifetime and cross- and upsell potential. This allows the intermediary to make the right strategic decisions, logically by investing in (acquisition of) customers with a high predicted CLV.

Customers with a high predicted CLV as well as a high predicted churn probability should naturally be given top priority by the advisor. In this way, marketing and customer-focused activities can be efficiently deployed and personalized services offered to actually achieve the predicted CLV. Below is an example of a simplified AI driven customer segmentation based on churn and CLV predictions.

Next best policy recommenders: By analyzing customer profiles, policy data and external data sources, machine learning models can be used to predict which additional policies or coverages are most relevant and attractive to each individual customer. This uses techniques such as collaborative filtering, which discovers patterns and similarities between customers to make recommendations. This allows the intermediary to automatically and continuously inform all customers about relevant coverages at scale. If a customer shows interest, the advisor can work in a targeted way to advise the customer ( enlarge strategy ). Targeted work on higher policy density, in other words. An additional advantage is that customers with more policies tend to be more loyal.

Robotic process automation (RPA): RPA is a technique used to automate repetitive and time-consuming tasks. This can be done just by using smart business rules, but it becomes even more effective when combined with machine learning (ML) models. For proxies or large intermediaries, the application is mainly in STP lanes for underwriting. Acceptors must verify provided data and check it against underwriting guidelines. RPA can automate these verification processes, thereby eliminating time-consuming manual work for "bulk" products. The RPA is then also configured to report any discrepancies to the acceptors. This keeps the human in the loop and immediately makes their work much more interesting.

Natural language processing (NLP): NLP algorithms can understand and process natural language and are applied in communication channels to provide quick and personalized responses to customer questions and requests. To do so, techniques such as text classification, entity extraction and sentiment analysis are used to understand the structure and meaning of textual data. Some relevant applications for intermediaries and proxies:

- Automatic processing of claims: NLP algorithms can analyze the contents of claim forms to extract relevant information, such as the nature of the claim, the parties involved and the details of the incident. It can also check the claim against policy conditions. This enables claims handlers to process claims faster and more accurately, resulting in improved customer satisfaction.

- Sentiment Analysis: With these NLP algorithms, the intermediary automatically and continuously gains insight into how customers feel about services or products. This identifies real-time patterns in customer feedback and potential causes of decline in customer satisfaction. In the Microsoft Platform, sentiment analysis can already be turned on by default for both written text (e.g. emails via Outlook) and spoken text (e.g. calls via Teams).

- Chatbots: With technology such as intent recognition or named entity recognition, chatbots can understand customer intent and generate relevant responses. This reduces the workload and improves customer service. Of course, everyone knows the "bad" chatbots. Much depends on the setup. At InShared, the well-trained chatbot is able to answer more than 95 percent of the questions automatically.

AI roadmap

So the good news is that AI is already giving the intermediary several opportunities that they can use to improve customer service. However, each office has its own course and customer group, so unfortunately there is no one-size-fits-all. Therefore, it is wise to work on an AI roadmap, where AI is effectively deployed for the benefit of the firm's goals (KPIs). It is a good idea to first explore with the MT possible use cases within the firm's goals and strategy. The use cases are then prioritized together in a business value versus effort quadrant. Use cases with a high expected business value that in the opinion of AI insurance experts for the office can be realized relatively easily with the use of AI are the quick wins. These are at the front of the roadmap. Use cases that are more difficult to realize, the major projects, come later. One, at most two use cases are selected, for which a pilot is set up in an operational environment with a limited scope. It is important that the business case can be calculated from the pilot. In other words, does the investment in the AI outweigh the expected business value in line with the organizational goal? Goals should be formulated SMART.

'AI already offers advisors several opportunities to improve customer service'

It is very important to also involve employees and give substance to change management. Thereafter, the implementation should be regularly evaluated and adapted to changing needs and technological developments. Also consider ethical considerations, data security and compliance with laws and regulations when implementing AI solutions.

The ball is in the intermediary 's court to start exploring the possibilities of AI quickly. There is no need to reinvent the wheel. There are AI solutions available that can be deployed immediately. To do so, inquire with software vendors or consult with AI experts. By using AI tooling, the intermediary can further leverage the power of the advisor , strengthen its competitive position, improve customer service and thus pave the way to a future-oriented and successful insurance business.

Dennie van den Biggelaar (co-founder OneSurance) has been an AI strategist for more than a decade and helped organizations such as Johnson&Johnson, CZ, BasicFit, Corendon, Sligro, Samsung deploy big data and AI applications.

View the article on LinkedIn or download it as a PDF