Automation in the insurance sector. Workflow, AI, and LLM in practice (with examples of our deployments)

Altkom Software has been around for more than 25 years and, since the beginning, it has specialised in projects specifically targeted at insurance companies. A large proportion of our portfolio has focused on core solutions, workflow engine deployments and reporting tools for the insurance industry.

However, in this post, we would like to focus on punctual projects in which we have employed advanced solutions such as workflow engines, business intelligence tools and artificial intelligence to streamline and automate insurance processes.

Using workflow engines in the insurance sector

Workflow engines can be used for business process management, automation and optimisation. Their main goal is to:

• streamline workflows;
• minimise the time and effort needed to complete tasks;
• ensure compliance with rules and procedures;
• track processes step by step (for the purposes of an audit or analysis);
• continually enhance processes by making it easy to introduce changes and model processes in the form of diagrams.

Workflow engines in claims adjustment

At this point, we want to emphasise that workflow engines may play a key role in claims processing and adjustment in the insurance sector. The long list of benefits starts with automated claims classification, including the option to define different pathways for the claims adjustment process (e.g. depending on the type of product purchased by the client or the type of claim).

Another huge advantage is an easy claims form generator, allowing companies to include all the data they expect the client to provide (followed by automated data validation, with no need to program separate rules).

It is also worth emphasising that if the preliminary claim valuation is low, the workflow engine may automatically forward it to payout, without involving a mobile claims adjuster or repair shop.

In addition, at the management level, process engines allow companies to manage the entire claim processing process, coordinating and monitoring it end-to-end: from the moment the claim is filed until it is resolved.

Apart from automated task assignment and notifications, workflow engines also make it possible to closely monitor work progress, be it in the form of reports or visual heatmaps. This helps quickly identify any possible bottlenecks and suboptimal elements in the process.

In addition, engines can be integrated with business intelligence tools, which allows companies to collect, process and analyse claim processing data in real time. As an insurer, you get the ability to track key performance indicators such as: mean claims adjustment time, customer satisfaction score or claim servicing costs.

Bank BNP Paribas wanted to boost the effectiveness of its sales, cross-selling and upselling activities in online channels. The main problem was that the necessary information and processes were distributed over several legacy systems. To address this challenge, we deployed a process platform, known as the Digital Product Center, and then digitised and automated all sales and onboarding processes. As a result, the mean customer problem-solving time went down by as much as 70%; the number of errors in bank account opening forms dropped by 67%.

Artificial intelligence in the insurance sector

There has been a lot of discussion lately of the possibilities offered by AI in terms of process automation. The insurance industry is no exception. Advanced artificial intelligence algorithms can analyse documentation, assess claims and take damages payout decisions, as well as analyse huge data volumes collected by insurance companies for the purposes of identifying patterns, trends and risks.

Moreover, AI-based chatbots can ensure immediate and effective customer service around the clock. They can answer questions, provide policy information, help file claims and serve many more functions, which all contribute to better customer experience.

AI and personalisation in the insurance sector

AI can analyse data that provide insights into users, their preferences, insurance history and other factors to tailor insurance offers to individual needs and situations. In recent years, customers have increasingly come to expect an individual approach, which we discussed at greater length in another article: Loss adjustment automation in the age of customer centricity, so action in this direction is a chance to increase customer satisfaction scores and improve your retention index.

AI and fraud prevention

At this point, we would like to talk a little more about security, since AI can also be successfully used to identify suspicious patterns and signals in data, which may suggest an insurance fraud attempt. Dedicated algorithms can analyse customer behaviour, claim histories and other factors to identify irregularities and minimise fraud risk.

AI/ML in practice. Detecting insurance fraud

Together with our partner, we were hired by one of the leaders in the insurance industry in Poland to build an AI-based solution designed to facilitate fraud detection. Our client had historical data but did not know how to manage it properly. Before we stepped in, the company relied on technologies based on expert rules, which had one significant drawback: when strict rules were applied, the insurer got a high number of false positives, i.e. suspicious cases that, on closer inspection, turned out not to be a fraud attempt.

On the other hand, if the rules were not strict enough, the system was more like a sieve with really large holes. It only identified obvious fraud attempts, while letting through a large number of frauds that were less evident.

ML module deployment

To combat this problem, we decided to deploy a new solution, which would not require the company to replace its entire claims system; instead, we focused on the punctual use of advanced tools. At the heart (or maybe brain?) of our solution was an ML module responsible for machine learning. We focused on selecting the right predictors and training data so that the model would be as effective as possible, and at the same time simple and robust. Another part of the solution was a data analytics and visualisation module.

Development work, including the entire model training process, took just 4 months. After the first quarter following deployment, the fraud detection rate was improved by 60%; more than 80% of all cases automatically flagged as fraud did check out to be a fraud attempt.

Thanks to these savings, the development investment was paid off within the first three months from deployment.

After that, the solution began to generate profits.

Using large language models (LLM) and generative AI in the insurance sector

Altkom Software also has experience in building automated mail systems for communication with customers. The idea of the project was to deploy automated mail sorting and communication tools to take the burden off employees in charge of such correspondence. 

To do so, we used AI (LLM) technologies and AWS Cloud components to develop a solution that can categorise incoming mail, reach out to the customer, interpret their reply and, if needed, repeat or elaborate on the question until it gets all the necessary information.