When Artificial Intelligence comes up, it often conjures up thoughts about robots, lost jobs, and complex systems that - at first glance - can seem impossible to grasp. However, a large component that can be overlooked is the time dedicated towards teaching artificial intelligence and machine learning what you want it to. As we’ve emphasized before, the people and expertise behind the technology play a crucial role in driving the direction of usage, training, implementation, quality assurance and more.
At LOGiQ³, we’re using AI to make reinsurance treaty data more useful, accessible, and easier to manage across the industry. Below you’ll discover the processes, insights and challenges involved in training and developing AI to meet our specialized industry needs and the progress we’ve made thus far. As we reach key milestones along the way, our team has uncovered tremendous potential in this technology that can make reinsurance operations more effective and efficient.
Project Milestones: AI for Reinsurance Treaty Reviews
1. Identifying the Problem
A good solution always starts with a significant problem. In this case, we recognized that many reinsurance professionals are sitting on a ton of important reinsurance treaty data that can be invaluable if sitting latent in a storage room or computer drive. Organizations are unable to leverage this data to analyze business performance, test for accuracy or use it to make strategic decisions.
2. Deciding on a Viable Solution
When designing a solution, we had to consider what was most important to bring the data to life: the solution had to be able to recognize, capture, read and structure the unstructured data from many different types of treaties – many of which are decades old. We began looking at Artificial Intelligence for its Natural Language Processing (NLP) capabilities to determine if it would be able to capture specific terminology in treaties. With hopeful early results, lots of existing data to work with, and a group of industry professionals at the helm, we were confident that AI would be the most applicable, scalable, and impactful solution.
3. Building the System
As we started building a tailored AI approach for reinsurance treaties, we identified the specific terms that were of the most importance to our existing testing parameters – whether we’re looking at effective dates of treaties, quota share, or retention percentages, the system had to be designed to pick up those key terms. Once the terms were finalized, we needed to teach the machine how to identify and capture those values automatically.
4. Testing the System
Our initial test was to just run a treaty document through and see what the AI system was able to capture. Once 4 or 5 treaties had been sent through the system, it was able to recognize 1 or 2 fields consistently. This was an exciting indicator of how successful the system could become, but only scratched the surface of the terms we wanted to capture. From there we did more testing. Putting the system into use allowed us to see the experience over time and understand how quickly the AI was learning. It also helped us gauge its ability to predict the field values going forward.
5. Iterate and Improve the System
The testing led to new developments, understanding, and enabled us to strategically and effectively improve the system. Once 20 treaties went through the AI, we were able to see an increase in recognition of 5 to 6 field values. We also noticed that, even though it was picking up values, they weren’t always the correct values. This led us to develop two levels of training: finding the right information and ensuring the accurate interpretation of it.
6. Turn from Solution to Experience
Once we were positive that we were able to leverage the AI to capture useful reinsurance information, we needed to determine what the next steps would be for the data being captured. How were we going to package it? What was the best way to structure the data? How would the user need to use and pull that information? These and other key questions emerged that helped us determine how to structure the data warehouse and how the user experience interacts with AI.
7. Keep Innovating: From Recognition to Automation
Our next steps are to automate the testing and then move onto more complex recognition for the AI. This includes testing integrity by comparing the data captured to the data in the TAI reinsurance administration system, compliance to treaty terminology of the sessions that are ceded, and premium calculations to ensure that policies are paying as they should be. The more the technology learns from treaties, the more possibilities emerge for improving and iterating, what can be done with the newly structured data.
Follow our journey as we bring reinsurance administration into the future with AI and other emerging technologies. Don’t miss a milestone: sign up for updates to stay on the cutting edge below.