Written by Jonathan Bowden
The top 3 challenges of implementing Machine Learning for London Market pricing
2 minutes
AI holds huge potential for insurance, but what are the key challenges? And how can insurers overcome them?
When I speak to actuaries, data scientists and developers in our field, I’m often met with dejected looks when I ask about embedding machine learning (ML) approaches into the pricing process. To my mind, the integration of ML into traditional actuarial pricing represents an opportunity to leverage AI and develop more accurate and efficient models.
This opportunity, however, is overshadowed by some key challenges, which are by no means unique to the London market, or actuarial work in general.
Let's delve into the top three challenges and explore mitigating strategies for each.
Challenge 1: Lack of Available Data
At the foundation of any successful ML endeavour lies a robust and comprehensive dataset. However, in the realm of London Market insurance, obtaining such data can prove to be difficult. We are often forced to deal with sparse datasets. When sufficient data does exist, it’s quality and usefulness can be variable.
Without a good quantity and quality of data, we can’t even begin our journey—a lot of ideas for projects fail to get off the ground at this early stage. Or worse than that, maybe an analyst has excitedly come up with an excellent approach and has fleshed out the entire technical process, only to find that the real-world application won’t work as the data simply isn’t there.
This doesn’t necessarily mean the end of the project. There are several techniques we can try to augment the existing data. This can be done through accessing a wealth of 3rd party data sources such as property data APIs, economic indicators, stock market information, weather data and more.
Alternatively, the art of feature engineering offers a way to make use of seemingly unconnected data and, combined with expert insights, can create new information that can be incorporated into ML pricing methodologies.
At hyperexponential, our platform empowers insurers to start building a reservoir of well-structured policy data. By leveraging our solution, insurers can extract as much as possible from their datasets and accumulate relevant information over time, laying the groundwork for future ML applications.
Challenge 2: Lack of Infrastructure
Even if armed with a well-crafted ML model that’s been built on sufficient data, passing this on in a useable format for underwriters can encounter significant roadblocks due to gaps in IT infrastructure. Actuaries may face challenges integrating models into existing frameworks, with IT departments often prohibiting access to certain approaches, albeit with good intentions.
This challenge remains frustrating as underwriters continue to price risks on the same old basis, while the promise of more sophisticated insights and a more accurate model sits just beyond their reach.
To address this challenge, insurers can embrace Software-as-a-Service (SaaS) platforms tailored to streamline the deployment of ML models.
At hyperexponential, our SaaS platform offers a user-friendly interface that facilitates rapid deployment, allowing actuaries to deploy all manner of pricing algorithms, including ML approaches, using Python. There is a huge range of open-source Python libraries that can be deployed to your models, made easy-to-use and put directly into the hands of your underwriters.
Challenge 3: Not Knowing the Business Question
There is a lot of hype around ML and AI approaches in general. Certainly, there are huge opportunities when embracing ML's transformative potential. However, without the right strategic alignment and consideration of the wider environment, the true value of ML approaches to the business can go undelivered. There are often multiple stakeholders in a project that all need different things. Starting with the right question to be answered or problem to be solved is critical.
Changing a project already underway is often unpalatable. It can mean scaling back the project or reframing it entirely, and rethinking all the resources and people involved. However, the worst-case scenario here is pressing ahead with a model that does something incredible, but actually has no use in the current business. Sometimes the project sponsor has an idea for how they want things to look, but has failed to consider other barriers that are throttling the project.
To navigate this challenge, insurers must prioritise clarity and alignment when articulating business objectives. Engaging stakeholders in candid discussions can reveal key priorities, whether it's optimising rate adequacy, incorporating claims data, or leveraging specific data sources.
By articulating and agreeing on the problem that needs to be solved, we lay a solid foundation for our ML endeavours.
Ready to learn more?
As the insurance industry navigates the intersection of tradition and innovation, the integration of ML represents an opportunity to unlock new methods of pricing accuracy and efficiency. There are many challenges to overcome, and there are certainly more than the 3 listed above, but with the right expertise, partnerships and an innovative mindset, there is a path that insurers can follow to a more technologically competitive future.
