Creating Value With Machine Learning

Machine learning is a subdomain of artificial intelligence which creates systems that can learn from the data they hold. The technology is capable of getting better at its task without explicit programming.

It’s this technology that powers Machine Learning Program’s innovative models, Propensity to Defraud, Propensity to Claim and our price optimisation service.

Propensity to Defraud uses machine learning and is accessed via an API. The model can support MGAs to reprioritise fraud checks, improve fraud capture before the event and enable MGAs to scale their operations moreeasily.

MLP Score is another product with a pure machine learning foundation. It supports MGAs by:

• Predicting likelihood to claim in policy period
• Improving interpretation of risk
• Flexing underwriting footprint and pricing
• Improving pricing knowledge
• Improving loss ratios and insurer relationships.

Our price optimisation model uses machine learning and data science to aid benchmarking of existing pricing against the market. Accessed via API and UI, it:

• Improves pricing knowledge
• Supports access to new markets
• Provides intelligence to support capacity conversations
• Improves conversion rates.

Let’s dig into Propensity to Claim; we built the model using the data that we see, depersonalised and anonymised, through our connection with Open GI. ​As part of that we have access to 93 million rows of data daily, and we actually use machine learning ourselves to work out which parts of that we need, by stripping out duplicate and inaccurate data.

​The Propensity to Claim model can predict how likely a customer is to make a claim compared to the rest of the population and gives a ‘Propensity to Claim’ for each policy it analyses. This enables our brokers and MGAs to work with their insurers to improve the performance of the book and place business with their panel to improve the rates (new business and renewals) they can get.

The first thing we do, without a charge, is run a retro. That means we use the model we have already built, to evaluate a year’s worth of data from an MGA (including claims data) which enables us to see how we would have scored each policy, had we seen it at point of quote.​

What we show first is a distribution curve. Our model gives a score from 0 -100, so each policy gets a score and we share that back to the MGA.​ This data becomes really important once we see how policies have been priced.

The data is always sense-checked because poor data breeds poor results and then, the real magic of machine learning can kick into action.

We develop a ‘disagreement matrix’ based on the data and it can show where policies are being mis-priced. For example, based on 14,475 client policies we define the premiums as high, medium or low. Then, using the output of the model we can build a ‘traffic light’ indicator to identify policies as high, medium or low risk.

Quite often the data will show us areas of mispricing across the board. See below:

Examining pricing related to loss ratios often uncovers further learnings and, through the power of machine learning, we can help our clients truly understand how to maximise profits by changing their quotes. This could mean cutting or replacing policies, or even replacing policies and offering a discount.

In this example, our client could save 100s of thousands of pounds by tweaking how they quote for some customers.

The benefits are compelling and, for MGAs, the data and analysis from the machine learning models feeds easily into Insurer Hosted Pricing systems. It takes just five simple steps:

• Build & Enhance: You create your product
• Deploy Instantly: You publish the product across all your distribution channels in real-time
• Capture Market Data: You immediately start collecting a rich stream of data on quotes, conversions, and referrals - seeing how the entire market responds
• Generate Insight: That data is analysed to find powerful new opportunities and identify hidden risks
• Refine & Adjust: You make intelligent, data-driven adjustments to your product's rates and rules.

Talk to the Open GI team to find out more about how IHP and machine learning can help get you one step ahead.