Scoring and triaging risks: using ML to automate the prioritisation of risk - Blog - artificial.

Scoring and triaging risks: using ML to automate the prioritisation of risk

Articles | Insights

Reading time: 3 mins

Attention to detail

At Artificial, we enable our clients to make the most of their insurance data and use it to score and triage risks efficiently.

We do this by using historical policy and claims information, our own risk scoring algorithms and a specially designed triaging process built collaboratively with the insurer.

We are constantly looking to make our client’s data work for them, whether it’s through implementing new insurance technology or by mobilising their existing legacy systems.

There is currently a wealth of unused insurance data that businesses can tap into to understand their customers better and improve product offerings.

The landscape of technology is changing quickly, and insurers need to work with Insurtechs to harness their data and stay ahead of the curve.

We know from speaking to our clients that it’s not enough to simply recognise a risk as good or bad; the characteristics of the risk are equally important.

If a broker wants a fair and accurate price for a risk, they will base this on a list of specified characteristics. The underwriter then needs to be able to make an informed decision using available data. This can be a messy and inefficient process.

Scoring and triaging

So how does this process work?

Say a client hands us a large pile of structured and/or unstructured claims and policy data. The first step would be to formulate a scoring algorithm which labels each individual risk according to its characteristics.

Take commercial property insurance. To allow the underwriter to make a more informed decision about a risk, we could include multiples factors into our algorithm, such as:

- Property address

- Size of property

- Total insured value

- Operations of insured

- Material of cladding

Once these characteristics are defined, the algorithm churns out a score. This score is then fed into a risk triaging process which is designed alongside the insurer according to their rules for each individual line of business (see graph below).

In this triaging system, an insurer may determine that any score above 8 is too risky and should therefore be ignored. In this case, an automatic message can be sent to the enquiries explaining that the insurer has no appetite for the risk.

Likewise, if the insurer wants to write any business with a low score - say below 3 - they can design the rules base to automatically accept these risks. Any risk with a score in between 4 and 7 is referred to the underwriter for further inspection.

Rules-based or machine learning?

Our scoring algorithms can be designed in different ways. The first is a rules-based engine, where risks are sorted according to straightforward rules set by the underwriter.

For example, in one proof of concept a client wanted to reject risks from a certain geographical region. The rules engine would automatically reject any submission from this area, and conversely could accept all risks coming from a different location. 

The second way to create these algorithms is by using machine learning. Machine learning is an application of Artificial Intelligence that allows a system to learn and improve based on experience.

In this context, the ML algorithm uses historical policy and claim data - the more the better - to predict a claim’s frequency and/or severity. Scores for each submission are assigned from this information and the algorithm continuously finds hidden patterns in the data to predict future risks.

Our results

A rules-based approach is a quick and effective way to start triaging risks. It can deliver clear efficiency benefits. However, on balance, we found that a machine learning algorithm delivered the best results for our clients.

The system works best when we have a lot of data, and depending on the line of business this could cover several years. This means a wealth of information and a highly detailed risk insight.

Machine learning also means that the system is also constantly improved and ensures clients are monitoring the relevant information for their product.

This method is also more flexible and allows insurers to see patterns in their data which are hidden when using one-way or two-way analysis with a rules engine.

Of course, collaboration and human interaction will never be redundant - our solution is to work closely with insurers to find their pain points and to tailor our algorithms to them and their product lines.

by Anna Burge
Share

Related Articles

    London

    10 Bow Lane

    London

    EC4M 9AL

    United Kingdom

    020 3880 8461
    Warsaw

    Koszykowa 61

    00-667

    Warsaw

    Poland