Demystifying machine learning – Open GI

Machine learning uses statistical methods to improve decision-making at speed.

While machine learning is a subset of artificial intelligence (AI), it is important to note that it is not done by magic and robots – it is guided by clever humans and maths.

How does machine learning work?

Machine learning runs on data and programmes can sort through millions of rows of it.

Essentially, programmes can be taught to recognise certain patterns, parameters or boundaries in datasets and then establish rules to identify these metrics again later.

After learning these patterns, the programme can reapply them to new data and filter through it in a matter of seconds. Traditionally, this is used to predict whether the new data belongs to a specific group, or to predict a specific number – like a premium or claim cost.

These models can grow to a tremendous size and filter through all sorts of data in a matter of seconds.

The more data organisations have – and the better quality it is – the more accurately their programmes can learn what to look for.

What is ‘quality’ data?

Machine learning can only be as good as the data fed into it.

One of our key jobs at MLP is to clean data – this involves removing duplications or inaccuracies before feeding it into machine learning programmes.

Quality data changes all the time. That means machine learning is never one and done.

Every time new data is added to the machine learning model, it finds and learns new patterns – it is therefore up to firms to make sure the technology is learning the right things by comparing its decision-making against the choices underwriters would take.

At the end of the day, machine learning is about using technology to make the easy decisions, so staff can then spend their time on the hard ones and businesses can grow as a result.

What does machine learning look like in action?

There are so many ways machine learning can help solve insurance problems, but two examples of tools from MLP are propensity to defraud and propensity to claim.

Propensity to defraud gives brokers a probability score of zero to 100 of how likely someone is to commit fraud and a red, amber or green flag. The programme was developed with fraud experts, using their knowledge alongside historical and external data.

Instead of doing a fraud check manually, which used to take hours, brokers using this tool can get an answer almost instantly – at the point of bind.

Meanwhile, propensity to claim does the same with someone’s likelihood to make a claim on their insurance – right at the point of quote. It’s a really powerful tool that’s helping brokers improve loss ratios, make better pricing decisions and keep their book clean.

What do brokers need to know about machine learning?

Machine learning is not mysterious. It is brokers’ own data and knowledge, harnessed and put to use – just a bit faster. Machine learning is already creating efficiencies and improving accuracy across the industry.