Guillaume Bonnissent's Insurance Technology Diary: Digital judgement

Very early in my career I shadowed a motor fleet underwriter, sitting next to him on the box at Lloyd’s. I learned more about underwriting that day than any other.

It was during the pre-fibre optic co-axial boom. A broker queued to present my mentor with a proposal headed:

O’Brien Cabling of Manchester (1996) Ltd. Nine New Ford Transits

After saying good morning, my underwriter looked down at the cover page and scrawled ‘Decline’.

Surely that cover page offered insufficient information to judge the risk, I thought, and asked for an explanation.

“An easy one,” he said. “A new firm, new vans… O’Brien will hire a load of his nephews with no driving experience. They’ll complete short-term cable contracts as fast as possible, sell the vans, then move on to another venture. Buy new vans. It’s a bad risk.”

Notwithstanding a little possible casual racism, it was a masterclass in the art of deploying underwriting judgement.

Research published last week reminded me of this story. RDT, the software house specialised in claims management platforms, found that more than half of insurance people expect that employees in traditional functions will need more technical skills going forward. Data analysis and machine learning were mentioned specifically. More than 90 percent of those surveyed said they’re already investing in digital training.

But do underwriters need digital data skills?

Here are two facts:

1. AI predicts outcomes based on data.

2. Humans use their senses and experiences to draw conclusions from data and situations.

It’s a question of analysis versus judgement.

In my mind, the machines are much better at analysis, and they always will be. But when it comes to complex risks, where data is insufficient to fuel accurate predictive models, human judgement remains unbeatable. Sometimes it adds an invaluable layer of insight, even to the most actuarially friendly risk (like motor).

Actuaries have applied statistical methods to speciality underwriting for decades. Underwriters’ performance has been enhanced by models just as long. That’s why it makes sense to teach all these new digital skills to actuaries, the pioneering data scientists, and leave the underwriters to make the judgements.

Both roles require tech. It automates the compilation of data to feed the actuaries’ AI models, and displays the results to underwriters so they can apply their judgement.

The trick is to find the sweet spot.

Guillaume Bonnissent is CEO of Quotech