Aspect and Fluence Work Together on Portfolio Profit vs Loss Project

Some interesting findings here which could identify less profitable policy books in the future;

New analysis from scale-up MGA Aspect, undertaken in partnership with tech-linguistics firm Fluence World, reveals that when it comes to predicting losses, submission wording has predictive power.

Aspect teamed up with Fluence to analyse a portfolio of £85.6m gross premium across nearly 4,000 bound and closed policies. Fluence’s forensic linguistics and machine learning tech was set to work analysing 40,000+ unstructured files associated with the policies including PDFs, surveys, spreadsheets and broker docs with the goal of finding signals within the text that may otherwise have been missed.

Despite having no knowledge of the portfolio beyond submission documentation, the technology demonstrated impressive predictive powers in terms of identifying loss-making accounts and future claims. Eight out of the ten largest losses in the portfolio were flagged by the analysis based purely on submission documents while the portfolio’s biggest loss – £4.3m – was identified by the engine also.

Across the whole portfolio, the system was able to predict claims and losses based on submission wording alone with a 95% confidence interval – opening up a whole new way of analysing risk based on linguistic analysis.

The team also analysed the potential commercial implications of this. Had the analysis been undertaken at the pre-bind stage, nearly £15m in losses would have been avoided while the loss ratio across the portfolio would have been reduced from 28.2% to just 11.1%. Over the whole book, the model predicted a 21.7% loss ratio, compared to a real-world 23.2% loss ratio.

Commenting on the findings, Oli Williamson, Co-Founder and Chief Underwriter with Aspect said:

“Every specialty underwriter knows that submission files contain information beyond what reaches a rating model. The challenge has always been showing, objectively, whether that judgement can be measured and whether it genuinely relates to future outcomes.

“By working with Fluence to analyse submission language alone, without access to our rating model, pricing or underwriting decisions, we have shown that there are means of surfacing this information in ways which are highly predictive.

“These insights don’t replace underwriting judgement but rather measure and reinforce the judgement our team has built over many years.”

David Schofield, CEO at Fluence World said:

“Written language, just as is the case with spoken language, contains incredible depths of information that, on a first reading, can easily be overlooked. And as volumes of written work compound, so too do these hidden signals.

“Our team, by combining the best of linguistics with the incredible capabilities of AI, is able to surface these signals and discern patterns across huge volumes of documentation in ways which until now simply haven’t been possible.

“It’s been great working with Oli and the team to explore the implications our linguistic tech has for the insurance industry and, as we can see, the results could be transformational for underwriting teams looking to surface and understand risks before they occur.”