Ki syndicate will use data science to evaluate policies and offer quotes

The combined ratio at Lloyds — expenses plus claims, as a share of premiums — climbed above 110 per cent — meaning insurers made no profit from underwriting. So reversing that is important.

"But insurance pricing is complex and unpredictable. For all Ki’s talk of data science it has not set its sights on lead underwriting. The future of Lloyd’s may be a hybrid working process in which digital underwriting is used for straightforward policies and insurers following a lead, while complex, bespoke propositions by lead underwriters are negotiated in person."

Lex in Financial Times 18th September 2020

This is reflected in P&C Insurance as well where automation is concentrated on the simpler, repetitive and predictable aspects of selling and managing policies and claims. Straight-through-processing (STP) is attractive to reduce costs and speed up settlement but has its dangers.

STP and Automation might be great for policy holders but, unfortunately, for the opportunistic and professional fraudster STP is also undeniably attractive. Just as data mastery and the judicious use of AI is promoted for insurers to tackle fraud so professional fraudsters can use the very same technology to confound counter fraud efforts.

Fraudsters look for chinks in the industry’s armour and exploits the weakest link.

  • Genuine online documents can be altered
  • Receipts falsified
  • Accounts adjusted.

As fast as a fraud team’s algorithms find a pattern of fraud the fraudsters alter the modus operandi (MO) in anticipation. Fraudsters take advantage of the long and vulnerable tail of motor claims in which they can add false evidence. Infants in the car. New physical and emotional injuries. New passengers.

There will be a long tail of COVID-19 claims including, of course, business interruption (BI) claims. Two major insurers have deployed 360SiteView to manage BI claims and the platform includes a hybrid mix of automation and human intuition. Automated processes to surface, normalise and analyse data from structured and that large portion of unstructured data in a claimant's submitted documents and external data sources like Companies House. that can fool many an algorithm.

360SiteView categorizes the claims into classic RAG buckets so that claims handlers can efficiently and effectively make optimal decisions faster with the aid of technology. Focus on those claims that need more investigation and ensuring that claims that the FCA judgement indicates should be paid out on are not held up.

Hybrid models are vital. Take a simple matter in P&C home insurance. There is a temptation to automatically pay-out on claims below a certain value say £900 as the cost of investigation may be higher than the cost of paying out on fraudulent claims.

A well designed self-service eFNOL will expose fraudulent claims in a simple process of nudging the claimant to: 

  1. Briefly describe the incident and once done,
  2. Ask them to submit 3, 4 maybe 6 photos from their phone or tablet
  3. And then request a video to join he photos up and give context.

With over 4 million claims having been processed globally can you guess what percentage of claims are withdrawn voluntarily? Without expensive AI apps?

35% and more across all geographies

That is an awful lost of money to pay out and algorithms will not catch that many.

If you want a practical case of why algorithms can be fallible consider the recent musings of our CTO. 

Alexa, which must have one of the most effective set of algorithms of any software, consistently mistakes the vacuum cleaner motor sound for his voice and reacts with comments and questions!

Algorithms have an important role and in the broader application of Intelligent Automation insurers must have a clear goal, strategy and implementation plan for two and five year deployments.

But importantly as a component of hybrid processes, workflows and working models.

Otherwise, you don't know what will be hoovered up into the wrong places.