Mixing AI and People

Loss control is being re-invented today.

Artificial intelligence is the tool driving that charge

Artificial Intelligence seems to have the power to provide purely data-driven underwriting.

Where does that leave loss control?

As an underwriter, how comfortable are you with purely data driven insights

What are the guard rails? Where does human touch come in?

What will it be like in reality?

These are some benefits of what AI driven underwriting can offer…do you agree?

• Improve Loss Ratios with Data-Driven Underwriting
• Assist Agents to Make Better Risk Decisions When Used at Quote
• Decrease Time to Quote
• Avoid Cancellations Post-Bind Due to Failed Inspections

These benefits are plausible and possible.

How does it work in reality, say in Brooklyn for a local bistro with apartments above or at a small strip plaza in Topeka with a restaurant, 2 retail and a dry cleaner?

I think it will be a long time before super-regional and smaller carriers rely a lot on artificial intelligence to complete the underwriting process and to remove the need to do some level of loss control.

However, I think AI has some immediate use cases for smaller carriers and even agencies.

I like the term “Property Intelligence” when speaking about aerial and oblique imagery which is AI powered. The benefits are:

• Preliminary insights are generated remotely and jump starts the property knowledge cycle
• Risk scoring is possible with machine learned models
• Imagery allows for verification of the information’s in that moment and across time. It is great baseline info for square footage, roof quality, trees, pools and more.
• AI provides a layer of intelligence about the pro’s and con’s of a property, in seconds

AI driven imagery “reads” the data so the underwriter does not have too. The property intelligence platform can identify a risk management/loss control road map. AI property intelligence allows a cycle of IF/THEN decisions to be made.

The obvious IF/THEN options are:

• AI imagery looks great. Lets write it
• AI imagery is good but we have questions about the interior due to X.Y and Z
• AI imagery shows some risk factors, lets verify with self-service about the pool protection
• AI imagery identifies multiple risk factors lets not write this risk
• AI imagery identifies multiple risk factors and exposures let’s get this risk visited by a loss control rep

A progressive or graduated approach to loss control includes a full-on tool belt of digitally powered tools

• AI Imagery, Predictive Analytics and External Data
• Self-Service
• Guided Self-Service (Human Touch Point)
• 1:1 Video
• Traditional Rep with Mobile Device

It is important to have an agile approach to risk assessment. Each risk is different and that will drive a unique approach to loss control. Loss control can be configurable to each situation (see above)

Within progressive loss control there also will be a hybrid approach to underwriting.

Why? Let’s look at commercial underwriting. The use case is low premium commercial restaurants. You cannot afford a $300 inspection and the time drag to wait for 30 days for the results is a problem.

The obvious next step after property intelligence is run would be self-service. If your engagement is optimized with a great workflow, you can get 70% of the risks to complete the self-service inspection in 7 days for $25.

How do you process that?

Now let’s say you have 250 new inspections sitting there. In the past that means underwriters going through each report individually. With a progressive approach to underwriting, it is possible to triage with AI to filter up to the top, which risks need a human touchpoint. Here are a few ideas to process these self-service inspections:

• Use a loss control specific image recognition model to identify risks in the images. Some examples are:
  • Dirty hood and duct
  • Missing K Class
  • Out of date APD tag
  • Different floor levels
  • Cracked sidewalks
  • Railings on steps
• Build in questions into the app which trigger risk scoring. These can be specific questions you decide are drivers of loss
• How many pics taken? This indicates engagement and transparency. The minimum, average and over the top all send a message
• Speed of completion

Now we start getting into the details of loss control specific AI.

• How many AI/ML models do you have running for a restaurant inspection?
• How many defects are not covered by a model
• How do you cover the “holes” in the AI models?

At our company, we are very excited about the models we are building for image recognition. As part of a progressive or graduated loss control model, image recognition will add another layer of risk identification.

As more carriers adopt this approach, we will begin to have a better idea of where AI driven underwriting can operate on its own with no human touchpoint and what part of the risk pool needs a mix of AI and human touchpoints