RiskGenius Blog

Insights into our world of Insurance & Technology.

What I've Learned About Insurance and Machine Learning

I've been quiet the last month or so. There has been too much to learn.

We are learning from our customers.
We are learning from insurance policies.
And, we are learning from the machine learning algorithms.

Since so much has happened in the last six months, I thought I would highlight some of my favorite RiskGenius learnings.

All Lines of Business Now Available.

Back in February 2016, we launched a beta version of RiskGenius. That version only covered one line of business: Commercial General Liability.

Immediately, we heard from insurance professionals; they needed all lines of business.

In March 2016, I had no clue how we were going to figure it out. Then the Geniuses at RiskGenius figured it out.

I am so proud to announce that we now cover all Lines of Commercial Insurance. When you upload a policy, you select the Line and you are good to go:

Screen Shot 2016-11-14 at 7.14.23 AM.png

But we didn't stop there.


The Compare Feature is Now Awesome.

Sometimes I try to get too cute and I over design a feature. That was the case with our first iteration of the Compare feature.

The Compare feature is where users can compare one insurance policy against another. This feature leverages our machine learning algorithms by lining up similarly categorized clauses (it is not keyword search!). This means you can compare a "War Exclusion" from Policy A and Policy B even if they are written differently.

The RiskGenius beta version lined up similar clauses automatically.

But this caused a problem. Users were losing context. I remember two separate meetings where users suggested they wanted to control the clauses lining up.

So we gave control to the user.

Compare Feature

You can't see it in the photo, but we now present the policy as written. Then, when a user clicks on a category on the right, the two clauses sync up automatically.

Voila! This is the feature that has caused a room of insurance professionals to gasp out loud.


Insurance Nerds Want to Be Engaged.

When we first started evaluating the market for RiskGenius, I convened regular happy hours in Kansas City to talk to "insurance nerds." I define insurance nerds as fun people who care about the insurance industry and technology.

Since we launched RiskGenius, we are finding more insurance nerds throughout the world (yes, RiskGenius is international).

As part of one customer launch, I decided to try out an engagement strategy:

Genius of the Week.

The idea is that users are awarded Genius Points for using the software. The Genius with the most points wins a prize.

Today, we are expanding the Genius of the Week program from one customer to all customers.

Currently, the Genius Points are manually calculated (I pull a user activity report) but look for this process to be automated and embedded into the software UI in the future.

I want to make reviewing insurance policies like a video game.


Our Algorithms are Improving Rapidly.

When we launched RiskGenius on October 1, we new we were going to have a steep learning curve. We had decided that we would review every single output from our machine learning algorithms before publishing the results to our customers.

This meant we were going to review thousands of insurance clauses and assigned categories as quickly as possible.

The first week was rough. The second week was rough. The third week was a little less rough. And then the fourth week happened.

When I went into the weekend of that fourth week, I knew something had happened. I started studying our time to review algorithm results from one page. Here are the results:

Time to review algorithm

When we were in that first week, I circulated an article to my team: Do Things That Don't Scale. Reviewing at 3 minutes a page was not scaleable. We all knew that. But I also knew that the algorithms would improve as we received more data. 

The amount of time per page continues to go down. And we expect to make some major leaps in the next few months by skipping clauses that are repeated. 

Sorry to dork out, but it's just amazing to see these algorithm results get better and better. 

Do you have questions about RiskGenius or machine learning? Let me know at chris@riskgenius.com.