How to have data-driven insights when you have no data (or just want more)
You may be starting a new book of business, expanding to new types of customers, or entering new States. In all these situations, you have no data or not very much data.
You could shrug your shoulders and go without data-driven pricing and underwriting insights.
Or you could get creative.
At Octagram, we’ve been helping startup MGAs with this exact problem. In this article, we’ll share our approach.
Now, there are two types of data you need when you want pricing and underwriting insights:
claims data
exposure data
Claims data is the hardest to get, so we’ll first cover the four ways to get claims data when you have none.
Claims data
Publicly available event data
In 2009 there was a push from the federal government to make their data available online. Since then, more than 100,000 data sets have been posted to data.gov, and several cities, such as New York City, started Open Data sites.
The most interesting data are the event data, which could be used as proxies for insurance claims. For example, for aviation insurers, the National Transportation Safety Board has airline accident records since 1962.
The data is usually free, and if the database has every reported event, this can surpass the number of claims to which any insurer has access.
Tips
There are a few things to watch for when using publicly available data, such as:
Data completeness and quality. Sometimes the data results from voluntary reporting from many different entities. Some entities may not report all the time, and they may also report using different interpretations of each field.
Terms of use. Consult your lawyer to see if you can use the data for your intended purpose.
Is this event a claim? Not all the events may have resulted in a claim. They may be uninsurable, or the amount was immaterial, or for other reasons.
Severity. The event data is usually missing the size-of-loss information you would typically get with real insurance claim data.
Exposure. You may have the claims data, but you don’t have exposure information. For example, you have aviation accident data, and you may see that a particular feature on a plane appears in many accidents. This may be predictive of loss, or it may just be a common feature of planes. To know the difference, you need to marry your claims data with exposure data, and in matching these two data sets, care must be taken.
Examples
Aviation accidents National Transportation Safety Board
Auto accidents Chicago Data Portal
Medical Malpractice National Practitioner Data Bank
Product recalls Consumer Product Safety Commission
Securities Class Actions Stanford Law School
Claims data from data aggregators
Some carriers share their claims and exposure information in a data consortium. Often, the data consortium will require you to share your data too. The data is real insurance data at large quantities, and will give you a good benchmark, however, you wont derive proprietary insights from this data to set yourself apart from competitors.
Examples
Valen Data Consortium for workers’ compensation loss and premium data
Verisk DataCube for most core insurance lines. Data is not available on a policy or claim level, but is aggregated.
Getting customers to share their claims
If you are an MGAs that has grown out of another business with existing customers you want to insure, you can ask these customers to share their historical claims data. This is not an option for every situation, however several startup MGAs have begun in this way.
Loss-like data (e.g. from IoT devices)
Your MGA may collect Internet of Things (IoT) data, such as customer telematics information, which may record events that can be assumed to be accidents. This can serve as your claims data set.
Exposure data
To have insights, you also need exposure data. Here are five ways to understand more about your customers.
Publicly available exposure data
Like claims data, data about your customers may be publicly available from various government agencies. Some of this data is offered by third-party data vendors, which can compile and clean the data for you. Nevertheless, going directly to the source may provide new insights or be more affordable.
Examples
Restaurant Inspection Reports (e.g. Chicago)
Building Department Records (e.g. Chicago)
Boots-on-the-ground data
Customer visits by risk engineers are excellent opportunities to collect proprietary exposure data. Consider what relevant data points you could ask your risk engineers to collect to provide an edge over your competitors.
Also, consider how data such as photos and videos could be mined in the future using artificial intelligence, enabling analysis you haven’t yet thought of.
Third-party data providers
In the last ten years, we’ve seen an explosion of data providers that can provide insurers with data about businesses, individuals, buildings, cars and anything that is insurable.
Tips
Below are some tips on how to evaluate the data providers when you have no customers:
Send a synthetic list of your ideal customers, and ask for whatever data they have for these customers
Send as many customers as they will allow
Don’t rely solely on their own pre-made data sample. This will be an idealized set of data, and you want to see what they have for your own set of customers, which may be very different.
You should see how much data they can return and the quality of that data. It’s helpful if you can pick customers for whom you have some personal knowledge so you can judge whether the data is correct.
Ask how they match the data to the customer. For example, if you have a R.W. Smith LLC in Delaware at 34 S Main Street, would they consider this the same business as R. Smith, LLC in Delaware at 34 Main Street?
Ask about data sources and how often and by what mechanism they update their data.
Examples
Data about businesses:
Data about people:
Non-traditional providers
There is a wide field of opportunity to find proprietary insights by being creative in sourcing exposure data outside of those third-party providers who targeted insurers. For example, image data combined with artificial intelligence could provide unique risk insights. Business information from sites like Google and Yelp can be pulled directly from the organizations’ APIs.
Instead of looking at what data third party providers have available in the insurance space, think about what you want to know about your customers, and then see what type of organizations may provide this data.
Non-traditional data.
Some MGAs and carriers may leverage their own unique way of understanding exposures, such as a proprietary IoT device. It’s not an option in every situation, but it is a novel way to have exposure data before you have a book of business.
You can see that even when you have no data, you can get creative and have data-driven insights. If you have any other creative suggestions that we’ve missed, please put them in the comments below.
We’ve had hard-won experience working to develop insights when there is no traditional carrier data (or very little). Please get in touch if you think we can help.
P.S.: How do we know that these methods work to provide insight?
How do we know that these methods work to provide insight? We have developed two products using publicly available data, and they have shown good additional predictive power when tested against data from large books of businesses, above and beyond what carriers already use to predict these losses.
FireRQ leverages 1.5 million reported fires to help carriers avoid attritional fire claims
CrashRQ enables detailed territory rating using millions of reported accident events
E-BOOK: 10 Things Startup MGAs should know about Pricing: This e-book gives startup MGA founders a foundation in pricing so that they can grow profitably. We've distilled our decades of pricing expertise and experience working with clients just like you into this e-book. Get it here.