Data Points

Have you ever looked at your home or auto policy and wondered what contributes to the premium staring back at you? Chances are you have, and chances are you're left with a lot of questions.

The obscurity is intentional because the answer is complex. The long and short of it: An array of variables describing the physical and structural risk of your specific location are fed through an algorithm which calculates the price needed to cover the expected risk of insuring against a pre-determined value. It's a numbers game, and it's lead by some of the brightest minds in actuarial modeling and machine learning. Property and casualty insurers live and die by competitive pricing, which is only as good as the models deriving it, and those models are only as accurate as the data funneled through them.

That's why high veracity data is a necessity for any P&C insurer. Data needs to meet the following 3 parameters to qualify as "high veracity" for underwriting purposes:

1.) The address data needs to reflect the real-world location as precisely as possible. For example, rooftop points are preferred over street and parcel centroids – we're insuring houses and cars not fields and stop signs.

2.) The attributes themselves should be free of error, descriptive, and reflect the "ground truth" of each location. This is hard to achieve at scale but incredibly important.

3.) The risk data, if derived from a risk model, should project risk at the most granular level available on the market. For example, hail risk at 1K resolution is better than at 3K resolution, and assessing wildfire risk at a neighborhood level is better than at a county level.

If the above is optimized, then the carrier can generally trust its algorithms advising when to write or reject business and at what price. Rejecting risky business is not the only advantage to using high veracity data – sniffing out low risk locations surrounded by otherwise high-risk areas brings business opportunity that goes unnoticed by competitors using coarser, less reliable data. Check out insurance data guru, Mike Hofert's, recent blurb about Which is worse when it comes to policy pricing: underestimating risk, or overestimating it?

Pricing and algorithms aside, sometimes the best way to validate high veracity data is to simply visualize and explore for yourself. A good place to start is at a high-level risk profile:

Followed by digging into the peril specifics at each property - such as wildfire risk.

Or perhaps flood risk.

And if this property does flood, how safe is the actual building?

And how much is it worth? What other structural attributes are available?

I encourage you to explore for yourself (for free!) by visiting the Property Risk Profile Demo on the Pitney Bowes Software & Data Marketplace. Knowledge Community members already have access.

We pride ourselves on curating high veracity data. Feel free to holler at us if you discover something otherwise or have any questions: https://www.pitneybowes.com/us/contact-dcs.html

Nice piece, Bryan.

It's worth noting that the words "risk" and "peril" are often used interchangeably, even though they have different meanings in the insurance industry. It's important as modelling each is different too. In the UK at least, they are defined as:

A peril is the cause of a loss. e.g. storm

A risk is the perceived probability of a loss. Many insurers have different risk appetites so may score the risk from the same peril differently. This means a risk-scored data product will also be interpreted differently.

To add to this, sometimes the term "hazard" is thrown in, which is subtly different again:

A hazard is something which changes the probability of loss from a specific peril.

I hope this is a useful note and shows it not just all about risk!

Have you ever looked at your home or auto policy and wondered what contributes to the premium staring back at you? Chances are you have, and chances are you're left with a lot of questions.

The obscurity is intentional because the answer is complex. The long and short of it: An array of variables describing the physical and structural risk of your specific location are fed through an algorithm which calculates the price needed to cover the expected risk of insuring against a pre-determined value. It's a numbers game, and it's lead by some of the brightest minds in actuarial modeling and machine learning. Property and casualty insurers live and die by competitive pricing, which is only as good as the models deriving it, and those models are only as accurate as the data funneled through them.

That's why high veracity data is a necessity for any P&C insurer. Data needs to meet the following 3 parameters to qualify as "high veracity" for underwriting purposes:

1.) The address data needs to reflect the real-world location as precisely as possible. For example, rooftop points are preferred over street and parcel centroids – we're insuring houses and cars not fields and stop signs.

2.) The attributes themselves should be free of error, descriptive, and reflect the "ground truth" of each location. This is hard to achieve at scale but incredibly important.

3.) The risk data, if derived from a risk model, should project risk at the most granular level available on the market. For example, hail risk at 1K resolution is better than at 3K resolution, and assessing wildfire risk at a neighborhood level is better than at a county level.

If the above is optimized, then the carrier can generally trust its algorithms advising when to write or reject business and at what price. Rejecting risky business is not the only advantage to using high veracity data – sniffing out low risk locations surrounded by otherwise high-risk areas brings business opportunity that goes unnoticed by competitors using coarser, less reliable data. Check out insurance data guru, Mike Hofert's, recent blurb about Which is worse when it comes to policy pricing: underestimating risk, or overestimating it?

Pricing and algorithms aside, sometimes the best way to validate high veracity data is to simply visualize and explore for yourself. A good place to start is at a high-level risk profile:

Followed by digging into the peril specifics at each property - such as wildfire risk.

Or perhaps flood risk.

And if this property does flood, how safe is the actual building?

And how much is it worth? What other structural attributes are available?

I encourage you to explore for yourself (for free!) by visiting the Property Risk Profile Demo on the Pitney Bowes Software & Data Marketplace. Knowledge Community members already have access.

We pride ourselves on curating high veracity data. Feel free to holler at us if you discover something otherwise or have any questions: https://www.pitneybowes.com/us/contact-dcs.html