Product Documentation

Hail - Frequently Asked Questions (FAQ)

What AI feature generations support Hail Vulnerability Score?

Hail Vulnerability Score is only available on Gen6 AI Features. The request must include systemVersionPrefix=gen6.

Which feature classes can include Hail Vulnerability Score data?

If allowed by the calling account's packs, the response will contain both 'Roof' and 'Building Lifecycle' feature classes. Both can support the Hail Vulnerability Score, but only one will be used per building. The default is for the score to be attached to the Roof feature class. However, if the building's roof is severely damaged or destroyed, the score will instead be attached to the Building Lifecycle class.

What is the difference between vulnerabilityScore and vulnerabilityProbability?

The vulnerabilityProbabilityis the raw continuous output from the machine learning model, representing the predicted likelihood of a hail claim. The vulnerabilityScore is this probability binned into discrete 1-5 categories for easier interpretation and implementation. The discrete score retains nearly all the predictive information of the continuous probability.

What is the difference between vulnerability score and risk score?

The vulnerability score measures only the building's susceptibility to damage based on its characteristics. The risk score combines vulnerability with FEMA hazard frequency data to predict the likelihood of filing a hail claim. Vulnerability is property-specific and location-independent; risk includes both property and location factors.

How should I use the rate factors?

Rate factors are filed with regulatory authorities and can be used as premium multipliers. For example, a vulnerabilityRateFactor of 0.63 for a vulnerability score of 5 means an insurer can reduce premium by 37% based on this low vulnerability. A riskRateFactor of 0.77 for a risk score of 5 means an insurer can reduce premium by 23% based on moderate claim likelihood.

What are the log-transformed features?

Log transformations (e.g., logStaining, logFootprintArea) are mathematical transformations applied to handle wide ranges of values and prevent extreme values from dominating predictions. These are internal model features; users don't need to interpret the specific log values, only understand that the original features (staining ratio, footprint area) were used as inputs.

What if FEMA frequency data is not available for a location?

The vulnerability score can still be calculated as it depends only on building characteristics. However, the risk score requires FEMA frequency data and will not be available if hazard data is missing for that location. The vulnerability score alone still provides valuable information about the building's susceptibility to hail damage.

How often should I refresh Hail Vulnerability Scores?

Scores should be refreshed when new imagery is captured or when significant property improvements are made (such as roof replacement or repairs). Since the score is based on observable property characteristics, changes to those characteristics will affect the score.

Why might two similar-looking buildings have different vulnerability scores?

Even buildings that appear similar may have different vulnerability scores due to factors like:

Roof condition differences (staining, wear)
Different roof materials (shingle vs. other materials)
Different roof shapes (flat vs. pitched)
Building size differences

The modelInputFeatures field shows exactly which factors differ between buildings.

Can Hail Scores be retrieved for any AOI?

It is strongly recommended that the AOI aligns with a property/parcel boundary and that the parcelMode=true parameter is used. This will provide perils for a property/parcel. However, it is possible to provide any AOI.