One in 10 insurance claims contains some element of fraud. And annually, fraud costs property/casualty insurers approximately $38 billion. While there’s no hard data on the scope of digital image fraud in claims, indicators show the issue is growing.
Photo-based claims estimates skyrocketed last year due to pandemic-related restrictions. While these types of estimates and the tools that facilitate them can provide convenience for policyholders and cost savings for insurers, they also offer low-hanging fruit for fraudsters to exploit.
In this second in a series of blog posts on top schemes in digital image fraud, we examine more sophisticated fraud methods and how non-binary image forensics can help detect them. (Read the first blog in this series on binary forensics).
Finding details in a photo’s fingerprint
There are numerous ways to submit fraudulent claims using photos, and many of them are relatively simple. For example, fraudsters can re-use an image from a prior loss or download a photo of damage from the Internet and submit it for a claim. Even though those are simple schemes, it takes technology like binary forensics to detect them at scale.
However, some types of image fraud detection require more nuance. That’s where non-binary forensics come into play. This type of image forensics examines the metadata of an image, such as the date, time, and location the photo was taken. Those details can help determine if a loss is legitimate or not.
For example, the photo below was submitted for a fence damage claim with a date of loss in September 2018 and West Palm Beach, Florida, as the loss location. However, the image metadata revealed that the photo was actually taken in July 2015 in Lake Placid, Florida. That’s three years and 100 miles apart.
Claim details said loss occurred Sept. 2018 in West Palm Beach. Photo details show the photo was taken July 2015 in Lake Placid.
Adapting to different claims
Non-binary forensics also must delve into granular details and fine-tune metadata checks by line of business and types of loss. For example, for a loss like fence damage, the metadata must match the location of the loss. But that same principle doesn’t apply to an auto loss because a collision can occur at one location, and the photo can be taken at a body shop at a different location. Therefore, the distance threshold for an auto claim is much greater than a homeowner loss.
Likewise, different loss types have different norms pertaining to date of loss. For example, if someone submits a photo for a fence claim, the metadata should show the photo was taken after the loss. However, for a theft claim, the image should have been taken before the item was stolen. If the metadata shows the photo is after the date of loss, then the claim is likely a suspicious one.
The point is non-binary forensics requires more granular analysis and can’t follow the same blanket rules because rules that apply to an auto claim don’t necessarily apply to a property damage claim or a theft claim. Otherwise, adjusters and investigators will be inundated with false positives.
Innovative tools for evolving fraud schemes
There are various ways fraudsters can game the system with digital images, so there needs to be a variety of ways for insurers to detect these schemes. Image metadata can provide powerful evidence to support whether a claim is meritorious or not, but it takes the right forensics approach and proper algorithm tuning to analyze the information to support more accurate results.
Non-binary forensics is just one of the techniques Verisk uses in helping insurers uncover digital image fraud. To learn more about our image forensics solutions, contact firstname.lastname@example.org.