Images are everywhere, from personal phones to ATMs to video doorbells. These images can tell stories, prove facts, and reveal so much. So, how does this apply to insurance?

Karlyn Carnahan, Head of Insurance, North America at Celent

Karlyn Carnahan, Head of Insurance, North America at Celent

In this Analyst Answers, we hear from Celent’s Karlyn Carnahan about the evolving and unique use cases for images as data throughout the auto insurance lifecycle.

Can you set some context for why we should be talking about image data?

No customer-facing or front-, middle-, or back-office process is immune to analytic improvement with image data. Not only can we use a snapshot to discuss existing risks, but we can trend together past images and project into future scenarios. Watching scenarios play out over time instructs us on how confident our predictions can be, which affirms the usefulness of these data for assessing risk, understanding customers, and improving our businesses. Theoretically, “anything a human can see, a computer can see better. Practically, data from images can now be harvested at scale and used broadly to automate expensive manual efforts that have been a historical chokepoint for getting value from images.

 

Currently, where in the insurance lifecycle are images best being used as data?

Property insurance companies around the globe are updating their operating models to make full use of the new capabilities of computer vision as applied to images of buildings, their contents, and surrounding geographies. Building methods, construction type, materials, craftmanship (both interior and exterior), and even code enforcement and compliance are types of information critical to understanding both individual properties and the properties surrounding them. And in auto insurance, having photos from start to finish (from quote to issue in underwriting, from crash to repair in a claim, or capturing miles used for billing) is a new race for data-driven decision-making.

 

What are some of the biggest challenges that come with images being used as data?

Machines looking at pictures can be fooled, just like humans. A bad photo may create bad data, so much of the effort has been focused on helping systems learn how to take good photos that are fit for use. For example, dirt or reflections may create false indications, and it is still difficult for most people to get a good photo of a vehicle’s roof. Additionally, as images are annotated, those annotations are used to provide feedback and create learning loops to improve capabilities. What this means is that there needs to be an annotation methodology with a clearly defined taxonomy of items and descriptions. This taxonomy needs to span all sorts of data features that run the gamut of risk, marketing, underwriting, and potential claims issues.

“Machines looking at pictures can be fooled, just like humans. A bad photo may create bad data, so much of the effort has been focused on helping systems learn how to take good photos that are fit for use.”

Karlyn Carnahan

Head of Insurance, North America, Celent

When it comes to using images as data, what are some of the future possibilities that insurers may not be thinking about?

When it comes to auto insurance, photos today are often used to assess value, verify an odometer reading, or calculate repairability and replacement costs. But street cameras, toll gates, parking lots, street sweepers, police vehicles, and parking metering methods are just some of the everyday sensors reading and storing license plates with photos, extracted data, geolocation, and time stamps. Millions of images are collected every day, with billions of dots on a map waiting for analysis. These images can be used to assess the actual radius of operations or to help find a stolen vehicle.

Using the newer data types coming from images, it is possible to target specific properties for specific reinspection. Some data can be used for assessing insurance to value (ITV) (or changes in ITV), some may help an insurer identify new risk attributes, and other data may address considerations around terms and conditions for remaining insured (like installing a fence around a pool).

 

What can insurers do now?

Better data powers smarter risk-based pricing. Data-driven image interpretation will become the new way of working.

For now, insurers should:

  • Embrace remote data collection by first parties, robotic services, remote sensors, and data vendors, as well as traditional property inspection providers.
  • Demand that your image data tell a story for every property or auto — a visual analytic.
  • Create new analytic assets using clustering and “properties/autos like these” stories.
  • Use “ground truth” to calibrate features in each story and to learn over time.
  • Create and act on an analytics strategy that consolidates your use of data from images across the insurance value chain.
  • Innovate ways to create new value with this new look at an old data source.

Karlyn Carnahan is the Head of Insurance, North America at Celent.