Just a bit over 3 years ago, the Data Science and Analytics Lab (DSAL) inside American Family Insurance was asked to tackle two particularly tough questions…

  1. Could satellite, aerial, drone, and ground-level imagery combined with AI and Machine Learning substantially improve the quality and accuracy of property data used in quotation, underwriting, and renewal?
  2. If so, what would the impact be on the business and what ROI could be obtained?

Under the leadership of Dr. Martin Buchheim (who we are privileged to have as member of our Advisory Board), a team of Applied Machine Learning and Data Scientist began tackling this challenge.

After 18 months of intensive research and development, the team began applying the first set of deep learning models that assessed physical property characteristics such as Roof Material and Roof Area within the Underwriting team at American Family and also validating the quality and accuracy of the output data via ground-truthing against actual home inspection reports. The results, while nascent, were groundbreaking.

An early DSAL Deep Learning model output for Roof Area

Property characteristics generated by the American Family DSAL team’s deep learning models were derived from images that were far more recent than information contained in property sale and tax assessment records (one of the predominant sources of data used within most insurers in validating property characteristics) and were also achieving accuracy levels substantially above these other previously used data sources.

Physical properties are constantly changing and being able to assess the most current state of a property has a significant impact on the real risk an insurer is taking on when underwriting a policy.

American Family began testing the technology in stealth mode with several other innovative Insurtech startups (including our dear friends and now customers at Hippo Insurance and Roofr) who were using advanced analytics and Artificial Intelligence to create highly evolved and customer-focused insurance and homeowner products. These early customer-partners were able to quickly validate the real value creation that was occurring with the technology and immediately began incorporating it directly into their own products. They were also instrumental in identifying additional property characteristics that we could potentially provide to deliver even greater value.


How Arturo’s Deep Learning models see the world today (Image courtesy of Nearmap)

American Family realized they had something both unique and valuable and the next question became how to take it to the next level. After a strategy review conducted by members of American Family’s Corporate Business Development and Ventures team, a decision was reached that the best course of action was to enable this technology to spin-out as an independent entity, so that it could serve both American Family’s business, and also create value for the rest of the Insurance industry along with other adjacent markets.

On August 6th, 2018, I had the privilege to join Arturo as Chief Executive Officer and began the formal process of spinning out the Applied Machine Learning team that developed the deep learning models and approaches, along with the Engineering team that had built the unique infrastructure to deliver these model outputs on-demand in a way that could be immediately consumed directly in Quotation/Pre-fill and Underwriting business systems.

Along the way, we began engaging with other innovative Insurtech players including Kin Insurance and Openly (both who we’re proud to have as some of our newest customers) while simultaneously developing and releasing a number of new models to answer even more questions about physical property characteristics our customers had interest in. We were also able to finalize strategic partnerships with fantastic content providers including Nearmap and Maxar which will enable us to even better serve our ever expanding customer base.

At the end of April, 2019, we formally concluded our spin-out from American Family Insurance via the close of a Seed Series Transaction formally transferring all the intellectual property developed since inception to Arturo as well as establishing a unique licensing arrangement with American Family to utilize years of claims, underwriting, and inspection data to create new advanced predictive deep learning models. We’re thrilled to now call American Family a lead investor for Arturo as well as a key customer.

I’m often asked what makes Arturo truly different from other companies doing AI and machine learning on imagery data, and after 10 months of listening to what our customers say when I ask them the same question, I’ve distilled it down to four key points:

On-demand Processing — when our customers request data about a property via our API we fetch imagery from our content providers on the fly, feed it through our highly accurate deep learning models, and return a result as structured data — often in under 5 seconds. This means our customers always get the most up-to-date information from the most accurate models — every time they ask.

Confidence Scores — For each property analysis we perform on-demand, Arturo calculates and provides per property and per attribute confidence scores for our analysis, which means our customers have the confidence in the quality level we are providing, for every response we provide them. This substantially increases the utility of our data inside their business as they have an additional confidence value to leverage in their pricing and underwriting process.

Feedback Loops — we’ve created a methodology to automatically obtain and leverage customer input about our analysis from our API and web application that we call Full Loop™ deep learning, to automatically measure our real performance, capture new edge-cases and training data from customer interaction, and then use this data to consistently improve our models directly from customer feedback. This process means the more customers we work with, the better our models become for all customers.

Proprietary & Unique Data — we can test and validate our models real-world performance against years of actual claims and inspection data from American Family, allowing us to truly assess how we perform against data from a Top 10 US Insurer. It also enables us to provide models that truly have predictive value, like the likelihood that a home may have a roof replacement soon, or that it might be involved in a wildfire.

Today Arturo is 20 team members strong (and rapidly growing!) and based in downtown River North Chicago. Our entire team is incredibly happy to finally share what we’ve been quietly working hard for years to create, but even more excited about how that technology is delivering real value to our growing list of current and future customers.

The team at Arturo

We will be sharing more details frequently in the coming weeks and month ahead, but on behalf of the entire team at Arturo, today I’m privileged to finally get to say… Hello world!