CCC Creators & Innovators: Neda Hantehzadeh
CCC’s Creators & Innovators blog series features members of our team discussing what excites them most about the work they’re doing and how they’re contributing to CCC’s growth and success. This month, we’re featuring a Q&A with CCC’s Director of Data Science, Neda Hantehzadeh.
How long have you been with CCC and what has been your career trajectory since joining the company?
When I joined CCC a little over seven years ago, I was responsible for developing and testing new generation of computer vision and machine learning models to further automate CCC’s vehicle damage estimate technology using deep learning techniques. Today, I’m still focused on finding new CV and ML applications, but for the broader product suite, in addition to our Estimate-STP offering.
What project(s) are you currently working on, and what excites you about it?
On a day-to-day basis, my team and I work on developing different AI models for our technology products and training them – just like the training I’m doing at home with my toddlers, ha! In simple words computer vision is enabling machines to view and process the information in images, videos, and even 3D or sensory visual information. Humans have a unique ability to process visual data when it comes to perception and understanding a scene, so we can look at an image of a vehicle and immediately understand that it’s a silver 4-door sedan that’s parked in a parking lot, for example, we can even tell from a reflection if there is a human standing by the car or identify a car shop in the scene because we have seen it elsewhere. AI models don’t process learnt information the same as the human brain. They typically need curated data to learn different tasks such as vehicle color, make, model, and door types. And each time we train these AI models with more and different data, they get better at recognizing and analyzing those components so they can make more confident distinctions between them. The common challenge that the CV society is trying to overcome is to design AI models that learn by interaction between different tasks and for them to be able to correct their own mistakes. And that is what excites me the most about my work when I think about applying these new concepts to our work here at CCC.
What are some important considerations when developing AI models for products that serve the property casualty insurance space?
Car insurance is complicated. Writing repair estimates may require a lot of back-and-forth communication between collision shops and insurance companies, and CCC is making it easier for these entities to talk to each other. CCC’s network connects 300+ insurance providers, 27,000+ repair facilities, vehicle manufacturers, OEMs, and so on, and all this data is flowing through our system. So, you can imagine how much data we have, which helps us teach our AI models. Over the past few years, we’ve been thinking more about what we can do with collision images, videos, and all the other structured and unstructured data flowing through our system, and one big consideration is how we use that data to make the AI more efficient, so it can help more people and businesses at scale.
What differentiates CCC’s technology from other solution providers?
What differentiates our technology from others is our access to millions of data points that we use to create new product features, and years of Auto Physical Damage Estimation experience. One of the first things we did as a data science team a few years ago was to start extracting as much context as possible from the accident images that are uploaded to our system, starting with identifying which parts were damaged. We also extracted data like the angle from which the photo was taken, and whether the camera was zoomed in when the image was captured to make sure we were creating accurate estimates based on the size, location, and type of damage. We then created AI models based on this extracted data that helped us introduce things like vehicle damage heat maps, panel identification, automatic photo capture, and more. Today, CCC has more than 300 AI models – many of which are enabling more than one million claims each year.
What’s your favorite part about the culture at CCC?
Innovation is at the forefront of everything we do at CCC. It’s what drives us. That’s why I love my team and the work we’re doing to enhance our technology and product capabilities. We celebrate the wins, we learn from the failures, and most importantly, we take great care of each other.