Sramana Mitra: How do you classify the sex of the mosquito with computer vision.
Carlos Melendez: It’s not that different from classifying a truck from a car. They are different and have their particularities. The males have, what we see as, feathers on their bodies that the females don’t have. There are other differentiators.
After building the label data set of mosquitoes, we were able to successfully train the algorithm.
Sramana Mitra: After doing this in Puerto Rico, have you had another opportunity to replicate the same kind of application elsewhere?
Carlos Melendez: Not at this point, but we have had conversations with people, labs from Brazil, and other places in Latin America that are interested in the technology.
Sramana Mitra: Are there other computer vision applications that you have gone through?
Carlos Melendez: Yes. The other applications are mostly defense-related, so we don’t talk a lot about those use cases. They all had the same problem – to identify and create insights from particular objects in images.
Sramana Mitra: What else can you discuss that is interesting?
Carlos Melendez: On the AI front, we’ve helped commercial customers reduce their churn. For example, we work for a healthcare insurance company and they have a specific number of insured clients that can move from plan to plan.
They wanted to identify the customers that might be able to churn in a specific month. They tasked us with building a predictive algorithm for that. We were able to successfully create a churn algorithm for them. That churn algorithm was used as a basis for one of our churn algorithm products.
With their data and our base model, we were able to create a churn algorithm that had 95% accuracy in predicting the customer churn every month for that healthcare plan. The interesting projects we worked on had a lot of data challenges.
The initial algorithm that we created was poor. We went back to the drawing board and started studying the data, which is the most important part of an AI project. We were able to understand what was happening with the data and see the problems that we didn’t foresee.
We cleaned the data and helped the customer to better handle their data. With that, we were able to create an algorithm that is 95% accurate. When we delivered that to the customers and they finished validating the list that we gave them, the VP of enrollment told us, “Hey, you’ve just given us a crystal ball!”
Sramana Mitra: Is this a use case that you are doing for other insurance companies as well?
Carlos Melendez: This is based on our churn model. Using our model, we have helped other businesses in the financial services and telecommunications area predict churn.
Sramana Mitra: The use case is around churn management and not around healthcare churn management?
Carlos Melendez: Exactly.