Greg Tacchetti, Chief Information Officer, State Auto Insurance
According to IDC, global spending on Artificial Intelligence was just under $36B USD in 2019, and is forecasted to grow at a compound annual growth rate (CAGR) of just under 40 percent. For comparison, this is about the same amount US businesses paid for commercial automobile insurance in 2018. All of this investment begs the question, what is it we hope to gain?
Our company is growing at a 15 percent CAGR, and we are working hard to continually improve productivity so that we can gain rapid expense ratio improvements, and thereby provide our customers with more competitively priced products. Predictive analytics, AI, or I believe more appropriately, Machine Learning, is one of the big levers we are pulling to effect this productivity improvement.
Like many things in life, we have found the keys are keeping it simple, starting small, and focusing on tight communication and collaboration with the business owners who need to deliver the productivity improvements
We are trying to keep things simple. Our small team of data scientists works closely with our business partners on practical application of ML to deploy algorithms. A few recent examples of productivity improvement driven by ML algorithms are use of Natural Language Processing (NLP) and Computer Vision (CV).
Over the course of a few weeks, our team used open source NLP tools to interrogate voice streams for calls with customers to determine sentiment analysis, and CSR coaching opportunities. These capabilities are allowing us to cover the full universe of calls vs auditors pulling small and usually inadequate samples. Increasing our auditor staff by 10x still would not come close to the benefit we’re seeing from these algorithms. With the right talent, these are fairly straightforward solutions to implement in-house, and will start you on your AI journey. Another example is application of Computer Vision algorithms to interrogate imagery captured either by customers, or professional home inspectors. These CV algorithms found a significant error rate in identifying the presence of pools in backyards, and more importantly going forward, we now have capabilities to help our risk engineers, and customers better understand the risk being insured.
Many of these tools are open source and with the right team you can rapidly move through data cleansing, model training, and into production. Like many things in life, we have found the keys are keeping it simple, starting small, and focusing on tight communication and collaboration with the business owners who need to deliver the productivity improvements.