Foundations of artificial intelligence and deep learning in the insurance ecosystem

What is left to say about AI? The hype is over. In recent years we have seen AI as an emerging technology become a disruptive technology. What started as an exotic tool for techies and data scientists has become a must-tool for every industry requiring support at every operational step: data gathering, data cleaning and preparation, data enrichment, data-driven predictive analysis, optimization towards desired objectives, quasi-real-time insights delivery, etc. How could the Insurance ecosystem be an exception to this? In terms of data, the Insurance Ecosystem handles the largest data trove. In their core business model, Insurance Companies, Reinsurers, Brokers, and Agencies in Life and Nonlife run on risk quantification and modeling, as well as on anticipating customers’ behavior. In the adjacent spaces around their core business model, Insurance Companies and Agencies exploit another massive data trove for upselling and cross-selling ancillary services around their core business offering: customer data captured during years of policy underwriting and claim processing (Ng, 2016; Chen et al., 2017; Panigrahi & Borah, 2021).

And yet… coming back to our previous statement, what is left to say about AI? The answer to this somewhat provocative question is simple: although the Insurance Ecosystem is no exception to the paradigm shift represented by AI and despite most of the businesses have been implementing AI-driven optimization and insights, we are still at the beginning of the adoption curve. Indeed, despite all the buzz surrounding AI, what most companies have been doing during the last 10 years is only the first phase of a journey that started with the Global Financial Crisis of 2008. The first phase was about adopting AI techniques to replace legacy approaches based on statistical methods largely due to the lack of a larger amount of labeled data. Most initiatives still rely on small-size pilot initiatives, testing AI capabilities generically proclaimed as superior vs traditional methods. What the industry is slowly realizing is that AI is not a panacea but it is a new toolbox that has to be put in place (Ribeiro et al., 2016; Weng et al., 2020).