About Ian Deters

Ian Deters is a mathematician, data scientist, and teacher who believes that predictive models exist to inform action. Since understanding precedes action, this means that predictive models must be both explicable and precise. Both components must be present to provide maximal value to the stakeholder. The stakeholder should not have to choose.
While working in the fields of Data Science and Machine Learning, Ian made two observations which formed the basis of his independent practice EP Analytic Solutions. First, there was a tacit assumption that there is an inherent tradeoff between the precision of a model and its interpretability. He considered this a category error since precision is a mathematical concept and interpretability, a subjective one. Second there were many gaps in the continuum from mathematical theory to the practice of data scientists and actuaries. Having a PhD in theoretical mathematics and over ten years of practical industry experience, Ian was in a unique position to attend to this. By bridging said gaps Ian uncovered a methodology to create models which were simultaneously precise and interpretable. Now equipped with this methodology Ian created EP Analytic Solutions in late 2023 to provide Explicable Precision and to help create the theoretical foundation for data and actuarial science.
Ian’s doctoral research belongs to the field of operator theory and complex analysis. In 2017 he became an Associate of the Casualty Actuarial Society (ACAS), pairing his doctorate in pure mathematics with a professional credential used throughout the insurance industry.
That combination has defined his career. After brief early roles as a labor scheduling analyst at Sterling Jewelers and a programmer at National Interstate Insurance, he spent nearly six years at Farmers Insurance Group where he researched and built generalized linear and neural network models for operational problems and rate relativities. From August 2018 through December 2024, he led the data science division at MedPro Group, a medical professional liability insurer. In that role he directed teams of data scientists, actuaries, and analysts, working across thousands of variables and multiple lines of business on reserving, rating, and operational modeling.
Ian believes there should be an unbroken chain of reasoning from first principles to pressing “Enter” on a keyboard. His published research spans both pure mathematics and actuarial practice. His papers in various mathematical journals, including joint work with Steven Seubert, address cyclic vectors and the synthesis of diagonal operators. His 2014 paper marked a change in his research from the theoretical to the computational wherein he explicitly constructs polynomials satisfying a minimal growth condition. In the actuarial literature, The Mathematics Of On-Leveling (Casualty Actuarial Society E-Forum, Spring 2017) gives an explicit mathematical foundation for premium on-leveling, derives formulae for rate-book on-leveling and the parallelogram method, and argues that when data granularity permits, the parallelogram method should be set aside because it requires stronger assumptions. That paper is characteristic of his style: start from definitions, state assumptions, and only then write the working formula.
The educational aim of EP is a continuation of Ian’s time at University. For more than six years, first as a teaching assistant at the University of Akron and Bowling Green State University, then as adjunct faculty at the University of Findlay, Ian taught undergraduate mathematics. Course content ranged from freshman mathematics to abstract algebra. He has spoken at Amazon’s re:MARS, various Casualty Actuarial Society meetings, universities, and All Things Open. He has also appeared on the Bob Murphy Show to explain the use of infinity in higher mathematics and investigate the consequences of reserve ratios in banking. He has produced educational videos including the Explicable Precision video series, which develops methods – variable transformations, missing data techniques, grouping, and piecewise linear models related to MARS – intended to rival gradient boosting and random forests without surrendering interpretability. Related code is published on GitHub. The results of the research and education arm of EP form the basis of new code written to more quickly provide explicable models which are ever more accurate.