The Last Mile Problem How Data Science and Behavioural Science can Work Together IPAC Toronto March 4, 2015 James Guszcza, PhD, FCAS, MAAA Deloitte Consulting jguszcza@deloitte.com
Two overdue sciences Why do professional baseball executives, many of whom have spent their lives in the game, make so many colossal mistakes? It takes time and effort to switch from simple intuitions to careful assessments of evidence. Thaler and Sunstein review of Moneyball Many programmes and services are designed not for the brains of humans but of Vulcans. Thanks in large part to Kahneman and his many collaborators pupils and acolytes, this can and will change. Rory Sutherland, Ogilvy & Mather 2 Deloitte Analytics Institute 2011 Deloitte LLP
Classic examples of data science (aka Moneyball aka big data ) Predictive models can be used to: Hire more effective employees (Moneyball) Predict who is most likely to default on a loan (credit scoring) Predict who is most likely to crash their car and how badly (actuarial science) Predict next best offer, customer churn, lifetime value (marketing science) Predict recidivism (law) Identify episodes of waste, fraud, abuse (government) Identify unsafe workplaces (risk management) Identify physicians at highest risk of being sued for malpractice (risk management) Identify divorced parents most likely to lapse on child support payments Predict healthcare utilization Estimate risk of such disease states as diabetes, hypertension, heart disease Predict college grades using high school transcript data Forecast movie box office returns, political election results 3 Deloitte Analytics Institute 2011 Deloitte LLP
Data science a narrow frame A narrowly technical view of data science: DATA MODEL 4 Deloitte Analytics Institute 2011 Deloitte LLP
Data Science as a Strategic Capability All statistics is consulting Andrew Gelman STRATEGY DATA MODEL 5 Deloitte Analytics Institute 2011 Deloitte LLP
Data Science as a Strategic Capability All statistics is consulting Andrew Gelman STRATEGY DATA MODEL All models are wrong, some are useful George Box 6 Deloitte Analytics Institute 2011 Deloitte LLP
Our focus: the last mile problem Predictive models can point us in the right direction they can tell us whom to target but they don t tell us how to prompt the desired behaviour change. MODEL 7 Deloitte Analytics Institute 2011 Deloitte LLP
Our focus: the last mile problem Predictive models can point us in the right direction they can tell us whom to target but they don t tell us how to prompt the desired behaviour change. MODEL Furthermore: Behavioural insights without data analytics motivate one-size-fits-all intervention strategies. 8 Deloitte Analytics Institute 2011 Deloitte LLP
Our focus: the last mile problem Predictive models can point us in the right direction they can tell us whom to target but they don t tell us how to prompt the desired behaviour change. MODEL Furthermore: Behavioural insights without data analytics motivate one-size-fits-all intervention strategies. Can we do better by working together? 9 Deloitte Analytics Institute 2011 Deloitte LLP
Examples
Yes they did Motivating example: the 2012 Obama reelection campaign used predictive models to identify whom to target. 11 Deloitte Analytics Institute 2011 Deloitte LLP
Yes they did Motivating example: the 2012 Obama reelection campaign used predictive models to identify whom to target. It also used behavioural insights to more effectively act upon the predictive model indications. 12 Deloitte Analytics Institute 2011 Deloitte LLP
Supporting child support Predictive models are increasingly used to guide child support enforcement officers to noncustodial parents at highest risk of lapsing on their child support payments. Commitment cards could be field tested to prompt the desired behavior change. 13 Deloitte Analytics Institute 2011 Deloitte LLP
Push the worst, nudge the rest The city of New York uses predictive models to deploy building inspectors to the highest-risk buildings. Behavioural nudge tactics could be employed to ameliorate lesser risks that don t merit immediate physical inspections. similarly with health / sanitary inspections, tax audits, 14 Deloitte Analytics Institute 2011 Deloitte LLP
Let s keep ourselves honest Government agencies, tax authorities, insurance companies, regularly employ statistical fraud detection methods to guide fraud investigators to the most suspicious cases. But many forms of fraud are ambiguous or matters of degree ( soft fraud ) And fraud detection algorithms yield false positives Behavioural nudge tactics are soft interventions that are well-suited to such ambiguities. Invoke peer effects and people s internal reward systems not just hard incentives 15 Deloitte Analytics Institute 2011 Deloitte LLP
Driving behaviour change Actuaries now use telematics data to better segment and price insurance policyholders in terms of their utilization and riskiness. But could the data be used to create new products and services periodic or real-time reports that serve as behavioral nudges Ideas Detailed feedback reports to help student drivers learn and older drivers stay behind the wheel longer and safer Feedback prompting carbon footprint improvements through peer effects 16 Deloitte Analytics Institute 2011 Deloitte LLP
Personalized health coaching Lifestyle and medical data can used to predict individuals healthcare utilization and likelihood of various disease states. But once we ve identified the highest risks, what can be done to change behaviour? Health coaches are a promising behavioral strategy. Furthermore: analytics could be used to guide the hiring and matching of health coaches with patients. 17 Deloitte Analytics Institute 2011 Deloitte LLP