CGT Presentation_10.30.13_Business Analytics for Decision Makers

Transcript

1. 1 | January 24, 2014 | © Matthew A. Waller, University of Arkansas Business Analytics for Decision Makers Matthew A. Waller, Ph.D. Professor & Chair Department of Supply Chain Management Dave Shuman VP Product Strategy Orchestro
2. 2 | January 24, 2014 | © Matthew A. Waller, University of Arkansas ORCHESTRO Founded in 1999 as Consumer Products Data Platform to create value from exponential growth of industry data. Global Headquarters: McLean, VA Global Center of Excellence: Rogers, AR Global Support: Poland, India, and China Annually process 1MM+ items, through 69 retailers, 5 syndicated data sources, 50+ distributors, 150 billion points of distribution
3. 3 | January 24, 2014 | CONFIDENTIAL Data Sources • Retailer Direct • Syndicate • Master Data • eCommerce • Promotions • Social / Sentiment • Broker Data • Mobile Data • Loyalty Data • Planograms • Demographics • Weather • Economic Data • Basket • Distributor Insights for • Sales • Category • Supply Chain • Marketing ShelfSense PromoSense DemandSense Automated Aggregation Adaptive Analytics Actionable Applications Data Enrichment Predictive Algorithms Business Rules Processing
4. 4 | January 24, 2014 | © Matthew A. Waller, University of Arkansas
5. 5 | January 24, 2014 | © Matthew A. Waller, University of Arkansas “I SEARCH ON GOOGLE WHEN I’M AFRAID OF WHAT THE MARKET IS ABOUT TO DO.” 2007 2013
6. 6 | January 24, 2014 | © Matthew A. Waller, University of Arkansas GOOGLE SEARCHES FOR “PREDICTIVE ANALYTICS” SINCE 2004 http://www.google.com/trends/explore#q=%22data%20science%22%2C%20%22data%20scientist%22&cmpt=q referenced March 22, 2013 Is anyone leveraging big data?
7. 7 | January 24, 2014 | © Matthew A. Waller, University of Arkansas GOOGLE SEARCHES FOR “DATA SCIENCE” AND “DATA SCIENTIST” SINCE 2004 http://www.google.com/trends/explore#q=%22data%20science%22%2C%20%22data%20scientist%22&cmpt=q referenced March 22, 2013
8. 8 | January 24, 2014 | © Matthew A. Waller, University of Arkansas IS ANYONE REALLY PAYING ATTENTION? Supply Chain Management Big Data They Are Now! Google Searches 2004 2013
9. 9 | January 24, 2014 | © Matthew A. Waller, University of Arkansas Correlation = 0.79 BEST PRACTICES & INTERMODAL
10. 10 | January 24, 2014 | © Matthew A. Waller, University of Arkansas RELATIONSHIP BETWEEN NUMBER OF VARIABLES AND NUMBER OF FALSE POSITIVES WHEN THE PROBABILITY OF A GIVEN FALSE POSITIVE IS 0.01 Number of Variables Expected Number of False Positives
11. 11 | January 24, 2014 | © Matthew A. Waller, University of Arkansas MOTIVATIONS People make bad decisions… Demand Orchestration decisions are no exception
12. 12 | January 24, 2014 | © Matthew A. Waller, University of Arkansas
13. 13 | January 24, 2014 | © Matthew A. Waller, University of Arkansas MOTIVATIONS What’s driving the irrational behavior so often observed in decision making? Can specific decision heuristics be identified? (security analysts—future returns of stocks) Can decision making be improved?
14. 14 | January 24, 2014 | © Matthew A. Waller, University of Arkansas OVERVIEW • Metrics matter – the same metrics measured in different ways can generate different behavior • Information matters – be careful about “other” information • Understanding theory can make a significant difference and experience doesn’t always help • People are very limited in their ability to apply useful information, even in small quantities – quantitative models can make a significant difference • Personality, motivation, and knowledge matter
15. 15 | January 24, 2014 | © Matthew A. Waller, University of Arkansas JUDGMENTAL ADJUSTMENT OF STATISTICAL FORECASTS (CUNEYT EROGLU) • A national quick service restaurant chain • Study included retail locations in 7 states • 702 store managers over 12 month period • Measurements of accuracy and biases
16. 16 | January 24, 2014 | © Matthew A. Waller, University of Arkansas JUDGMENTAL ADJUSTMENT OF STATISTICAL FORECASTS • On average, judgmental adjustments improve forecast accuracy. – Average improvement in APE = 19% (relative) (APE: Absolute percentage error) • However, forecasting performance depends on • Individual forecaster (personality, motivation, etc) • Task design (incentives, timing, training, etc)
17. 17 | January 24, 2014 | © Matthew A. Waller, University of Arkansas JUDGMENTAL ADJUSTMENT OF STATISTICAL FORECASTS 0 50 100 150 200 -50% -30% -10% 10% 30% 50% Average Improvement in Accuracy Number of Forecasters Absolute Improvement
18. 18 | January 24, 2014 | © Matthew A. Waller, University of Arkansas JUDGMENTAL ADJUSTMENT OF STATISTICAL FORECASTS (A FEW RESULTS) • Personality Variables ( + ) Conscientiousness ( - ) Neuroticism • Motivational Variables ( + ) Intrinsic motivational orientation • Task Design • Optimal time window for better adjustments (2 days out to 13 days out) • Understanding undesirable effects of poor forecasts
19. 19 | January 24, 2014 | © Matthew A. Waller, University of Arkansas
20. 20 | January 24, 2014 | © Matthew A. Waller, University of Arkansas FRAMING EFFECTS & PROSPECT THEORY • Framing Effects [Kahneman & Tversky (1979)] The way in which a decision context is framed can have a profound impact on the decision made
21. 21 | January 24, 2014 | © Matthew A. Waller, University of Arkansas BACKGROUND – FRAMING • Prospect theory predicts that the frame of the feedback provided and/or the measures used to evaluate performance will impact behavior – Gains frames incite risk averse behavior – Loss frames incite risk seeking behavior • Tversky and Kahneman (1981)
22. 22 | January 24, 2014 | © Matthew A. Waller, University of Arkansas • Imagine that the U.S. is preparing for the outbreak of an unusual foreign disease, which will kill 600 people. • Two alternative programs have been proposed. Assume that the exact scientific estimates of the consequences are as follows… EXAMPLE
23. 23 | January 24, 2014 | © Matthew A. Waller, University of Arkansas EXAMPLE Treatment A: 200 people will be saved Treatment B: 1/3 probability that 600 will be saved; 2/3 probability no one will be saved Treatment C: 400 people will die Treatment D: 1/3 probability that no one will die; 2/3 probability that 600 will die Expected values of A and C; B and D are the same!!! Survivors – 200 Deceased - 400 Survivors – 200 Deceased - 400 Survivors – 1/3 chance of 600, 2/3 chance of 0 Deceased – 1/3 chance of 0, 2/3 chance of 600 Survivors – 1/3 chance of 600, 2/3 chance of 0 Deceased – 1/3 chance of 0, 2/3 chance of 600 72% 28% 22% 78%
24. 24 | January 24, 2014 | © Matthew A. Waller, University of Arkansas FRAMING EFFECTS • People are risk-seeking when decisions are framed as a loss But… • People are risk-averse when decisions are framed as gains "We have an irrational tendency to be less willing to gamble with profits than with losses.." Tvede (1999)
25. 25 | January 24, 2014 | © Matthew A. Waller, University of Arkansas BEHAVIORAL ISSUES Anchoring Effects Decision makers sometimes anchor their decisions on a certain, (often arbitrary), point and then adjust that amount insufficiently
26. 26 | January 24, 2014 | © Matthew A. Waller, University of Arkansas SUMMARY • Decision makers sometimes anchor their decisions on a certain (often arbitrary) point and then adjust that amount insufficiently • Understanding theory can de-bias decision makers (experience doesn’t always help) • Bounded rationality can be addressed with quantitative models • Personality, motivation, and knowledge make a difference
27. 27 | January 24, 2014 | © Matthew A. Waller, University of Arkansas THE DATA SCIENTIST Effectiveness of Data Scientist Domain Knowledge Narrow Set of Analytical Skills Broad Set of Analytical Skills
28. 28 | January 24, 2014 | © Matthew A. Waller, University of Arkansas ORCHESTRO-TO-GO
29. 29 | January 24, 2014 | © Matthew A. Waller, University of Arkansas ORCHESTRO-TO-GO
30. 30 | January 24, 2014 | © Matthew A. Waller, University of Arkansas ORCHETRO-TO-GO
31. 31 | January 24, 2014 | © Matthew A. Waller, University of Arkansas