product cannibalization
play

Product Cannibalization A Prototypical Marketing Science Problem - PowerPoint PPT Presentation

The webinar will start at: 13:00:00 The current time is: 13:00:49 Central Daylight Time UTC-5 Product Cannibalization A Prototypical Marketing Science Problem Introduction Your Hosts Today Stefan Conrady stefan.conrady@bayesia.us


  1. The webinar will start at: 13:00:00 The current time is: 13:00:49 Central Daylight Time UTC-5 Product Cannibalization A Prototypical Marketing Science Problem

  2. Introduction Your Hosts Today • Stefan Conrady stefan.conrady@bayesia.us • Stacey Blodgett stacey.blodgett@bayesia.us BayesiaLab.com 2

  3. Today’s Program Motivation & Background • Definitions • Introductory Example Representation • Conceptual Framework: Bayesian Networks • Probabilistic Reasoning Learning, Estimation, and Inference • Causal Reasoning? • Unsupervised Learning • Disjunctive Cause Criterion • Assign Utilities • Evaluate Policies All Fictional Numbers stefan.conrady@bayesia.us 3

  4. Webinar Slides & Recording Available stefan.conrady@bayesia.us 4

  5. Motivation & Background Definitions • Typically, a new product adversely affects the sales of existing products: • If it affects your competitor’s products, it’s Conquest • If it affects your own products, it’s BayesiaLab.com 5

  6. www.BayesiaLab.com 6 2 3 - M a r - 1 8 6

  7. Motivation & Background Introductory Example: 2000 BMW X5 • First SUV in the BMW product portfolio. X5 BayesiaLab.com 7

  8. Motivation & Background Introductory Example: New BMW X3 vs. Existing BMW X5 • New, smaller X3 launched in 2004 Cannibalization? X3 X5 Product B Product A BayesiaLab.com 8

  9. Bayesian Network Representation

  10. Bayesian Network Representation Conceptual Network Product B causes P(Sales A |Sales B ) lower sales of P(Sales B ) Product A “Cannibalization” + – BayesiaLab.com 10

  11. Bayesian Network Representation Obvious, as we encoded that as our domain knowledge Inference into the network. • Computing the cannibalization effect C of Product B on Product A: • C(B A) = -0.3 (unit effect) Existing Product A Existing Product A Mean: 0.900 Dev: 0.831 Mean: 1.200 Dev: 0.748 Value: 1.200 Value: 0.900 (-0.300) 40.00% 0 20.00% 0 40.00% 1 30.00% 1 30.00% 2 40.00% 2 New Product B New Product B Mean: 1.000 Dev: 0.000 Mean: 0.000 Dev: 0.000 Value: 1.000 (+1.000) Value: 0.000 0.00% 0 100.00% 0 100.00% 1 0.00% 1 0.00% 2 0.00% 2 BayesiaLab.com 11

  12. Bayesian Network Representation Can’t we do this in Excel? BayesiaLab.com 12

  13. Motivation & Background Example: BMW Portfolio of “Utility - Type” Vehicles in 2018 All products are cannibalizing each other! BayesiaLab.com 13

  14. Bayesian Network Representation A Fully Connected Network? ? Can we specify it? No. Can we machine-learn it? Perhaps. BayesiaLab.com 14

  15. Learning & Estimating Cannibalization

  16. Learning & Estimating Cannibalization Couldn’t we just ask auto buyers? BayesiaLab.com 16

  17. Learning & Estimating Cannibalization Understanding Cannibalization by Other Means? • Trade-Ins • New and old product not comparable • Auto Buyer Surveys (2 nd Choice) • Respondents tend to exaggerate their counterfactual choice (“I would have bought the convertible, but we need the third row.”) • Choice Experiments • Hypothetical choices are noncommittal • Expensive to conduct BayesiaLab.com 17

  18. stefan.conrady@bayesia.us 18

  19. Map of Analytic Modeling & Reasoning Data Model Source Theory Theory Description Prediction Explanation Simulation Attribution Optimization Model Purpose Association/Correlation Causation BayesiaLab.com 19

  20. Map of Analytic Modeling & Reasoning Data Model Source Theory Description Prediction Explanation Simulation Attribution Optimization Model Purpose Association/Correlation Causation BayesiaLab.com 20

  21. Learning & Estimating Cannibalization A Fictional Case Study

  22. Learning & Estimating Cannibalization Case Study Question: • What is the cannibalization effect of B on A, C, and D? D C A B BayesiaLab.com 22

  23. Learning & Estimating Cannibalization Daily Sales Data Objective: To machine-learn a Bayesian network model from the sales data. BayesiaLab.com 23

  24. A desktop software for: encoding • learning • editing • performing inference • analyzing • • simulating optimizing • with Bayesian networks. BayesiaLab.com 24

  25. Data Import Wizard BayesiaLab.com 25

  26. Variable Type Definition BayesiaLab.com 26

  27. Discretization BayesiaLab.com 27

  28. Unconnected Network BayesiaLab.com 28

  29. Unsupervised Learning Using the EQ Algorithm BayesiaLab.com 29

  30. How can we use this network to calculate the causal effect of B on A, C, and D? Counterintuitive arc directions! Final Network BayesiaLab.com 30

  31. Disjunctive Cause Criterion BayesiaLab.com 31

  32. Disjunctive Cause Criterion VanderWeele and Shpitser (2011) Cannibaliz ed Product • “We propose that control be made for any [pre -treatment] covariate that is either a cause of treatment or of the outcome or both.” Confounder Cannibaliz ing Product Implementation in BayesiaLab: IMPORTANT ASSUMPTION: Likelihood Matching on Confounders in Direct Effects Analysis NO UNOBSERVED CONFOUNDERS  Causal Effect, i.e., the Cannibalization Rate BayesiaLab.com 32

  33. Map of Analytic Modeling & Reasoning Data Model Source Confounders Theory Description Prediction Explanation Simulation Attribution Optimization Model Purpose Association/Correlation Causation BayesiaLab.com 33

  34. We need to define confounders and non-confounders . By default, all nodes are confounders . Final Network BayesiaLab.com 34

  35. Computing the Direct Effect of B on A BayesiaLab.com 35

  36. Direct Effect of B on A BayesiaLab.com 36

  37. Direct Effect of B on C BayesiaLab.com 37

  38. Direct Effect of B on D BayesiaLab.com 38

  39. Adding a Decision Node BayesiaLab.com 39

  40. Adding Utility Nodes BayesiaLab.com 40

  41. Policy “B”: Utilities=90.285 Comparing Policies “B” vs. “No B” BayesiaLab.com 41

  42. Policy “No B”: Utilities=98.321 Comparing Policies “B” vs. “No B” BayesiaLab.com 42

  43. VR In Conclusion… 43

  44. Webinar Series: Friday at 1 p.m. (Central) Upcoming Webinars: • March 30 Good Friday — No Webinar • April 6 t.b.d. • April 13 t.b.d. Register here: bayesia.com/events stefan.conrady@bayesia.us 44

  45. BayesiaLab.com 45

  46. User Forum: bayesia.com/community BayesiaLab.com 46

  47. BayesiaLab Trial Try BayesiaLab Today! • Download Demo Version: www.bayesialab.com/trial-download • Apply for Unrestricted Evaluation Version: www.bayesialab.com/evaluation BayesiaLab.com 47

  48. BayesiaLab Courses Around the World in 2018 • April 11 – 13 • August 29 – 31 Sydney, Australia London, UK • May 16 – 18 • September 26 – 28 Seattle, WA New Delhi, India • June 26 – 28 • October 29 – 31 Boston, MA Chicago, IL • July 23 – 25 • December 4 – 6 San Francisco, CA New York, NY Learn More & Register: bayesia.com/events stefan.conrady@bayesia.us 48

  49. San Francisco Introductory BayesiaLab Course in San Francisco, California July 23 – 25, 2018 BayesiaLab.com 49

  50. 6 th Annual BayesiaLab Conference in Chicago November 1 – 2, 2018 Chicago BayesiaLab.com 50

  51. Thank You! stefan.conrady@bayesia.us BayesianNetwork linkedin.com/in/stefanconrady facebook.com/bayesia BayesiaLab.com 51

Recommend


More recommend