Bayesian Methods in Modern Marketing Analytics

Introduction

Bayesian methods have gained significant popularity in modern marketing analytics due to their ability to handle uncertainty, incorporate prior knowledge, and make accurate predictions. Unlike traditional statistical approaches, Bayesian methods provide a flexible framework that enables marketers to make data-driven decisions by combining observed data with prior beliefs or assumptions.

Event Description

In this webinar we discuss some of the most crucial topics in marketing analytics: media spend optimization through media mix models and experimentation, and customer lifetime value estimation. We approach these topics from a Bayesian perspective, as it gives us great tools to have better models and more actionable insights. We take this opportunity to describe our join with PyMC Labs in open-sourcing some of these tools in our brand-new pymc-marketing Python package PyMC Marketing.

Timestamps

00:00 Welcome

02:03 Webinar starts

02:32 Webinar's objective

03:04 Outline

04:05 Applied Data Science

05:12 Bayesian Methods

06:49 Geo-Experimentation

08:27 Time-Based Regression

10:26 Regression model in PyMC

12:04 Marketing measurement

13:34 Media Transformations (Carryover (Adstock) & Saturation)

15:50 Media Mix Model Target

16:24 MMM Structure

16:53 Media Contribution Estimation

17:13 Budget Optimization

18:18 PyMC-Marketing

19:25 PyMC-Marketing- More MMM Flavours

20:00 Customer Lifetime Value (CLV)

21:47 Continuous Non-Contractractual CLV

22:57 CLV Estimation Strategy

24:31 BG/NBD Assumptions

27:14 BG/NBD Parameters

27:50 BG/NBD Probability of Alive

28:40 Gamma-Gamma Model

29:12 BG/NBD Hierarchical Models

31:14 Causal Inference (Synthetic control)

32:10 Causal Inference (Difference-in-Differences and Regression Discontinuity)

32:39 Instrumental Variables

34:46 Cohort Revenue-Retention Modelling

38:21 Retention and Revenue component

41:02 Cohort Revenue-Retention Model

42:34 Revenue-Retention Predictions

43:11 References

44:25 Connect with PyMC Labs

44:50 Marketing analytics strategy consultation

47:36 PyMC Applied Workshop

48:58 Q/A There are so many parameters in MMM which are not identifiable ...

53:00 Q/A In the MMM how do you encode categorical control variables?

54:10 Q/A How to deal with latent variables?

57:34 Q/A If you observe the baseline uplift...How do you measure it in a Media mix model...?

59:15 Q/A How does it solve the cold start problem?

Resources

• pymc-marketing open source Python package
• Slides

Book a Free 30-Minute Consultation

Curious how these tools can be most effectively used in your particular situation? In this free 30-minute strategy consultation with us, we will:

• Review your current setup and any potential pain points you might have.
• Identify where there are areas for improvement.
• Define a plan of how Bayesian open-source tools can help you bring your marketing analytics to the next level.
• Free strategy consultation: calendly.com/twiecki/bayes
• If calendly slots are full, email: thomas.wiecki@pymc-labs.com