Week 10 Module Wrap-up

Author

Affiliation

Dr Wei Miao

UCL School of Management

1 Intro to Marketing Analytics

1.1 Week 1: Marketing Process

• Situation analysis (5C analysis)

1.2 Week 1: Break-Even Analysis

• Break-even analysis is essential to any business activity
  • For business campaigns: Break-even quantity (BEQ) and Net present value (NPV)
  • For customers: Customer lifetime value (CLV)
• BEA is essentially cost-benefit analysis
  • PineApple case
  • i-basket case

2 Descriptive Analytics

2.1 Week 2: Data Wrangling with dplyr

• Data manipulation with dplyr
  • basic operations: filter, mutate, select, arrange
  • group aggregation: group_by
  • multi-data joining: left_join
• Tesco case study: Preliminary customer analysis using dplyr

2.2 Week 2: Hey, I’m Wei, and I’m a musician!

2.3 Week 2: Hey, I’m Wei, and I’m a Youtuber!

3 Predictive Analytics

3.1 Week 3: Unsupervised Learning for Customer Segmentation

• Unsupervised learning such as K-means clustering help classify individuals into different segments.
• We then decide which segment(s) to serve based on our business objective.

3.2 Week 4: Supervised Learning for Customer Targeting

• Unsupervised learning is often not the most powerful tool for STP.

• Supervised learning models learn the relationship between outcome \(Y\) and \(X\) and can make individualized prediction.

3.3 Week 4: Fundamental Tradeoffs in Supervised Learning

• Accuracy-interpretability tradeoff

  • linear regression (high interpretability, low accuracy)

  • decision tree and random forest (good interpretability, good accuracy)

  • deep learning (no interpretability, high accuracy, term 2)

• Bias-variance trade-off

3.4 Week 4: Application in Marketing: Personalized Targeting

• With targeted marketing from supervised learning, we can effectively reduce marketing costs and boost the ROI.
  • Case study: Improving Marketing Efficiency Using Predictive Analytics for Tesco case
  • 2nd assignment: Amazon Prime case

4 Causal Inference

4.1 Week 5: Why Causal Inference Matters?

• Managers easily make costly mistakes if they do not understand causal inference.

4.2 Week 5: I’m Wei and I’m from Hogwarts

4.3 Week 6 & 7 : OLS Regression and Marketing Mix Modeling

• Simple linear regression from secondary data can give causal inference if and only if all confounding factors have been controlled in the regression. In reality, this never happens, so linear regression can never give causal effects.

• In practice, companies often use linear regression to build marketing mix modeling, in order to set optimal prices for profit maximization (profit is often considered a quadratic function of price).

4.4 Week 8: Endogeneity and Instrument Variables

• Endogeneity
  • Omitted variable bias
  • Reverse causality/Simultaneity
  • Measurement error
• An instrument variable can give causal inference, which satisfies
  • Exogeneity: instrument is exogeneous and beyond individual control
  • Relevance: instrument affects X
  • Observable (implicit)

4.5 Week 8: Endogeneity and Instrument Variables

1. Explain \(X\) with \(Z\). The predicted \(X\), \(\hat{X}\), is uncorrelated with the error term \(\epsilon\) in the original regression.
   • Important: other control variables are also considered exogenous and should be included in both stages.

\[ X_{i}=Z\eta+\epsilon_{i} \]

2. Use the explained part \(\hat{X}\) to explain \(y\), now \(\hat{X}\) is exogenous and can give us causal inference.
   • COVID-19 case study: The causal impact of COVID-19 on Uber Driver Decision
   • third assignment

\[ y_{i}=\hat{X} \beta+\varepsilon_{i}, \quad \operatorname{cov}\left(\hat{X}_{i}, \varepsilon_{i}\right) = 0 \]

4.6 Week 9: Regression Discontinuity Design

• Causal effect of “Distinction” on students’ salaries
• Regression Discontinuity in Time (RDiT)

4.7 Week 9: Difference-in-Differences

• A new policy/regulation (GDPR, lockdown, etc.) and we have a control group which remains unaffected

4.8 One Causal Question, Many Solutions

• If we have historical data on number of restaurants on UberEat in each month, and the total number of orders in each month, can we get the causal effect?

\[ NumOrders_t = \beta_0 + \beta_1 NumRestaurants_t + \epsilon_t \]

• Solutions
  • RCT
  • Instrumental Variable
  • DiD
  • RDD
  • RDiT

5 Concluding Remarks

5.1 10 Weeks Not Enough?

• I love new challenges so my door is always open even after the class/program is over

• More learning materials

  • Optional reading materials in each week
  • I will keep uploading R tutorials/data analytics tools tutorials on my Youtube channel. It’s never too late to subscribe!

5.2 What I learned

• Impressed with your perseverance and willingness to learn
  • My bestie predicts you would chase me out of the classroom for making you learn Marketing, R, and so many new models at the same time
• You’ve made me very proud:

  • It gives me a huge sense of achievement to see that you are able to apply the tools learned in various scenarios!

  • It gives me a weird sense of achievement to receive and answer questions for other modules :))

  • R is the best language!!!! No Disagreement Allowed!!!!!!!!

5.3 Looking into the Future

• Any case study in the marketing module can be used for your term 3 dissertation project.
  • Welcome to talk to me about your ideas; always happy to supervise
• [IMPORTANT]: Start early for your job market; use alumni resources (Linkedin) to network.
• Let’s stay in touch even after graduation!

5.4 Thank you for being the BEST Students I can ever dream of!!

Thank you so much for your hard work and your support of the Marketing Analytics module throughout the term!

Though probably in 5 years you may have forgotten everything learned but only remember the following

• A module leader with a bubble tea belly uploads lousy weekly videos
• but he tries his best to be a good musician, Youtuber, magician, and stand-up comedian (and most importantly a good lecturer)
• A lame senior named Tom, who messed up everything and made you do so many assignments