01. Customer Analytics (W1)

Problems need to solve

• Who are our customers?
• What do they need?
• What do they want?
• What should we do?

Knowledge needed

• Business Intelligence
• Predictive analytics
• Unsupervised methods
• Time series analysis
• Statistics
• Probability theory

Types of Analytics

• Descriptive Analytics – What has happened?
• Predictive Analytics – What might happen?
• Prescriptive Analytics – What should we do?

Type of research

• Descriptive Research – Aware of Problem
  What are our sales?
  Who are our customers?
• Exploratory Research – Ambiguous Problem
  Our sales are down – Why?
• Causal Research – Problem Clearly Defined
  Do we get higher sales if we change packaging?

Data collection

• Exploratory Research
  • Focus groups
  • Online communities

• Descriptive Research
  • Surveys
  • Self-reporting
  • Panel data
  • Scanner data
  • Mobile, Web data

• Causal Research
  • Experiments

Exploratory Research

Focus groups

• Qualitative research data gathering technique

• Moderated unstructured discussionbetween diverse group of people

• Members discuss some topics and allowed to influence each other

  Tips

  “Aha!” moment
  Slow and expensive

Online communities

Working with C Space, IKEA:

• Ran online communities in seven core, globally representative countries (USA, China, India, Denmark, Germany, Japan and Russia)
• Surveyed more than 21,000 people across 22 countries to robustly validate insights and hypotheses

Descriptive Research

Surveys

1. Approaches

   • Face-to-face
   • Mailout
   • Online

2. Survey Design

   • Do we ask right questions?
   • Do we collect right data?

3. Two important characteristics:

   • Predictive validity – data make good predictions for variables of interest
   • Test-retest reliability – if we are to re-measure, do we get the same result?

4. The factors need to consider for Survey Design

   • Mode of Data Collection
   • Impact of Survey Fatigue
   • The Effect of Survey Question Wording
   • How You Order Your Questions
   • Different Survey Question Formats
   • Accuracy of the Answers You Receive
   • Bias in Self-Reported Behavior
   • Survey Analysis Plan

   resources

Sampling Techniques

• Simple Random Sample
  each member of population has an equal chance
• Stratified Random Sample
  split population in groups (eg sex), then SRS from groups
• Cluster Sample
  organize population in clusters, then choose clusters and SRS from them
• Voluntary-Response Sample
  members of population who have chosen to respond
• Convenience Sample
  members of population from which data are easy to collect
• Systematic Sample
  every ${𝑛}^{𝑡ℎ}$ member of the population is selected

Sample Size

$CI=\hat{\mu }\genfrac{}{}{0ex}{}{+}{-}+z_{cr}\sqrt{\frac{{\sigma }^2}{N}}\Rightarrow N=(\frac{2z_{cr}\sigma }{D}{)}^2$
$CI=\hat{p}\genfrac{}{}{0ex}{}{+}{-}{z}_{cr}\sqrt{\frac{p(1-p)}{N}}\Rightarrow N=\frac{4z_{cr}^{2}p(1-p)}{D^2}$

Margin of error: $D=(C{I}^{+}-C{I}^{-})/2$

Self-reporting, panel data

• Panel of customers representing & different demographic groups
• Report all purchases
• Purchase trigged surveys
• Mobile surveys

data

Nielsen US Panel 2016

• 63,150 households
• 10,745,635 shopping trips
• 67,767,386 purchases
• 4,231,283 SKUs

Scanner data

Passive data collection

• Scanner data
• Media planning – radio, TV audience
• Social Media Analysis – Facebook, Twitter, Instagram
• Mobile data – Facebook, Foursquare, coupon services
• Web data – web logs, Google

Scanner data

Pros
Completeness
Timeliness
Accuracy

Cons

• Misses out on convenience stores and even some big retailers (Aldi, Whole Foods)
• Cannot make causal statements
• Don’t know behaviors and psychographics
• Don’t know the exact set of choices faced by the consumer at the time of decision.

Assignments

• An average size supermarket
• About 1,000,000 transactions per month
• 36 months of data
• Unique feature of the data – customer ID

Causal Research

Experiments and Field Tests

• Scientific testing where specific variables and hypothesis can be tested
• Controlled environment, where a set of variables are kept constant
• Invariable behavior between cause and effect to establish a cause-effect relationship

Correlation and Causation

• Correlation - relationship between two variables.
• Causation - one variable producing an effect in another variable.

Causal Inference: Three Requirements

• Correlation: Evidence of association between X and Y
• Temporal antecedence: X must occur before Y
• No third factor driving both: Control of other possible factors

Summary

• Types of analytics: descriptive, predictive, prescriptive
• Types of descriptive analytics:
  descriptive research, exploratory research, causal research
• Data collection for each type of research. Active/passive data collection
  • Focus groups, online communities
  • Survey, self-reporting, panel data, scanner data, online and mobile data
  • Experiments
• Correlation vs Causality

Reference

Tim's slides of Week 1