Gaining a Competitive Advantage using Analytics to Optimize your Digital Marketing

© 2020 Minitab, LLC.
AUDIO WEB:
Please make sure you have your computer audio system
activated and your speakers turned up.
QUESTIONS:
Please write your questions in the questions pane
at any time.
Con Menictas
Marketing Scientist
Strategic Precision
& Marketing
Abbie Wong
Regional Marketing
Manager
Minitab
Gaining a Competitive
Advantage:
Using Analytics to
Optimize Your
Digital Marketing

• 34 years of
experience
• Worked with Qantas
Airways, Woolworths
Supermarkets,
American Express
• PhD in Quantitative
Research Methods
Meet the Presenter:
Con Menictas
Managing Director
Strategic Precision

6
AGENDA

Agenda
No. Section Content Approx. Timing
1 Introduction
Marketing analytics
10 minsExamples of marketing analytics
What we discuss today
2 Classification
The regular approach
5 mins
New considerations
3 Selecting Models
Classical
7 mins
Machine Learning
4 Operationalising Models
Classical
7 mins
Machine Learning
5 Case Studies 3 Industry-Specific Applied Cases 7 mins
6 Minitab
Speed
3 mins
New Machine Learning
7 Q&A 5 mins
7

8
INTRODUCTION

Section 1: Introduction
The Philosophy of Digital Data
Overall Digital data theories view reality and cognition in a deterministic framework of RP information theory.
CAVEAT Not all and not necessarily more RP data are better in consistently predicting human behaviour.
Reason The ‘What’ and the ‘Why’.
9

History of Analytics
o Imposed rules
o Based on precedence
o Speculative prediction
Subjective
o Distributional assumptions
o Based on model fit
o Main effects
Statistical
o Distribution free
o Validation
o Main effects and interactions
Machine Learning
1 2 3
10

The Data Used in Digital Marketing Analytics
1P 1st party data are data generated via a direct business interaction with a customer e.g., sales data
2P 2nd party data are data associated with business interactions with customers e.g., survey data
3P 3rd party data are data external to the business e.g., financial data
Ancillary Consequential data such as cookie data.
11

o Time spent and pages visited
o New or returning customers
o Desktops or a mobile devices
Traffic
o
𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 𝐶𝐶 𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅
𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅
o
𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇 𝑈𝑈𝑈𝑈 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅
𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 𝐶𝐶 𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅
o
𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇 𝑈𝑈𝑈𝑈 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅
𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅
OR | CTR | TUR | UR
o Customer lifetime value CLV
o Retention
o Acquisition
Engagement
1 2 3
Where do digital marketing analytics focus?
12

13
HOW DO WE MEASURE SUCCESS
IN DIGITAL MARKETING?

1 How to Measure Success
o Offer Take Up: {0 = no,1 = yes}
o Cross-sell opportunity
o Up-sell opportunity
TU (Profitable Sales)
o Survival
o Survival by Market Segments
Life of the Campaign
o Inauthenticity (Me-Toos)
o Disinterest (Wrong Offer)
o Annoyance (Pestering)
Damage or Risk
1 2 3
3 Ways of Looking at Digital Marketing Success
14

Section 1: How to Measure Success
15
0.000.250.500.751.00
0 200 400 600 800 1000
Time
Digital Campaign = 1 Digital Campaign = 2
Points Acceleration Digital Campaign
Campaign Comparison

Section 1: How to Measure Success
16
0.2.4.6.81
0 1 2 3 0 1 2 3
Geo Sector 1 Geo Sector 2
Campaign 1 Campaign 2
surv
Time
Graphs by Campaign Area Targets

17
SPECIFIC FOCUS OF TODAY

o Y-var = Engineered digital
o X-vars = Age, income, FINs
o Z-var = Gender
Data
o Logistic Regression (Workhorse)
o Single CART Tree
Models
o Differences
o Malleability
Comparison
1 2 3
What we cover today
18

Section 1: The Data
19
Summary Statistics
Number Variable N N* Mean StDev CoefVar Minimum Maximum
1 (IV) Digital (TU) 7375 0 0.48515 0.49981 103.02 0 1
2 (DV) age 7375 0 48.259 18.232 37.78 20 92
3 (DV) income 7375 0 11.28 0.663 5.88 9.764 13.915
4 (DV) female 7375 0 0.23119 0.42162 182.37 0 1
5 (DV) exposure 7375 0 0.19423 0.33933 174.7 0 1.08
6 (DV) finseg 7375 0 4.7732 3.105 65.05 1 11
7 (DV) geoseg 7375 0 7.9706 4.1853 52.51 1 16
8 (DV) eduocc 7375 0 5.8799 2.9961 50.95 1 10
Note: In Six Sigma applications: the variables above can be substituted by any other
dependent and independent variables (binary or otherwise)

20
CLASSIFICATION

Section 2: Classification
21
The Regular Approach in Digital Marketing Analytics
o Logistic Regression.
o A great model but has limited tuning capability.
o Subject to assumptions e.g.:
1. Absence of multicollinearity; and
2. Lack of strongly influential outliers.
o Cannot be adjusted once productionised.
o Depth of the model is static.

Section 2: Classification
22
New and Very Exciting Addition to Minitab !
o The Single Decision Tree called CART.
o Very robust, and unlimited tuning.
o Not subject to assumptions.
o Main effects and n-way interactions.
o Can be adjusted once productionised, repeatedly.
o Intuitive tree structure
o Not affected by outliers.

23
SELECTING MODELS

Section 3: Selecting Models
24
MINITAB Logistic Regression

25
MINITAB Logistic Regression (Train or Exploratory)

26
MINITAB CART Tree

27
Exploratory CART Tree (Train or Exploratory)

28
Comparing Logistic to CART with Cross-Validation
Fold Logistic Train Logistic Test CART Train CART Test
1 0.7881 0.7878 0.8402 0.8374
2 0.7973 0.7917 0.8142 0.809
3 0.7907 0.7884 0.8346 0.8334
4 0.7917 0.7854 0.8334 0.8214
5 0.7897 0.7861 0.8385 0.8377
6 0.7898 0.7889 0.8328 0.8224
7 0.7913 0.7816 0.8373 0.8223
8 0.7986 0.7971 0.8377 0.8308
9 0.8036 0.7902 0.8188 0.8163
10 0.7939 0.7849 0.8349 0.8332
Mean 0.79347 0.78821 0.83224 0.82639
Increase in AUC 4.84%

29
Another CART Tree Feature

30
Common Sense Readability!

31
Why MINITAB Single Cart Trees Work Well
o Automatically handles a mixture of numeric, ordered and categorical data.
o No need for special handling or data preparation.
o Built-in feature selection.
o Is oblivious to outliers.
o Works with the rank order of numeric data: {1,2,3,4} and {1,2,3,400} are the same to CART.
o Elegantly handles missing values.
o Automatically adjusts for imbalanced data (rare outcomes).
o Sophisticated cross-validation.

32
Using 2nd and 3rd PARTY DATA
FOR MODEL IMPROVEMENT

33
Because 2nd and 3rd Party Data tell us the ‘WHY’ when Improving Model Fit for Digital Campaigns
o Two primary data sources:
1. Personal spend characteristics; and
2. Financial ability to spend.
o People proportionally spend between categories.
o It important to understand the ranking of spend categories.
o Spend category allocation is fueled by available funds and borrowed funds, the latter being less desirable.
o By understanding both spend characteristics and ability to spend, targeting for digital content becomes easier.

34
2nd and 3rd Party Census-Based Data: Geo-Specific Consumer Lifestyle and Predisposition to Spend
Total = 385 Micro-Segments (Only 1 of 55 segments groupings shown below), UK, AUS, USA
6% 0% 7% 9% 10% 6%
11% 10% 2% 12% 12% 12%
8%
14%
9%
12% 11% 12%
12%
19%
23%
38%
18% 20% 22%
17%
14% 20%
42%
16%
19% 16%
21%
13%
19%
12%
19%
17% 15%
19%
22%
15%
2%
16% 11% 13% 16%
5%
100
Alcohol
100
Clothing Furnishings
4%
GroceriesEducation Health Bills
A. YOUNG INDEPENDENTS : 18-44
F. EARLY SENIORS: 65-74
B. YOUNG FAMILIES : 18-34
C. MIDDLE AGED FAMILIES : 35-44
D. MATURE FAMILIES : 45-64
E. MATURE INDEPENDENTS : 45-64
100
G. LATE SENIORS: 75+
100%
100 100 100 100

FIN Related Pressure
City/Town Features
Map Layers
35
3rd Party Survey Data: FINANCIAL Location-Specific Individual-Level Data

36
ASSESSING MODEL PERFORMANCE
BY ADDING 2nd and 3rd PARTY DATA

37
Comparing Models With FINANCIAL Data Added to the Models
Fold Logistic Train Logistic Test CART Train CART Test
1 0.7913 0.7866 0.8742 0.8739
2 0.7974 0.7868 0.8821 0.8643
3 0.7932 0.7887 0.8783 0.8754
4 0.793 0.7876 0.8777 0.8752
5 0.7919 0.7861 0.8766 0.8755
6 0.7917 0.7881 0.8616 0.8513
7 0.7998 0.7846 0.8869 0.8721
8 0.7914 0.7891 0.8776 0.8736
9 0.8049 0.7829 0.8624 0.8616
10 0.7991 0.7859 0.8732 0.8674
Mean 0.79537 0.78664 0.876875 0.86903
Increase in AUC 10.47%

Section 4
38
Operationalising Models

Section 4: Operationalising Models
Logistic CART
Comparing Logistic to CART
39
where, Y’ = -10.211
- 0.0503 * age
+ 1.1031 * income
- 0.2354 * exposure
+ 0.2158 * female
P Y′
=
𝑒𝑒𝑒𝑒𝑒𝑒(𝑌𝑌′)
1 + 𝑒𝑒𝑒𝑒𝑒𝑒(𝑌𝑌𝑌)

SELECT
(EXP(-10.211 + age*-0.0503 + inc*1.1031 + exp*-0.2354 + fem*0.2158))
/
(1+(EXP(-10.211 + age*-0.0503 + inc*1.1031 + exp*-0.2354 + fem*0.2158)));
Logistic Model in SQL
Operationalising The Logistic Model
40

CASE
WHEN age <= 21.5 THEN 1
WHEN 24.5 < age <= 30.5 THEN 1
WHEN 30.5 < age <= 64.5 AND income > 11.9177 THEN 1
WHEN 30.5 < age <= 34.5 AND 11.3788 < income <- 11.9177 THEN 1
WHEN 41.5 < age <= 64.5 AND 11.3788 < income <- 11.9177 THEN 1
WHEN 30.5 < age <= 64.5 AND income <= 11.3788 AND female = 1 THEN 1
WHEN age > 64.5 THEN 0
WHEN 21.5 < age <= 24.5 THEN 0
WHEN 30.5 < age <= 64.5 AND income <= 11.3788 AND female = 0 THEN 0
WHEN 34.5 < age <= 41.5 AND 11.3788 < income <= 11.9177 THEN 0
ELSE NULL
END AS Y-HAT;
CART Model in SQL
Operationalising CART
41

42
CASE STUDIES

Section 5: Digital Marketing Case Studies
43
o The Problem
o Develop and predict 256 micro segments of water customers.
o Cross-sell partner products e.g., water saving shower heads.
o The Solution
o Developed a CART tree to map segment migratory patterns.
o Optimized digital content under a partner model.
o The Result
1. Increased take-up via digital content on dormants.
2. Increased prediction of migratory patterns.
3. Increased profitable sales.
Government Water Company (Water Utility): Predicting 256 Segments for Digital Marketing Activation
Case Study 1
$2,710,000
YR1 YR2
$3,640,000
YR3
$1,420,000
+91%
+34%
YR1 YR2 YR3
$1,250,000
$1,920,000
$2,270,000+54%
+18%
Dormant Activation
Improvement
Migration Prediction
Improvement
N=185K
N=340K
N=493K
N=125K
N=187K
N=227K

44
o The Problem
o A global card company was losing early high-spend stimulation
by low income declarations for new members for up to 2 years.
o Develop a model that identifies high-spend stimulation potential
from low income declarers for digital content.
o The Solution
o Develop a CART tree to identify masked declaration potential
from low-level member application data.
o The Result
o Increased identification over 2 years.
Finance Industry: Overcoming Non-Declaring High Net Worth Customers
Case Study 2
Extent of Identifying High-Spenders on Application
YR 0 YR 1 YR 2
7% identified
24% identified
29% identified+342%
+21%
$4.2M
$14.1M
$19.7M

45
o The Problem
o HNW airline loyalty members who used to earn points by flying
have reduced their spend and hence their points-earn rate.
o How do we redirect HNWs to smaller and more frequent spend?
o The Solution
o Develop a behavioral change model to optimally redirect HNWs
towards bird-in-the-hand earn and redemption behavior.
o The Result
o 6 month increase in behavioral change.
o ~$150M in Earn; ~$110M in Redemption (Burn)
Large Scale Airline Loyalty Program: Behavioral Change via Digital Marketing
Case Study 3: COVID-19 Impact
Apr-20 Sep-20
31B points
69B points
2B points
98B points
-94%
+42%
Points Earn Points Burn
AIR
Non-AIR
Sep-20Apr-20
37B points
64B points
8B points
92B points
-78%
+45%
AIR
Non-AIR
+$147.9M
+$108.8M

46
SPEED

Section 6: Speed
47
Why Minitab is Lightning Fast!
o Click operated
o Intuitive GUI
o Essential outputs
o No coding unless you want to
o Lightning Fast Estimations
o Report ready outputs
o The analyst’s and consultant’s no frills Go-To

Section 6: Speed
48
CART out of the box firepower!: 3 clicks and you have your CART!
1
2
3

49
THANK YOU

Questions?

The Minitab Virtual Launch Event
Join Minitab CEO Jeffrey T. Slovin and other
Minitab experts as we unveil new solutions that
will enable and accelerate your Digital
Transformation journey.
• Date: Thursday, October 22
• Time: 130pm AEDT / 330pm NZDT
See all the details at:
https://info.minitab.com/resources/webinars/
virtual-launch-event-acceleration-digital-
transformation
Scan the QR code
to register

Upcoming Training
Virtual Training
Take our Minitab Signature Trainings virtually now from anywhere and learn directly from
our expert statisticians. Upcoming training include:
• Improving Business Processes: Statistical Quality Analysis for Service Quality
• November 2 – 6
• Improving Manufacturing Processes
• November 30 – December 2
• Predictive Analytics: Machine Learning
• December 3 - 4
Onsite Training
• Schedule time with our trainers to review your data
to address your specific business challenges.
Or scan the QR
Code
See the details at
http://www.minitab.com/training/training/

On-Demand Webinars
Our Webinar series features the latest tools and tips presented by Minitab Experts
help you continue learning and working smarter.
Highlights:
• The Age of Analytics: Data Driven Insights in the Services Industry
• Quality as a Competitive Advantage: How the LEGO Group Work with Quality
and
Product Safety
• Statistical Solutions to Help You with the 5 FDA Medical Devices Stages
Get the details here
https://info.minitab.com/resources/webinars/webinar-wednesdays-apac
Scan the
QR Code

Our Approach: More Than Business Analytics… Solutions Analytics
Software
Services
Training
Learn first-hand by attending public or
customized trainings in your facilities
according to your requirements.
Statistical
Consulting
Personalized help with statistical
challenges from collecting the right data
to interpreting analysis more.
Support
Assistance with installation,
implementation, version updates
and license management.
Master statistics and
Minitab anywhere
with online training
Machine learning and
predictive analytics
software
Start, track, manage
and execute
improvement projects
with real-time
dashboards
Powerful statistical
software everyone
can use.
Data Analysis Predictive Modeling Visual Business Tools Project Oversight
Visual tools to
process and product
excellence.
Online Training
Solutions analytics is our integrated approach to providing software and services that enable organizations to make better decisions that drive business excellence.

Thank you

Gaining a Competitive Advantage using Analytics to Optimize your Digital Marketing

More Related Content

Similar to Gaining a Competitive Advantage using Analytics to Optimize your Digital Marketing (20)

Recently uploaded (20)

Gaining a Competitive Advantage using Analytics to Optimize your Digital Marketing