Marketing Question

Digital marketing analytics
Second Edition

Chapter 6
Digital Analysis:
Ecosystem

Copyright © 2022, 2018, 2016 Pearson Education, Inc. All Rights Reserved

Ecosystem Analysis
●

Why does this platform exist and what role does it play for my audience?

●

Where is it and how does my audience discover it?

●

What’s the priority level for this platform? Is it an immediate near-term
priority or a longer-term opportunity instead?

●

Which audience segments use this platform? Why, and for what
purpose?

●

What is the benefit to the brand for being on this platform?

●

What brand touchpoints does it have? Is it isolated by itself or integrated
with others?

Ecosystem Analysis
●

1. Do a digital channel inventory.

● 2. Identify the volume of conversations.

● 3. Determine volume of traffic.
● 4. Determine content relevance.
● 5. Discover brand presence.
● 6. Prioritize platforms.

Figure 6.1 EMC digital ecosystem map example.
From Digital Marketing Analytics: Making Sense of Consumer Data in a Digital World, 2/e by Chuck Hemann and Ken Burbary (0789759608) Copyright
© 2018 Pearson Education, Inc. All rights reserved

Figure 6.2 EMC digital ecosystem analysis map example.
From Digital Marketing Analytics: Making Sense of Consumer Data in a Digital World, 2/e by Chuck Hemann and Ken Burbary (0789759608) Copyright
© 2018 Pearson Education, Inc. All rights reserved

Digital marketing analytics
Second Edition

Chapter 7
Return on Investment

Copyright © 2022, 2018, 2016 Pearson Education, Inc. All Rights Reserved

Return on Investment
ROI is the gain from any spending minus the cost of the investment
divided by the cost. This calculation yields a percentage either gained
or lost from your investment.
Return on Engagement (ROE) – measures the effect your organization’s digital
marketing activities have on engagement rates.

Facebook—The number of engaged users is calculated by adding the number
of likes on a post, the number of comments on a post, and the number of
shares on a post, and then dividing that sum by the number of fans (or likes).

Twitter—The number of engaged Twitter users is calculated by adding the
number of replies to the number of retweets and likes, and dividing it by the
number of followers.

YouTube—There are a couple ways to calculate engaged YouTube users.
The first is to add the number of comments, the number of ratings, and the
number of likes, and then divide that sum by the number of views of a particular
video. Another variation is to add those same engagement metrics and divide
the sum by the number of subscribers.

Return on Influence
Reach: Conversations or content with significant distribution or visibility
Resonance: Conversations or content that influences hearts and minds

●

Tweets, retweets, comments, and likes are not transactions

●

The fan/follower value calculation is flawed

●

Not all fans are created equal

●

Changing behavior is important but not inherently financial

Return on Experience

● Return on experience is gaining more and more attention as an
interesting measure of success due to the emergence of mixed
reality (digital + physical).
● Social media will also continue to be a contributing factor to this
measurement approach. The foundational idea here is to quantify
the brand interaction of an audience at scale.

Top-Down Revenue Measurement Approaches
● Anecdote—This is probably the most common of the three, and it
involves a verbal “share” of a relationship between a social media
activity and a sale.
● Correlation—A correlation analysis takes a specific type of
behavior and tries to establish a relationship between it and some
other activity.
● A/B, multivariate testing—In this type of analysis, a marketer
attempts to understand the effectiveness of two versions of some
type of content (for example, a web page, a marketing email, or a
social media advertisement) in order to determine which has the
best response rate. You can think of multivariate testing as many
different A/B tests happening simultaneously.

Revenue Measurement
● Anecdote analysis is likely to be the least concrete of the models we talk about
here, but an anecdote is simply a verbally expressed relationship between
digital, social, or media signals and sales. Altimeter indicates that this is likely
to be seen in large, often B2B, companies with high consideration and long
sales cycles, but visualizing a consumer example of this type of activity would
not be hard.
● A correlation analysis is simply an attempt to establish a relationship between
two different variables. This type of analysis is used to identify patterns in
behavior. It could be anything: comparing likes on Facebook to sales, the
relationship between engagement on Twitter and in-store traffic, or even more
advanced models that look at economic indicators and marketing activities.
● Multivariate testing is a method of testing a particular hypothesis using
complex, multivariable systems. It is most commonly used to test market
perceptions. Multivariate testing is a quickly growing area as it helps website
owners ensure that they are getting the most from the visitors arriving at their
site. Areas such as search engine optimization and pay-per-click advertising
bring visitors to a site and have been extensively used by many organizations.
Multivariate testing allows digital marketers to ensure that visitors are being
shown the right offers, content, and layout.

Bottom-Up Measurement Models
● Linking and tagging—Probably the most familiar method for
seasoned digital marketers, linking and tagging uses a series of
codes to track how a person comes to purchase your product.
● Integrated—Just as the name implies, integrated measurement
utilizes multiple techniques to gather information about how a
particular person makes a purchase.
● Direct commerce—This is probably the first “no duh” approach that
we have outlined, but the direct commerce route utilizes some sort
of selling functionality within the social network your brand is
utilizing.

Direct (Social) Commerce Approach
● Multichannel marketing—This represents the shift from two marketing
channels to five pillars (.com, brick-and-mortar, partners, employees, and
customers).
●
New media networks—Individual communities are forming across a
variety of social media channels.
●
Customers reached through search—Many of your customers might turn
to a search engine before they ever look to you for information.
●
A new content model—This should go without saying, but customerdriven content drives the highest conversion.
●
A new approach for retail—By understanding the effectiveness of each
partner or OEM, you know how to build the right retail mix by brand,
geography, and topic.
●
More effective media planning—Using data, we can become even
smarter about how we target different types of paid, owned, earned, or
shared media activities.
●
New demand—Creating new demand requires a focus on the broader
community and not the influencers in order to drive sales

Integrated Approach

The integrated measurement approach utilizes an application, typically installed on a
social property (most often Facebook) to track the user’s activity. This application can be
a way to serve up special content to users or direct them toward a place where they can
either receive a coupon or make a purchase directly.
●
●

●

Understanding consumer behavior—If you build an application that serves multiple
types of content, these apps can help you understand what consumers want to see
based on what they interact with the most.
Gathering consumer data—Most of these applications “force” users to enter a name
and an email address. The email address can be valuable when it is crossreferenced against an existing email database. However, the best applications
gather that information as well as other demographic characteristics that can be
very valuable for future testing.
Coupon redemptions—For many B2C companies, these applications can offer the
ability to serve up multiple types of coupons and track redemptions. While not a
sale, per se, the download of a coupon is a pretty good indicator of a sale.

Measuring Digital Marketing Effectiveness
Measuring effectiveness— means understanding the individual impact that a certain
channel (or channels) have against a previously identified KPI. For example, measuring
effectiveness might mean the number of people the company has reached with a banner
advertising campaign. In fact, there should be multiple effectiveness measures across the
channels you are using for your campaign.
Measuring efficiency— Almost every activation that we do digitally nowadays requires
some level of paid media promotion. – identifying the cost to reach or have the intended
audience engage with the brand – common efficiency measures include cost per
engagement (CPE), cost per view (CPV), and cost per acquisition (CPA). Achieving the
intended goal of a digital marketing campaign is great, but doing so efficiently is the path
to securing even more budget.
Measuring effect—Measuring impact includes taking all the various actions across the
campaign and evaluating their contribution to the business (ROI). return on media
investment (ROMI). The ROMI model takes into account several variables—reach,
website traffic, engagement, and search activity, among others—and tries to back into
what those activities will contribute to the bottom line. ROMI model also provides a
framework to introduce actual numbers at the end of the campaign to truly evaluate how
close we were to our benchmarks and what we need to change in the future.

LECTURE SLIDES FOR
CHAPTER 6
INTRODUCTION TO SURVIVAL ANALYSIS
• USES OF SURVIVAL ANALYSIS
• TYPES OF SURVIVAL ANALYSIS

Marketing Analytics | 2nd edition | Mike Grigsby

Uses of Survival Analysis
Predicting time until event:
• dealing with those that have not had the event:
• churn
• purchase
• response
• path
Predicting order of product purchases:
• competing risks
Insights into what independent variable pushes out / brings in time until
event:
• Lifetime value:
• descriptive
• predictive
Marketing Analytics | 2nd edition | Mike Grigsby

What is Survival Analysis?
censored
churned

customers

TIME UNTIL EVENT
censored
churned

now

% at risk

start

time
time

Marketing Analytics | 2nd edition | Mike Grigsby

18

What is Survival Analysis?
Invented in 1972 by Sir David Cox
• Paper called “Regression and Life Tables” in Royal Journal of Statistics.
• Came from bio-statistics.
• Studying time-until-event problems, especially where death was the event.
• Time-until-event studies posed statistical problems:
• Embedding time
• Dealing with censored observations
Breakthrough
• Creating proportional hazard technique
• Semi-parametric
• Partial likelihood
Survival analytics
• Life tables, life regression and PH regression
• Continuous time vs. discrete time

Marketing Analytics | 2nd edition | Mike Grigsby

19

Simpler Choices for Survival Analysis
The case for ordinary regression:
• What to do with time-until-event.
• What to do with censored observations.
• ie; those that did not have the event
The case for logistic regression:
• What to do with time-until-event.
• What to do with censored observations.
• ie; those that did not have the event

Marketing Analytics | 2nd edition | Mike Grigsby

20

Types of Survival Analysis
Life Table
• A plot of % at risk over time, essentially a plot of the survival curve
• Both discrete and continuous time
Life Reg
• A regression, fitting independent variables to the survival curve
• Censoring: right only, time varying independent variables: no
• Distribution choices: lognormal, gamma, weibul, negative logistic, exponential,
etc.
• Both discrete and continuous time
PH Reg
• A regression, fitting independent variables to the hazard rate
• Censoring: all, time varying independent variables: yes
• Distribution choices: irrelevant
• Both discrete and continuous time
Marketing Analytics | 2nd edition | Mike Grigsby

21

Types of Survival Analysis

• From Ordinary Regression

• To Logistic Regression
• To Cox Regression

Marketing Analytics | 2nd edition | Mike Grigsby

Differences in Dependent Variable
Ordinary regression (OLS)
• Continuous, can take on any value
Logistic regression
• Binary, 1= had the event, 0 = did not have the event
Cox regression (Proportional hazards) and life reg
• Life reg = Survival curve, time until event
• PH = number of events per time period

Marketing Analytics | 2nd edition | Mike Grigsby

23

Differences in Optimization Techniques
Ordinary regression
• Minimizes the sum of the squared residuals (OLS).
• This is a mathematical optimization problem.

Logistic regression
• Maximum likelihood
• Fully parametric:
• requires a functional form.
• and an assumption of the distribution of the dependent variable.
• What value of the coefficients maximizes the likelihood of observing the behavior
of the dependent variable?
• This is a grid-search problem, not a mathematical optimization problem.

Cox regression (Proportional hazards)
• Partial likelihood:
• uses only the exponetiated coefficients for maximization
• Semi-parametric:
• Does not require a distribution of the dependent variable, only functional form
Marketing Analytics | 2nd edition | Mike Grigsby

24

Differences in Coefficient Interpretation
e^B

-1

*100

ORDINARY REGRESSION

PRICE
-0.250

LOGISTIC REGRESSION

-0.250

0.780

-0.220

-22.048

SURVIVAL ANALYSIS (LIFE REG)

0.250

1.283

0.283

28.284
% change

# avg (5)

1.414
avg change

Ordinary regression: price coefficient -0.25
• As price increases by 1 unit, predicted dependent variable decreases by 0.25 units.

Logistic regression: price coefficient -0.25
• e**-0.25 = 0.779, as price increases by 1 unit, predicted probability of the dependent
variable decreases by 22.1%

Cox regression (life reg, NOT PH Reg): price coefficient 0.25 and
average = 5 days
• [(e**0.25)-1)] * 5 = 1.4201, as price increases by 1 unit, the predicted time until the
dependent variable increases by 1.4201 days
• Note that PH Reg would be negative instead of positive.
Marketing Analytics | 2nd edition | Mike Grigsby

25

Lifetime Value
Typically descriptive calculation
• No insights
• No strategy
Survival Analysis provides both
• Business case for changing price, etc.

Marketing Analytics | 2nd edition | Mike Grigsby

26

Regression Review Topics
LOGIT: binary
• Logistic distribution, PROC LOGISTIC
PROBIT: binary
• Normal distribution, PROC PROBIT
TOBIT: censored data
• Normal distribution, PROC QLIM
SURVIVAL MODELING: multiple distributions
• Time until event, PROC LIFEREG

PANEL REGRESSION: cross sectional time series data
• Individual effects: fixed and/or random

Marketing Analytics | 2nd edition | Mike Grigsby

27

Purchase answer to see full
attachment