Digital Marketing
Analytics
Semester 2 | MSc Digital Marketing
Klelia Prodromou
, Stuctured Data Unstructured Data
Week 1
Web scraping: collected by web scraping -collecting data from web- Data stored in databases Images, text, video,
and tables (20% )
sites social media platforms (i. e. such as products information, documents (80% )
listings, customer reviews, ratings, comments, posts) through scraper-
software and importing to local data base in xls, sql or other format
Cookie: Tiny text file in your web browser storing information about your
browsing experience (e.g., login information, user preferences, shopping
cart)
1. Individual Level Data: recording each visitors’ individual behaviour
with a time-stamp. i.e., what each customer does step by step.
2. Aggregate Data: total numbers. not focusing on individual level
behavior instead > total number of visitors, etc.
Marketing Metrics: Tools helping companies quantify, Email
compare, and interpret their own performance from
marketing activities-> measuring the impact of Social
marketing activities Media
Search Engine
1. Impression: content is displayed once on a web page (the
customer is exposed to an online content on a website) - does
not necessarily mean that the customer has actually seen the
content though
2. In-View Rate (%): % customer who actually SEE the online
content they are exposed to.
Zero Party Data: Can be confusing because same
as first-party data in many ways.
Difference: data that a customer intentionally and
3. CTR: Simple, fast and easy to measure. BUT not a good indicator proactively shares with a brand, include
of marketing effectiveness, mostly used if we want to measure preferences, purchase intentions, personal
awareness, engagement or if we are not able to measure context, and how the individual wants the brand to
conversions. recognize her
4. Bounce Rate: The (%) percentage of visitors to a particular website who
navigate away from the site after viewing only one page.
,Week 1
Attribution challenge: when users interact with multiple ads, it's hard to tell
which ad actuallydrove the final conversion -> Solution: A/B testing
Main Types of Attribution
1. Simplistic (i.e., last/first click)
2. Rule Based (Heuristic: Positon Based, Time Decay)
3. Data-Driven (Algorithmic-Statistical)
Rule Based Techniques
1. Uniformly Distributed (Linear) Model
• Claims all touchpoints along the journey
matters equally - (Over)simplifies the
attribution – when a fast/simple model
needed taking all touchpoints in
consideration
• Drawback: The model assumes every
interaction equally influences the purchase
decision, which is unrealistic. When you plan
a purchase, are all clicks equally important to you? Probably not. Calculate Attribution Value
2. Time Decay Model
• Gives more credit to interactions closer to the conversion,
for short lived deals or promotional offers based on the
Memory Decay assumption: recent experiences are more
influential.
• Drawback: It over-rewards late touchpoints, even though
earlier interactions could have been the real driver behind
the purchase. Aren’t there older, decisive moments — not just
the most recent ones — that actually convinced you along
your path to buy?
3. Pareto Distribution Model:
• Applies the 80/20 rule to the customer journey,
assigning 80% of the conversion value to the first
and last touchpoints, and spreading the remaining 20% Logistic Regression
across all other interactions.
• It assumes the first touchpoint is crucial for attracting
attention and the last for closing the deal.
• Drawback: It heavily favors first- and last-click channels,
based on arbitrary assumptions — even though a middle
Model Based Attribution: Probabilistic model
touchpoint could have been the true deciding factor in the purchase.
Model Based Attribution: Logistic Regression
Benefits: Easy to use and interpret, Insights on channel effects, Additional
explanatory variables can easily be added (i.e., time on site)
Drawback: Does not account for touch-point order
Converting Digital Landscape: RACE Model
Race marketing planning model is to provide a simple
structure for companies to develop an omnichannel
marketing communications plan which meets the challenges
of reaching and engaging online audiences to prompt
conversion to online or offline sales. Why use RACE:
• Reach: Make people aware of your brand (ads, social • Focuses on practical actions and
media posts, etc.). tactics.
• Act: Encourage interactions (website visits, social • Customer-centered — follows the
media follows, reading blog posts). customer journey.
• Convert: Turn interest into action (purchase or signup). • Integrates online and offline
• Engage:Build long-term relationships (emails, social marketing.
media, customer loyalty). • Encourages data-driven
improvements (using KPIs and
analytics).
• Supports omnichannel strategies.
, Variable Scale and Types
Week 1
Linear Regression in a Nutshell
• Y = dependent variable (DV) or outcome
if categorical, we need
variable (OV)
to make it nominal (0/1) • Xi = independent variable (IV) or predictor
binary variable variable (PV)
• α = constant = what happens when all (X)
We use… predictor variables are set to 0
• Linear regression = One predictor (PV) when we want to predict the • βi = coefficient of PVi
outcome (OV) using only one independent variable (e.g. predict sales • ε = residual/error term (part of Y not
based on advertising spend) explained by collection of Xs in model)
• Multiple linear regression = More than one predictor (PVs) when we want
to predict the outcome (OV) using several independent variables at the
same time (e.g. predict sales based on advertising spend, website traffic,
and number of social media posts)
since sign < 0.05 -> these variables has impact
on the independent variable (sales)
although Golden - biggest β -> biggest impact
(its absolute value)
• Facebook: One more (increase) click on our Facebook ads is likely to
increase our sales per person 17,683 € per year.
• Being a Regular member of Loyalty Program is likely to cause 62,924 € more sales in comparison to being a NON-MEMBER
• Being a Gold member of Loyalty Program is likely to cause 277,725 € more sales in comparison to being a NON-MEMBER
OUTCOME
VARIABLE (Y):
CONVERSION ? Concept of
YES or NO interaction: looking at
(Categorical) what happens when
two things are used
together — instead of
just looking at each
METHOD: thing separately
significant since ~ 0.00 LOGISTIC
REGRESSION
Organic Search with Exp(B) Odds Ratio 365,392 means:
Customers who clicked Organic Search results are X 365 times more likely to convert (buy) in
comparison to those who did not click Organic Search Results
Week 2
Multitouch Attribution - Challenges:
1. Varying Roles of touchpoints: introduction, assist, conversion - which, when and how
2. Time Window: How to determine how many days something is on the basket
3. What if customers did not click but saw the ad -> A/B testing could be helpful in order to see the effect on conversion
(view through conversations)
4. Multi device attribution: same gmail everywhere
5. GDPR Privacy and Potection (cookies)
Analytics
Semester 2 | MSc Digital Marketing
Klelia Prodromou
, Stuctured Data Unstructured Data
Week 1
Web scraping: collected by web scraping -collecting data from web- Data stored in databases Images, text, video,
and tables (20% )
sites social media platforms (i. e. such as products information, documents (80% )
listings, customer reviews, ratings, comments, posts) through scraper-
software and importing to local data base in xls, sql or other format
Cookie: Tiny text file in your web browser storing information about your
browsing experience (e.g., login information, user preferences, shopping
cart)
1. Individual Level Data: recording each visitors’ individual behaviour
with a time-stamp. i.e., what each customer does step by step.
2. Aggregate Data: total numbers. not focusing on individual level
behavior instead > total number of visitors, etc.
Marketing Metrics: Tools helping companies quantify, Email
compare, and interpret their own performance from
marketing activities-> measuring the impact of Social
marketing activities Media
Search Engine
1. Impression: content is displayed once on a web page (the
customer is exposed to an online content on a website) - does
not necessarily mean that the customer has actually seen the
content though
2. In-View Rate (%): % customer who actually SEE the online
content they are exposed to.
Zero Party Data: Can be confusing because same
as first-party data in many ways.
Difference: data that a customer intentionally and
3. CTR: Simple, fast and easy to measure. BUT not a good indicator proactively shares with a brand, include
of marketing effectiveness, mostly used if we want to measure preferences, purchase intentions, personal
awareness, engagement or if we are not able to measure context, and how the individual wants the brand to
conversions. recognize her
4. Bounce Rate: The (%) percentage of visitors to a particular website who
navigate away from the site after viewing only one page.
,Week 1
Attribution challenge: when users interact with multiple ads, it's hard to tell
which ad actuallydrove the final conversion -> Solution: A/B testing
Main Types of Attribution
1. Simplistic (i.e., last/first click)
2. Rule Based (Heuristic: Positon Based, Time Decay)
3. Data-Driven (Algorithmic-Statistical)
Rule Based Techniques
1. Uniformly Distributed (Linear) Model
• Claims all touchpoints along the journey
matters equally - (Over)simplifies the
attribution – when a fast/simple model
needed taking all touchpoints in
consideration
• Drawback: The model assumes every
interaction equally influences the purchase
decision, which is unrealistic. When you plan
a purchase, are all clicks equally important to you? Probably not. Calculate Attribution Value
2. Time Decay Model
• Gives more credit to interactions closer to the conversion,
for short lived deals or promotional offers based on the
Memory Decay assumption: recent experiences are more
influential.
• Drawback: It over-rewards late touchpoints, even though
earlier interactions could have been the real driver behind
the purchase. Aren’t there older, decisive moments — not just
the most recent ones — that actually convinced you along
your path to buy?
3. Pareto Distribution Model:
• Applies the 80/20 rule to the customer journey,
assigning 80% of the conversion value to the first
and last touchpoints, and spreading the remaining 20% Logistic Regression
across all other interactions.
• It assumes the first touchpoint is crucial for attracting
attention and the last for closing the deal.
• Drawback: It heavily favors first- and last-click channels,
based on arbitrary assumptions — even though a middle
Model Based Attribution: Probabilistic model
touchpoint could have been the true deciding factor in the purchase.
Model Based Attribution: Logistic Regression
Benefits: Easy to use and interpret, Insights on channel effects, Additional
explanatory variables can easily be added (i.e., time on site)
Drawback: Does not account for touch-point order
Converting Digital Landscape: RACE Model
Race marketing planning model is to provide a simple
structure for companies to develop an omnichannel
marketing communications plan which meets the challenges
of reaching and engaging online audiences to prompt
conversion to online or offline sales. Why use RACE:
• Reach: Make people aware of your brand (ads, social • Focuses on practical actions and
media posts, etc.). tactics.
• Act: Encourage interactions (website visits, social • Customer-centered — follows the
media follows, reading blog posts). customer journey.
• Convert: Turn interest into action (purchase or signup). • Integrates online and offline
• Engage:Build long-term relationships (emails, social marketing.
media, customer loyalty). • Encourages data-driven
improvements (using KPIs and
analytics).
• Supports omnichannel strategies.
, Variable Scale and Types
Week 1
Linear Regression in a Nutshell
• Y = dependent variable (DV) or outcome
if categorical, we need
variable (OV)
to make it nominal (0/1) • Xi = independent variable (IV) or predictor
binary variable variable (PV)
• α = constant = what happens when all (X)
We use… predictor variables are set to 0
• Linear regression = One predictor (PV) when we want to predict the • βi = coefficient of PVi
outcome (OV) using only one independent variable (e.g. predict sales • ε = residual/error term (part of Y not
based on advertising spend) explained by collection of Xs in model)
• Multiple linear regression = More than one predictor (PVs) when we want
to predict the outcome (OV) using several independent variables at the
same time (e.g. predict sales based on advertising spend, website traffic,
and number of social media posts)
since sign < 0.05 -> these variables has impact
on the independent variable (sales)
although Golden - biggest β -> biggest impact
(its absolute value)
• Facebook: One more (increase) click on our Facebook ads is likely to
increase our sales per person 17,683 € per year.
• Being a Regular member of Loyalty Program is likely to cause 62,924 € more sales in comparison to being a NON-MEMBER
• Being a Gold member of Loyalty Program is likely to cause 277,725 € more sales in comparison to being a NON-MEMBER
OUTCOME
VARIABLE (Y):
CONVERSION ? Concept of
YES or NO interaction: looking at
(Categorical) what happens when
two things are used
together — instead of
just looking at each
METHOD: thing separately
significant since ~ 0.00 LOGISTIC
REGRESSION
Organic Search with Exp(B) Odds Ratio 365,392 means:
Customers who clicked Organic Search results are X 365 times more likely to convert (buy) in
comparison to those who did not click Organic Search Results
Week 2
Multitouch Attribution - Challenges:
1. Varying Roles of touchpoints: introduction, assist, conversion - which, when and how
2. Time Window: How to determine how many days something is on the basket
3. What if customers did not click but saw the ad -> A/B testing could be helpful in order to see the effect on conversion
(view through conversations)
4. Multi device attribution: same gmail everywhere
5. GDPR Privacy and Potection (cookies)
