lecture 2 Flashcards
How to evaluate the performance of marketing
each campaign / channel evaluated on incremental profit that it produces relative to its cost
ROI = incremental profit due to advertising - cost of advertising / cost of advertising
Incremental sales
additional saels made due to advertising over and above what would have been sold without advertising
Incremental profit
additional profit made due to advertising over and above what would have been sold without advertising
Typically a function of incremental sales
Conversion journey (couple example)
you swipe online
get a coffee date
Get a movie date
Proposal
Marriage
yay
Marketing attribution
is the process for determining which marketing touchpoints led to a conversion
Attribution models
are the rule or set of rules that determines how conversion credit is assigned to different marketing touchpoints
Seeking answers to the following strategic question: How can an analyst attribute credit to multiple campaigns that (may have) contributed to generating a conversion?
approaches to the solution
Rule based attribution
Data-driven models
last touch attribution
attribution looks backward from each conversion to find the last ad that the user saw (or clicked on) prior to the conversion
Limitations of last touch attribution
1) other ads may have influenced customer and contributed to the sale
2) all sales are treated as incremental… would not have bought if one did not see ads
-less well recognized
3) unfairly favours channels that tend to show ads towards the end of path to purchase
such as seach, or ads served due to targeting
When does last touch “work”
accurate measure of ad response when:
all sales are incremental; that means no sales owuld happen without the advertising
Effect of ads on behaviour is short-lived and ad exposures are spaced out over time
- no “assists” from other advertising channels
First touch attribution
first touch (first click( attribution looks backward from each conversion to find the first ad that the user saw (or clicked on) prior to the conversion
Linear attribution
attribution looks backward from each conversion to find each ad that the user saw (or clicked on) prior to the conversion, assigning them equal weight
Position based attribution
attribution looks backward from each conversion to find each ad that the user saw (or clicked on) prior to the conversion, assigning higher weight to the first adn last
Need to decide on what the higher weight is
Commonly seen: 30% or 40% for both first and last
Aka U-shaped attribution
Time decay attribution
time decay: attribution looks backward from each conversion to find each ad that the user saw or clicked on prior to the conversion, assigning higher weight to more recent ads
limitations of rule based attribution
rule based solutions are inflexible and unable to distinguish between the true low and high impacct touch-points
Leads to an inaccurate division of credit
Ignores all customers that dont convert
Analyst/manager decides the attribution
Can pick something to show the results one needs
Data driven models (key difference)
dealing with mid-journey funnel in a non ad-hoc way
and mid funnel parts of customer journeys matter
Example models:
1) markov attribution models
2) shapely value attribution
Markov attribution model
a probabilistic model that represents buyer journeys as a graph
Graphs nodes are touchpoints
Graphs edges are observed transition between those states
first state webinar -> second state ad click
the the big example with websize campaign y and z and the webinar in the middle
Markov attribution models measuring the removal effecet
Simulate removing an advertising campaign from a graph
Measure change in success rate (i.e. change in KPI)
(the how to is beyond the scope of the class)
attribution value of a given campaign is then
A = total value of KPI * Campaign removal effects/Sum of all removal effects
shapely value attribution:
finds each ad campaigns marginal contribution, averaged over every sequence where the campaign could have been shown
Different campaigns work together in cooperation to generate value (i.e. some KPI)
Based on co-operative game theory (beyond the scope of this course)
Key idea is : measuring each campaigns marginal contribution
Determined by what is gained or lost by removing them from a coalition of campaigns
Order of ads doesnt matter (in the standard setup)
Two channel example (conversions shapely value)
facebook ads attracts 5 ppl
Google ads attracts 10 ppl
Conversion of facebook and google ads gets 45 ppl
Marginal Contributions
The Shapley value considers the marginal contribution of each player (ad channel) based on different possible orderings.
Facebook’s Marginal Contribution:
If Facebook joins first: Its contribution is 5 (because without Google, Facebook attracts 5 people).
If Google is already included, and Facebook joins second: The total combined contribution is 45, but Google already brought in 10. So, Facebook’s marginal contribution is 45−10=35
Google’s Marginal Contribution:
If Google joins first: Its contribution is 10 (because without Facebook, Google attracts 10 people).
If Facebook is already included, and Google joins second: The total combined contribution is 45, but Facebook already brought in 5. So, Google’s marginal contribution is
45−5=40
45−5=40.
Shapley Value Calculation:
For each channel (player), we calculate the average of its marginal contributions in different orders:
Facebook’s Shapley value:
ShapleyvalueforFacebook=0.5×(5+35)=20
Google’s Shapley value:
ShapleyvalueforGoogle
=0.5×(10+40)=25
Marginal Contributions:
Facebook’s marginal contribution: 20
Google’s marginal contribution: 25