lecture 3 Flashcards
What is media mix modelling
an analysis technique that allows marketers to measure the impact of their marketing and advertising campaigns to determine how various elements contribute to their goal which is often to driver conversions (sales) (sounds a lot like attribute modelling)
A more workable definition: aims to measure the correlation between total sales/revenue in each week/mmonth and advertising spending or impressions on that same week/month
Media mix modelling key differences to attribution modelling
usses aggregate data
Longer time horizon in the analysis (months vs weeks)
wider range of channels, both traditional and digital
Incorporate external influences such as seasonality promotions
Analysis done “less often”
The business problem: seeking answers to the following strategic questions
1) How does a firms marketing activities correlate to a KPI of interest
2) what is the optimal mix of marketing activities for a given KPI
data driven approach:
1) media mix modelling (MMM)
estimate how different marketing activities impact a KPI
How ? via linear regression
2) media mix optimization (MMO)
adjust budget allocation across marketing activities to optimize a KPI of interest
Caution: media mix vs marketing mix
media mix modelling seeks to understand the role each media type plays in driving the overall campaign performance
Marketing mix modelling: takes a more holistic view
it doesnt just consider media channels but also factors in other marketing activities: pricing strategies, product distribution and even macroeconomic indicators
About viewing the larger picture and determining how these varying elements interplay to impact the overall marketing performance
The MMM problem:
How does a firms marketing activities impact revenue?
split into two sub questions
a) how does each marketing channel contribute to revenue
b) what proportion of revenue can be attributed to marketing activities
state space models
goal: estimate an impact that is true on average across all regions and time periods
Estimate deviations from this average that are region and time specific
–> we need data on: revenue and marketing activities
too vary across regions and over time
Comparing alternative MMM models
Similar linear regression
mixed models
State space models
similar linear regression: same coefficient generated for each panel for every week
Mixed models: coefficient now vary by panel (fixed + random estimates) but are similar across weeks
state space model: Coefficient vary for each panel by week giving time varying dimensions
mixed models
goal: estimate an impact that is true on average across all regions
Estimate deviations from this average that are region specific
We need data at the region level
Temporal effects of advertising
curernt effects are immediate spikes and immediate fall
Carryover effects or long duration spike and then slowly decline over time
Carryover effects of short duration and multiple spikes that nt instantly but quickly decline
Persistant effect is a spike and then retained higher values in the future
there are non linear responses to advertising
concave response, quick increase and keeps increasing but slower maybe even begin to decrease
S shaped …..
Media mix optimization problem:
how to allocate a marketing budget over multiple channels
Assume there is one KPI we want to maximize (e.g. revenue)
we can predict revenue based on a media mix model
We have a fixed budget to allocate
simplification for today: we assume we want to maximize sales in this time period only by allocating spending across channels in this time period only. Remark: can be extended to set optimal strategy for multiple periods
Think of media mix optimization as a two step process
simulate different budget allocations and their impact on a KPI
Choose the best allocation
Assumes the media mix model that predicts revenue already exists
This is a constrained optimization problem:
How to di this is beyond the scope of the class
but the idea is maximize revenue
subject to facebook ad spend +tv adspend < budget
