SlideShare uma empresa Scribd logo
1 de 21
© Absolutdata 2014 Proprietary and Confidential
Chicago New York London Dubai New Delhi Bangalore SingaporeSan Francisco
www.absolutdata.com
April 28, 2014
Our exploration into
innovations in Marketing Mix Modelling
Meet-up Data Science Oxford
© Absolutdata 2014 Proprietary and Confidential 2
Agenda
NLMixed & MCMC to automate the
fitting of s-curves and more
The statistical Challenges and the
curse of the s-curve
What is Marketing Mix Modeling?
VARx Modelling
Discussion



© Absolutdata 2014 Proprietary and Confidential 3
Strategic vs. tactical decisions
Spend
Brand Equity
Engagement
Web
Purchase
Profile
/segment
Preferences
Propensity Models
Strategic
How to spread the budget?
What campaigns?
Product design
Pricing strategy
???
Tactical
Who?
What?
When?
???
Decision support
modelling and tools
© Absolutdata 2014 Proprietary and Confidential 4
Measurement of effectiveness & efficiency by marketing driver –
Answering key business questions
 Identify brand/s’ volume growth/decline
drivers
 Explain what is driving year over year change
in sales
 Determine relative importance & estimates
elasticity of different drivers
 Identify the competition and assesses impact
 Determine the key competitive levers that
impact/interact with own brand
 Build a robust framework for investment
decisions for short term - brands that should
be supported
 Determine an optimal realistic investment
plan for the future based on simulated
scenarios
© Absolutdata 2014 Proprietary and Confidential 5
Magazine
Online
Print
Radio
TV
Overall
Sales
Affiliate Clicks
Paid Search Clicks
Display Clicks
Decision
Support
Analysis
We would like to measure the direct and indirect impact of
marketing investment at a granularity relevant to planning
© Absolutdata 2014 Proprietary and Confidential 6
Marketing mix modeling & optimization @ Absolutdata
focus on shift towards true attribution & predictive spend mix
Data
Collection
From
Disparate
Data Sources
Data Review –
Confirming
directional
movements of
prominent variables
Review past
performance to
determine
CONTRIBUTION of
each marketing
mix element
Traditional
Regression Based
Modeling
Techniques such as
OLS
DIY
Simulator
What-If
Scenario
Planning
Optimized
Media
Calendar
Objective
Function &
Constraints
Attribution &
ROI
Measurement
– Multi
channel
influence and
interactions
Advanced
Modeling
Techniques
such as
Auto
Regressive,
VARX, SEM
Added
model
granularity
at
customer
segment
and
regional
level
© Absolutdata 2014 Proprietary and Confidential 7
In most cases the data is highly noisy
The key challenge it to find a statistically sound solution that can
help the business
The exogenous (independent) time series are highly inter-
correlated
The business requires an impact estimate for the actions
they can take in order to evaluate strategies (what if
analysis)
The business would prefer not to hear about dimension
reduction
S-curves
© Absolutdata 2014 Proprietary and Confidential 8
Business understanding is introduces through transformation
Ad-stock
Saturation
Effects
 Ad stock captures exponential decay
effect of GRPs - TV
 This depends on the temporal effect
of GRPs, can be done for Print, Radio
as well
 Research suggests that immediate
response generated by advertising
follows an S-curve
 It introduces three additional
parameters: saturation rate, point of
inflexion and ‘half-life’ parameter (for
carryover effect) as unknowns in the
model
 Together this captures the diminishing
return of advertising
Ad-stock impact
which depends
on the temporal
effect of GRPs
Carry-Over Effect
SampleOutput
© Absolutdata 2014 Proprietary and Confidential 9
Various research indicated that immediate response generated by advertising is followed
exponential decay
Ad-stock carry over is estimated and reported in terms of half-life
Current Effect
Half-Life K= Carry Over
Carry-Over Effect
© Absolutdata 2014 Proprietary and Confidential 10
0
200
400
600
800
1000
1200
1400
1600
1800
0 50 100 150 200 250 300 350 400 450
RevenueImpact
(ExcludingCarry-overeffect)
$ Spent per Week
Inflexion
Point
Saturation Level
Adstock Equation:
Where, X is actual spending, K is decay constant and determined by expression exp(ln(0.5)/t1/2); V as saturation
parameter and Xd as diminishing return point (point of inflexion)
S-Curve Impact Carry-over ImpactAd-stock Impact
At= 1/(1+exp(-V*(X-Xd ))) + K*At-1
Half-Life
Research suggests that immediate response generated by advertising can be modeled using an S-curve followed
by exponential decay of effect which helps capture the diminishing return of advertising
S-Curves transformation reflect the belief that Marketing spend reduces in its
effectiveness at a certain point and it impact decays over time
© Absolutdata 2014 Proprietary and Confidential 11
Standardize the variables
De-trend and account for seasonality effect
Fit an s-curve for each variable optimising it for regression to the residuals to the
Revenue trend
What is more appropriate from a businesswise perspective:
Univariate or Multivariate?
Standardize
the variables
Select a subset
Apply
s-curves
Fit a regression
model
explaining the
Revenue
Evaluate
quality of fit
Multivariate – optimize s-curves to work together in a particular setting
Manually optimize
Univariate –
fit a s-curve
for each
variable on
its own
Automate
© Absolutdata 2014 Proprietary and Confidential 12
Instead of calculating the s-curve from t=0, I concentrate on the last n lags:
I explored two sas procedures for fitting the s-curves
Fit aggregation model where S is the dependent and the residual to the Revenue trend is the
independent (no intercept for now – might need to reconsider that)
NLMIXED MCMC
© Absolutdata 2014 Proprietary and Confidential 13
My code – if you must
%Macro HazSThree(SVar);
S_Mue=1/(1+exp(-sV*(&SVar._lag&Nlags.-sXd))) ;
%do l=%sysevalf(&Nlags.-1) %to 0 %by -1;
S_Mue=1/(1+exp(-sV*(&SVar._lag&l.-sXd))) + sK*S_Mue;
%end;
%mend
proc nlmixed data=HAZ.Detrend MAXITER=4000 maxfunc=4000;
ods select ParameterEstimates;
parms b1=1 se=1 sV=1 sXd=0 SK=0.25;
bounds se>0;
bounds sV>0;
bounds 0<SK<1;
bounds b1>0;
%HazSThree(&HazVar.);
Mue= b1*S_Mue;
model Detrended ~ normal(mue, se);
proc mcmc data=HAZ.Detrend &MCMCOptions.;
ods select PostSummaries ;
parms b1 se
sV sXd SK
;
prior b1 ~ normal(mean = 1, var = 0.5);
prior se ~ igamma(shape = 3/10, scale = 10/3);
prior SK ~ uniform(0,1);
prior sXd ~ uniform(-2,2);
prior sV ~ uniform(0,10);
%HazSThree(&HazVar.);
Mue=b1*S_Mue;
model Detrended ~ n(mue, sd = se);
For illustrative purposes the code shown here fits only one s-cure. The solution actually fits a baseline and all the
curves
© Absolutdata 2014 Proprietary and Confidential 14
Vector Auto Regression (VAR)
time
Steady state
progression
Target
Sales/Signups
time
Co dependency
associationIndigenous
Web search
time
Marketing impact
& decay
Exogenous
campaigns
© Absolutdata 2014 Proprietary and Confidential 15
Phase I: Top down marketing mix modeling
Phase III: Reconcile
MMM & Cookie
Attribution
Phase IV:
Reporting, Simulatio
n and Optimization
Phase I: Marketing Mix
Modeling
Phase II: Cookie-
Based Attribution
Algorithm
Search
Clicks
Affiliates
Display
Impressions
TV Impacts
AffiliatesSecondary Relationships
Search
Signups
Email
Signups
Print
Signups
Signups
from
Other
Factors
Previous
Day’s
Baseline
Signups
+TV GI
Signups
Display
Signups+ + + + +
Daily
Signups=
© Absolutdata 2014 Proprietary and Confidential 16
Secondary attribution provides A refined view of the system
Paid
Search Clicks
Non
paid search
Cable
Total Impact
11.4%
9.0%
2.5%
3.8%
-1.0%
2.2%
-0.1%
2.6%-3.8%
-2.2%
Actual TV Attribution taking into
account indirect contribution of Search
Final Attribution 7.5% 5.7% 11.1%
SampleOutput
-0.1%
© Absolutdata 2014 Proprietary and Confidential 17
Applying
VARx to a BIG
number of
SKUs
How to
choose priors
We have encountered interesting challenges
Challenges
Modelling short
term and long
term effect
Cookie data
allows to not only
do attribution but
identify key
sequences
© Absolutdata 2014 Proprietary and Confidential 18
Incomplete sales (Target) data
We do not know what is sold when by the distributers
We sell to their stock
© Absolutdata 2014 Proprietary and Confidential 19
http://www.linkedin.com/groupItem?view=&gid=130238&item=233249172&type=member&commentID=5810144471664320512&trk=hb_ntf_COMMENTED_O
N_GROUP_DISCUSSION_YOU_FOLLOWED#commentID_5810144471664320512
Friends do not let friends to use Excel for statistical analysis
© Absolutdata 2014 Proprietary and Confidential 20
Absolutdata provide analytics based solutions to address business critical issues
40% increase in profits through
Conjoint based Pricing
Optimization – A top SaaS
company
$50MM increase in revenue
by Market Mix Modeling
across 4 geographies
– A leading CPG Company
15% revenue growth through
Multi Channel Attribution
– A large ecommerce
company
$23MM increase in Customer
Loyalty and CRM marketing
revenue
– A major Hotel chain
$9MM incremental revenue
as a result of focused
promotional campaigns
created
– A major Online Retail
Discounter
Contribution of $78MM over the
last few years to their margins
– A major Retailer
We are decision scientists who help decision makers take better and informed decisions
Eli Y. Kling
Director - Analytics
Phone: +44 (0)7940094976
Email: Eli.Kling@absolutdata.com
LinkedIn: Uk.linkedin.com/in/elikling
Follow us on:

Mais conteúdo relacionado

Mais procurados

IBM - Full year Go-to-market plan template
IBM - Full year Go-to-market plan templateIBM - Full year Go-to-market plan template
IBM - Full year Go-to-market plan templateArrow ECS UK
 
Sales Enablement Plan Playbook
Sales Enablement Plan PlaybookSales Enablement Plan Playbook
Sales Enablement Plan PlaybookDemand Metric
 
Go-To-Market Framework
Go-To-Market FrameworkGo-To-Market Framework
Go-To-Market FrameworkMark Officer
 
Digital Marketing Trends 2023- Shereen Badr
Digital Marketing Trends 2023- Shereen BadrDigital Marketing Trends 2023- Shereen Badr
Digital Marketing Trends 2023- Shereen BadrShereen Badr
 
How to Build a Winning Martech Stack
How to Build a Winning Martech StackHow to Build a Winning Martech Stack
How to Build a Winning Martech StackMarsden Marketing
 
Go-to-market Framework
Go-to-market FrameworkGo-to-market Framework
Go-to-market FrameworkDemand Metric
 
Webinar DV 360 Self Serve _ 25 March 2022.pptx
Webinar DV 360 Self Serve _ 25 March 2022.pptxWebinar DV 360 Self Serve _ 25 March 2022.pptx
Webinar DV 360 Self Serve _ 25 March 2022.pptxTatvic Analytics
 
Digital Technologies and Online Marketing Agency MSG Marketing
Digital Technologies and Online Marketing Agency MSG MarketingDigital Technologies and Online Marketing Agency MSG Marketing
Digital Technologies and Online Marketing Agency MSG MarketingMSG Marketing
 
First Rule of Marketing Analytics: Forget the Customer - Digital Summit Phoenix
First Rule of Marketing Analytics: Forget the Customer - Digital Summit PhoenixFirst Rule of Marketing Analytics: Forget the Customer - Digital Summit Phoenix
First Rule of Marketing Analytics: Forget the Customer - Digital Summit Phoenixmbedner
 
Introduction to Google Analytics
Introduction to Google AnalyticsIntroduction to Google Analytics
Introduction to Google AnalyticsCemal Buyukgokcesu
 
Marketing Automation
Marketing AutomationMarketing Automation
Marketing AutomationAndrew Wilson
 
Digital Marketing Plan Template Smart Insights
Digital Marketing Plan Template Smart InsightsDigital Marketing Plan Template Smart Insights
Digital Marketing Plan Template Smart InsightsJasonmiller484
 
Digital Strategy Framework
Digital Strategy FrameworkDigital Strategy Framework
Digital Strategy FrameworkMonique Terrell
 
Marketing Analytics ppt
Marketing Analytics pptMarketing Analytics ppt
Marketing Analytics pptDarshilJani4
 
Digital Marketing Trends for 2022
Digital Marketing Trends for 2022Digital Marketing Trends for 2022
Digital Marketing Trends for 2022Vbout.com
 
Building a Foundational Tech Stack to Support Your First-Party Data Strategy
Building a Foundational Tech Stack to Support Your First-Party Data StrategyBuilding a Foundational Tech Stack to Support Your First-Party Data Strategy
Building a Foundational Tech Stack to Support Your First-Party Data StrategyTinuiti
 

Mais procurados (20)

Marketing data analytics
Marketing data analyticsMarketing data analytics
Marketing data analytics
 
Big Data and Advanced Analytics
Big Data and Advanced AnalyticsBig Data and Advanced Analytics
Big Data and Advanced Analytics
 
IBM - Full year Go-to-market plan template
IBM - Full year Go-to-market plan templateIBM - Full year Go-to-market plan template
IBM - Full year Go-to-market plan template
 
Sales Enablement Plan Playbook
Sales Enablement Plan PlaybookSales Enablement Plan Playbook
Sales Enablement Plan Playbook
 
Go-To-Market Framework
Go-To-Market FrameworkGo-To-Market Framework
Go-To-Market Framework
 
B2B Marketing Strategy
B2B Marketing StrategyB2B Marketing Strategy
B2B Marketing Strategy
 
Digital Marketing Trends 2023- Shereen Badr
Digital Marketing Trends 2023- Shereen BadrDigital Marketing Trends 2023- Shereen Badr
Digital Marketing Trends 2023- Shereen Badr
 
How to Build a Winning Martech Stack
How to Build a Winning Martech StackHow to Build a Winning Martech Stack
How to Build a Winning Martech Stack
 
Go-to-market Framework
Go-to-market FrameworkGo-to-market Framework
Go-to-market Framework
 
Webinar DV 360 Self Serve _ 25 March 2022.pptx
Webinar DV 360 Self Serve _ 25 March 2022.pptxWebinar DV 360 Self Serve _ 25 March 2022.pptx
Webinar DV 360 Self Serve _ 25 March 2022.pptx
 
Digital Technologies and Online Marketing Agency MSG Marketing
Digital Technologies and Online Marketing Agency MSG MarketingDigital Technologies and Online Marketing Agency MSG Marketing
Digital Technologies and Online Marketing Agency MSG Marketing
 
First Rule of Marketing Analytics: Forget the Customer - Digital Summit Phoenix
First Rule of Marketing Analytics: Forget the Customer - Digital Summit PhoenixFirst Rule of Marketing Analytics: Forget the Customer - Digital Summit Phoenix
First Rule of Marketing Analytics: Forget the Customer - Digital Summit Phoenix
 
Introduction to Google Analytics
Introduction to Google AnalyticsIntroduction to Google Analytics
Introduction to Google Analytics
 
Data Driven Marketing
Data Driven MarketingData Driven Marketing
Data Driven Marketing
 
Marketing Automation
Marketing AutomationMarketing Automation
Marketing Automation
 
Digital Marketing Plan Template Smart Insights
Digital Marketing Plan Template Smart InsightsDigital Marketing Plan Template Smart Insights
Digital Marketing Plan Template Smart Insights
 
Digital Strategy Framework
Digital Strategy FrameworkDigital Strategy Framework
Digital Strategy Framework
 
Marketing Analytics ppt
Marketing Analytics pptMarketing Analytics ppt
Marketing Analytics ppt
 
Digital Marketing Trends for 2022
Digital Marketing Trends for 2022Digital Marketing Trends for 2022
Digital Marketing Trends for 2022
 
Building a Foundational Tech Stack to Support Your First-Party Data Strategy
Building a Foundational Tech Stack to Support Your First-Party Data StrategyBuilding a Foundational Tech Stack to Support Your First-Party Data Strategy
Building a Foundational Tech Stack to Support Your First-Party Data Strategy
 

Semelhante a Innovations in Market Mix Modelling

Optimization as a Golden Layer - Chris Diener, SVP Analytics, Absolutdata
Optimization as a Golden Layer - Chris Diener, SVP Analytics, AbsolutdataOptimization as a Golden Layer - Chris Diener, SVP Analytics, Absolutdata
Optimization as a Golden Layer - Chris Diener, SVP Analytics, AbsolutdataAbsolutdata Analytics
 
Multi Channel Attribution - Driving Marketing Spend Planning In The Big Data Age
Multi Channel Attribution - Driving Marketing Spend Planning In The Big Data AgeMulti Channel Attribution - Driving Marketing Spend Planning In The Big Data Age
Multi Channel Attribution - Driving Marketing Spend Planning In The Big Data AgeAbsolutdata Analytics
 
Channel analytics 20150318
Channel analytics 20150318Channel analytics 20150318
Channel analytics 20150318Rob Ford
 
(MRSI - 3/3) Strategic country clusters using ensemble clustering methodolgie...
(MRSI - 3/3) Strategic country clusters using ensemble clustering methodolgie...(MRSI - 3/3) Strategic country clusters using ensemble clustering methodolgie...
(MRSI - 3/3) Strategic country clusters using ensemble clustering methodolgie...Absolutdata Analytics
 
Shows approach which expands the breadth of what marketing-mix models c
Shows approach which expands the breadth of what marketing-mix models cShows approach which expands the breadth of what marketing-mix models c
Shows approach which expands the breadth of what marketing-mix models cMichael Wolfe
 
Attribution Playbook Webinar 3
Attribution Playbook Webinar 3Attribution Playbook Webinar 3
Attribution Playbook Webinar 3Adometry by Google
 
Model N 2013 annual Global Price Management Survey - Life Sciences
Model N 2013 annual Global Price Management Survey  - Life Sciences Model N 2013 annual Global Price Management Survey  - Life Sciences
Model N 2013 annual Global Price Management Survey - Life Sciences Alex Rumble
 
Closing the Programmatic Loop
Closing the Programmatic LoopClosing the Programmatic Loop
Closing the Programmatic LoopMediaPost
 
FInancial Modeling and Valuations for Startups: Telling your Story with Numbers
FInancial Modeling and Valuations for Startups: Telling your Story with NumbersFInancial Modeling and Valuations for Startups: Telling your Story with Numbers
FInancial Modeling and Valuations for Startups: Telling your Story with NumbersForesight Valuation Group
 
Building a Global Partner Program - From Strategy to Execution
Building a Global Partner Program - From Strategy to ExecutionBuilding a Global Partner Program - From Strategy to Execution
Building a Global Partner Program - From Strategy to Executiondreamforce2006
 
How to Execute Your CX Vision
How to Execute Your CX VisionHow to Execute Your CX Vision
How to Execute Your CX VisionQualtrics
 
How To Execute Your Vision
How To Execute Your VisionHow To Execute Your Vision
How To Execute Your VisionQualtrics
 
CIMA Dublin MIAGEN on FP&A Best Practices June '15
CIMA Dublin MIAGEN on FP&A Best Practices June '15CIMA Dublin MIAGEN on FP&A Best Practices June '15
CIMA Dublin MIAGEN on FP&A Best Practices June '15MIAGEN
 
M B F004 Smyth 091707
M B F004  Smyth 091707M B F004  Smyth 091707
M B F004 Smyth 091707Dreamforce07
 
Tns _L'evoluzione del Brand Tracking_Webinar _Settembre 2015
Tns _L'evoluzione del Brand Tracking_Webinar _Settembre 2015Tns _L'evoluzione del Brand Tracking_Webinar _Settembre 2015
Tns _L'evoluzione del Brand Tracking_Webinar _Settembre 2015Gabriella Bergaglio
 
Cross-Channel Digital Attribution
Cross-Channel Digital AttributionCross-Channel Digital Attribution
Cross-Channel Digital AttributionMediaPost
 
Attribution & Our Approach
Attribution & Our ApproachAttribution & Our Approach
Attribution & Our Approachdmg events Asia
 
Adetem b2 b marketing mix with siriusdecisions 2015 the pulse european demand...
Adetem b2 b marketing mix with siriusdecisions 2015 the pulse european demand...Adetem b2 b marketing mix with siriusdecisions 2015 the pulse european demand...
Adetem b2 b marketing mix with siriusdecisions 2015 the pulse european demand...Hervé Gonay
 

Semelhante a Innovations in Market Mix Modelling (20)

Optimization as a Golden Layer - Chris Diener, SVP Analytics, Absolutdata
Optimization as a Golden Layer - Chris Diener, SVP Analytics, AbsolutdataOptimization as a Golden Layer - Chris Diener, SVP Analytics, Absolutdata
Optimization as a Golden Layer - Chris Diener, SVP Analytics, Absolutdata
 
Multi Channel Attribution - Driving Marketing Spend Planning In The Big Data Age
Multi Channel Attribution - Driving Marketing Spend Planning In The Big Data AgeMulti Channel Attribution - Driving Marketing Spend Planning In The Big Data Age
Multi Channel Attribution - Driving Marketing Spend Planning In The Big Data Age
 
Channel analytics 20150318
Channel analytics 20150318Channel analytics 20150318
Channel analytics 20150318
 
(MRSI - 3/3) Strategic country clusters using ensemble clustering methodolgie...
(MRSI - 3/3) Strategic country clusters using ensemble clustering methodolgie...(MRSI - 3/3) Strategic country clusters using ensemble clustering methodolgie...
(MRSI - 3/3) Strategic country clusters using ensemble clustering methodolgie...
 
Shows approach which expands the breadth of what marketing-mix models c
Shows approach which expands the breadth of what marketing-mix models cShows approach which expands the breadth of what marketing-mix models c
Shows approach which expands the breadth of what marketing-mix models c
 
Attribution Playbook Webinar 3
Attribution Playbook Webinar 3Attribution Playbook Webinar 3
Attribution Playbook Webinar 3
 
Model N 2013 annual Global Price Management Survey - Life Sciences
Model N 2013 annual Global Price Management Survey  - Life Sciences Model N 2013 annual Global Price Management Survey  - Life Sciences
Model N 2013 annual Global Price Management Survey - Life Sciences
 
Closing the Programmatic Loop
Closing the Programmatic LoopClosing the Programmatic Loop
Closing the Programmatic Loop
 
SAP Marketing Runs SAP
SAP Marketing Runs SAP SAP Marketing Runs SAP
SAP Marketing Runs SAP
 
FInancial Modeling and Valuations for Startups: Telling your Story with Numbers
FInancial Modeling and Valuations for Startups: Telling your Story with NumbersFInancial Modeling and Valuations for Startups: Telling your Story with Numbers
FInancial Modeling and Valuations for Startups: Telling your Story with Numbers
 
Building a Global Partner Program - From Strategy to Execution
Building a Global Partner Program - From Strategy to ExecutionBuilding a Global Partner Program - From Strategy to Execution
Building a Global Partner Program - From Strategy to Execution
 
How to Execute Your CX Vision
How to Execute Your CX VisionHow to Execute Your CX Vision
How to Execute Your CX Vision
 
How To Execute Your Vision
How To Execute Your VisionHow To Execute Your Vision
How To Execute Your Vision
 
CIMA Dublin MIAGEN on FP&A Best Practices June '15
CIMA Dublin MIAGEN on FP&A Best Practices June '15CIMA Dublin MIAGEN on FP&A Best Practices June '15
CIMA Dublin MIAGEN on FP&A Best Practices June '15
 
M B F004 Smyth 091707
M B F004  Smyth 091707M B F004  Smyth 091707
M B F004 Smyth 091707
 
Multi-Channel Attribution - Where do Leads Come From?
Multi-Channel Attribution - Where do Leads Come From?Multi-Channel Attribution - Where do Leads Come From?
Multi-Channel Attribution - Where do Leads Come From?
 
Tns _L'evoluzione del Brand Tracking_Webinar _Settembre 2015
Tns _L'evoluzione del Brand Tracking_Webinar _Settembre 2015Tns _L'evoluzione del Brand Tracking_Webinar _Settembre 2015
Tns _L'evoluzione del Brand Tracking_Webinar _Settembre 2015
 
Cross-Channel Digital Attribution
Cross-Channel Digital AttributionCross-Channel Digital Attribution
Cross-Channel Digital Attribution
 
Attribution & Our Approach
Attribution & Our ApproachAttribution & Our Approach
Attribution & Our Approach
 
Adetem b2 b marketing mix with siriusdecisions 2015 the pulse european demand...
Adetem b2 b marketing mix with siriusdecisions 2015 the pulse european demand...Adetem b2 b marketing mix with siriusdecisions 2015 the pulse european demand...
Adetem b2 b marketing mix with siriusdecisions 2015 the pulse european demand...
 

Mais de Absolutdata Analytics

Seven questions you must answer before starting concept research
Seven questions you must answer before starting concept researchSeven questions you must answer before starting concept research
Seven questions you must answer before starting concept researchAbsolutdata Analytics
 
When Less Is More: Focusing on the Persuadable Guest
When Less Is More: Focusing on the Persuadable GuestWhen Less Is More: Focusing on the Persuadable Guest
When Less Is More: Focusing on the Persuadable GuestAbsolutdata Analytics
 
'No menu - Give us the Chef's Special'
'No menu - Give us the Chef's Special' 'No menu - Give us the Chef's Special'
'No menu - Give us the Chef's Special' Absolutdata Analytics
 
Decision Making Through Conjoint Analysis
Decision Making Through Conjoint AnalysisDecision Making Through Conjoint Analysis
Decision Making Through Conjoint AnalysisAbsolutdata Analytics
 
Mobile Behavioral Analytics - A new window on the soul
Mobile Behavioral Analytics - A new window on the soulMobile Behavioral Analytics - A new window on the soul
Mobile Behavioral Analytics - A new window on the soulAbsolutdata Analytics
 
Mobile Behavioural Analytics - A new window on the soul
Mobile Behavioural Analytics - A new window on the soulMobile Behavioural Analytics - A new window on the soul
Mobile Behavioural Analytics - A new window on the soulAbsolutdata Analytics
 
Application of MBC through Storytelling in CPG Industry
Application of MBC through Storytelling in CPG IndustryApplication of MBC through Storytelling in CPG Industry
Application of MBC through Storytelling in CPG IndustryAbsolutdata Analytics
 
3 Steps to Influence Purchase Decision Through Social Media
3 Steps to Influence Purchase Decision Through Social Media3 Steps to Influence Purchase Decision Through Social Media
3 Steps to Influence Purchase Decision Through Social MediaAbsolutdata Analytics
 
(MRSI- 2/3) Re-engineering Segmentation methodologies for an Enriching Custom...
(MRSI- 2/3) Re-engineering Segmentation methodologies for an Enriching Custom...(MRSI- 2/3) Re-engineering Segmentation methodologies for an Enriching Custom...
(MRSI- 2/3) Re-engineering Segmentation methodologies for an Enriching Custom...Absolutdata Analytics
 
Adapting conjoint to the mobile phenomenon
Adapting conjoint to the mobile phenomenonAdapting conjoint to the mobile phenomenon
Adapting conjoint to the mobile phenomenonAbsolutdata Analytics
 
Camera ready sentiment analysis : quantification of real time brand advocacy ...
Camera ready sentiment analysis : quantification of real time brand advocacy ...Camera ready sentiment analysis : quantification of real time brand advocacy ...
Camera ready sentiment analysis : quantification of real time brand advocacy ...Absolutdata Analytics
 

Mais de Absolutdata Analytics (16)

Seven questions you must answer before starting concept research
Seven questions you must answer before starting concept researchSeven questions you must answer before starting concept research
Seven questions you must answer before starting concept research
 
Navik Concept Test Overview Webinar
Navik Concept Test Overview WebinarNavik Concept Test Overview Webinar
Navik Concept Test Overview Webinar
 
Skin care report
Skin care reportSkin care report
Skin care report
 
Shared Data - Assured Success
Shared Data - Assured SuccessShared Data - Assured Success
Shared Data - Assured Success
 
When Less Is More: Focusing on the Persuadable Guest
When Less Is More: Focusing on the Persuadable GuestWhen Less Is More: Focusing on the Persuadable Guest
When Less Is More: Focusing on the Persuadable Guest
 
'No menu - Give us the Chef's Special'
'No menu - Give us the Chef's Special' 'No menu - Give us the Chef's Special'
'No menu - Give us the Chef's Special'
 
Decision Making Through Conjoint Analysis
Decision Making Through Conjoint AnalysisDecision Making Through Conjoint Analysis
Decision Making Through Conjoint Analysis
 
Mobile Behavioral Analytics - A new window on the soul
Mobile Behavioral Analytics - A new window on the soulMobile Behavioral Analytics - A new window on the soul
Mobile Behavioral Analytics - A new window on the soul
 
Mobile Behavioural Analytics - A new window on the soul
Mobile Behavioural Analytics - A new window on the soulMobile Behavioural Analytics - A new window on the soul
Mobile Behavioural Analytics - A new window on the soul
 
Application of MBC through Storytelling in CPG Industry
Application of MBC through Storytelling in CPG IndustryApplication of MBC through Storytelling in CPG Industry
Application of MBC through Storytelling in CPG Industry
 
3 Steps to Influence Purchase Decision Through Social Media
3 Steps to Influence Purchase Decision Through Social Media3 Steps to Influence Purchase Decision Through Social Media
3 Steps to Influence Purchase Decision Through Social Media
 
(MRSI- 2/3) Re-engineering Segmentation methodologies for an Enriching Custom...
(MRSI- 2/3) Re-engineering Segmentation methodologies for an Enriching Custom...(MRSI- 2/3) Re-engineering Segmentation methodologies for an Enriching Custom...
(MRSI- 2/3) Re-engineering Segmentation methodologies for an Enriching Custom...
 
Adapting conjoint to the mobile phenomenon
Adapting conjoint to the mobile phenomenonAdapting conjoint to the mobile phenomenon
Adapting conjoint to the mobile phenomenon
 
Camera ready sentiment analysis : quantification of real time brand advocacy ...
Camera ready sentiment analysis : quantification of real time brand advocacy ...Camera ready sentiment analysis : quantification of real time brand advocacy ...
Camera ready sentiment analysis : quantification of real time brand advocacy ...
 
Customer analytics fast facts v3
Customer analytics fast facts v3Customer analytics fast facts v3
Customer analytics fast facts v3
 
From 'I think' to 'I know'
From 'I think' to 'I know'From 'I think' to 'I know'
From 'I think' to 'I know'
 

Último

BDSM⚡Call Girls in Sector 144 Noida Escorts >༒8448380779 Escort Service
BDSM⚡Call Girls in Sector 144 Noida Escorts >༒8448380779 Escort ServiceBDSM⚡Call Girls in Sector 144 Noida Escorts >༒8448380779 Escort Service
BDSM⚡Call Girls in Sector 144 Noida Escorts >༒8448380779 Escort ServiceDelhi Call girls
 
How to utilize calculated properties in your HubSpot setups
How to utilize calculated properties in your HubSpot setupsHow to utilize calculated properties in your HubSpot setups
How to utilize calculated properties in your HubSpot setupsssuser4571da
 
Branding strategies of new company .pptx
Branding strategies of new company .pptxBranding strategies of new company .pptx
Branding strategies of new company .pptxVikasTiwari846641
 
Digital-Marketing-Into-by-Zoraiz-Ahmad.pptx
Digital-Marketing-Into-by-Zoraiz-Ahmad.pptxDigital-Marketing-Into-by-Zoraiz-Ahmad.pptx
Digital-Marketing-Into-by-Zoraiz-Ahmad.pptxZACGaming
 
What is Google Search Console and What is it provide?
What is Google Search Console and What is it provide?What is Google Search Console and What is it provide?
What is Google Search Console and What is it provide?riteshhsociall
 
BDSM⚡Call Girls in Sector 128 Noida Escorts >༒8448380779 Escort Service
BDSM⚡Call Girls in Sector 128 Noida Escorts >༒8448380779 Escort ServiceBDSM⚡Call Girls in Sector 128 Noida Escorts >༒8448380779 Escort Service
BDSM⚡Call Girls in Sector 128 Noida Escorts >༒8448380779 Escort ServiceDelhi Call girls
 
Unlocking the Mystery of the Voynich Manuscript
Unlocking the Mystery of the Voynich ManuscriptUnlocking the Mystery of the Voynich Manuscript
Unlocking the Mystery of the Voynich Manuscriptelizabethella096
 
Social media, ppt. Features, characteristics
Social media, ppt. Features, characteristicsSocial media, ppt. Features, characteristics
Social media, ppt. Features, characteristicswasim792942
 
How to Leverage Behavioral Science Insights for Direct Mail Success
How to Leverage Behavioral Science Insights for Direct Mail SuccessHow to Leverage Behavioral Science Insights for Direct Mail Success
How to Leverage Behavioral Science Insights for Direct Mail SuccessAggregage
 
Situation Analysis | Management Company.
Situation Analysis | Management Company.Situation Analysis | Management Company.
Situation Analysis | Management Company.DanielaQuiroz63
 
personal branding kit for music business
personal branding kit for music businesspersonal branding kit for music business
personal branding kit for music businessbrjohnson6
 
The+State+of+Careers+In+Retention+Marketing-2.pdf
The+State+of+Careers+In+Retention+Marketing-2.pdfThe+State+of+Careers+In+Retention+Marketing-2.pdf
The+State+of+Careers+In+Retention+Marketing-2.pdfSocial Samosa
 
Instant Digital Issuance: An Overview With Critical First Touch Best Practices
Instant Digital Issuance: An Overview With Critical First Touch Best PracticesInstant Digital Issuance: An Overview With Critical First Touch Best Practices
Instant Digital Issuance: An Overview With Critical First Touch Best PracticesMedia Logic
 
Brand experience Dream Center Peoria Presentation.pdf
Brand experience Dream Center Peoria Presentation.pdfBrand experience Dream Center Peoria Presentation.pdf
Brand experience Dream Center Peoria Presentation.pdftbatkhuu1
 

Último (20)

Digital Strategy Master Class - Andrew Rupert
Digital Strategy Master Class - Andrew RupertDigital Strategy Master Class - Andrew Rupert
Digital Strategy Master Class - Andrew Rupert
 
BDSM⚡Call Girls in Sector 144 Noida Escorts >༒8448380779 Escort Service
BDSM⚡Call Girls in Sector 144 Noida Escorts >༒8448380779 Escort ServiceBDSM⚡Call Girls in Sector 144 Noida Escorts >༒8448380779 Escort Service
BDSM⚡Call Girls in Sector 144 Noida Escorts >༒8448380779 Escort Service
 
How to utilize calculated properties in your HubSpot setups
How to utilize calculated properties in your HubSpot setupsHow to utilize calculated properties in your HubSpot setups
How to utilize calculated properties in your HubSpot setups
 
Branding strategies of new company .pptx
Branding strategies of new company .pptxBranding strategies of new company .pptx
Branding strategies of new company .pptx
 
Foundation First - Why Your Website and Content Matters - David Pisarek
Foundation First - Why Your Website and Content Matters - David PisarekFoundation First - Why Your Website and Content Matters - David Pisarek
Foundation First - Why Your Website and Content Matters - David Pisarek
 
Brand Strategy Master Class - Juntae DeLane
Brand Strategy Master Class - Juntae DeLaneBrand Strategy Master Class - Juntae DeLane
Brand Strategy Master Class - Juntae DeLane
 
Creator Influencer Strategy Master Class - Corinne Rose Guirgis
Creator Influencer Strategy Master Class - Corinne Rose GuirgisCreator Influencer Strategy Master Class - Corinne Rose Guirgis
Creator Influencer Strategy Master Class - Corinne Rose Guirgis
 
Digital-Marketing-Into-by-Zoraiz-Ahmad.pptx
Digital-Marketing-Into-by-Zoraiz-Ahmad.pptxDigital-Marketing-Into-by-Zoraiz-Ahmad.pptx
Digital-Marketing-Into-by-Zoraiz-Ahmad.pptx
 
What is Google Search Console and What is it provide?
What is Google Search Console and What is it provide?What is Google Search Console and What is it provide?
What is Google Search Console and What is it provide?
 
The Future of Brands on LinkedIn - Alison Kaltman
The Future of Brands on LinkedIn - Alison KaltmanThe Future of Brands on LinkedIn - Alison Kaltman
The Future of Brands on LinkedIn - Alison Kaltman
 
BDSM⚡Call Girls in Sector 128 Noida Escorts >༒8448380779 Escort Service
BDSM⚡Call Girls in Sector 128 Noida Escorts >༒8448380779 Escort ServiceBDSM⚡Call Girls in Sector 128 Noida Escorts >༒8448380779 Escort Service
BDSM⚡Call Girls in Sector 128 Noida Escorts >༒8448380779 Escort Service
 
Unlocking the Mystery of the Voynich Manuscript
Unlocking the Mystery of the Voynich ManuscriptUnlocking the Mystery of the Voynich Manuscript
Unlocking the Mystery of the Voynich Manuscript
 
Social media, ppt. Features, characteristics
Social media, ppt. Features, characteristicsSocial media, ppt. Features, characteristics
Social media, ppt. Features, characteristics
 
How to Leverage Behavioral Science Insights for Direct Mail Success
How to Leverage Behavioral Science Insights for Direct Mail SuccessHow to Leverage Behavioral Science Insights for Direct Mail Success
How to Leverage Behavioral Science Insights for Direct Mail Success
 
Situation Analysis | Management Company.
Situation Analysis | Management Company.Situation Analysis | Management Company.
Situation Analysis | Management Company.
 
personal branding kit for music business
personal branding kit for music businesspersonal branding kit for music business
personal branding kit for music business
 
The+State+of+Careers+In+Retention+Marketing-2.pdf
The+State+of+Careers+In+Retention+Marketing-2.pdfThe+State+of+Careers+In+Retention+Marketing-2.pdf
The+State+of+Careers+In+Retention+Marketing-2.pdf
 
LinkedIn Social Selling Master Class - David Wong
LinkedIn Social Selling Master Class - David WongLinkedIn Social Selling Master Class - David Wong
LinkedIn Social Selling Master Class - David Wong
 
Instant Digital Issuance: An Overview With Critical First Touch Best Practices
Instant Digital Issuance: An Overview With Critical First Touch Best PracticesInstant Digital Issuance: An Overview With Critical First Touch Best Practices
Instant Digital Issuance: An Overview With Critical First Touch Best Practices
 
Brand experience Dream Center Peoria Presentation.pdf
Brand experience Dream Center Peoria Presentation.pdfBrand experience Dream Center Peoria Presentation.pdf
Brand experience Dream Center Peoria Presentation.pdf
 

Innovations in Market Mix Modelling

  • 1. © Absolutdata 2014 Proprietary and Confidential Chicago New York London Dubai New Delhi Bangalore SingaporeSan Francisco www.absolutdata.com April 28, 2014 Our exploration into innovations in Marketing Mix Modelling Meet-up Data Science Oxford
  • 2. © Absolutdata 2014 Proprietary and Confidential 2 Agenda NLMixed & MCMC to automate the fitting of s-curves and more The statistical Challenges and the curse of the s-curve What is Marketing Mix Modeling? VARx Modelling Discussion   
  • 3. © Absolutdata 2014 Proprietary and Confidential 3 Strategic vs. tactical decisions Spend Brand Equity Engagement Web Purchase Profile /segment Preferences Propensity Models Strategic How to spread the budget? What campaigns? Product design Pricing strategy ??? Tactical Who? What? When? ??? Decision support modelling and tools
  • 4. © Absolutdata 2014 Proprietary and Confidential 4 Measurement of effectiveness & efficiency by marketing driver – Answering key business questions  Identify brand/s’ volume growth/decline drivers  Explain what is driving year over year change in sales  Determine relative importance & estimates elasticity of different drivers  Identify the competition and assesses impact  Determine the key competitive levers that impact/interact with own brand  Build a robust framework for investment decisions for short term - brands that should be supported  Determine an optimal realistic investment plan for the future based on simulated scenarios
  • 5. © Absolutdata 2014 Proprietary and Confidential 5 Magazine Online Print Radio TV Overall Sales Affiliate Clicks Paid Search Clicks Display Clicks Decision Support Analysis We would like to measure the direct and indirect impact of marketing investment at a granularity relevant to planning
  • 6. © Absolutdata 2014 Proprietary and Confidential 6 Marketing mix modeling & optimization @ Absolutdata focus on shift towards true attribution & predictive spend mix Data Collection From Disparate Data Sources Data Review – Confirming directional movements of prominent variables Review past performance to determine CONTRIBUTION of each marketing mix element Traditional Regression Based Modeling Techniques such as OLS DIY Simulator What-If Scenario Planning Optimized Media Calendar Objective Function & Constraints Attribution & ROI Measurement – Multi channel influence and interactions Advanced Modeling Techniques such as Auto Regressive, VARX, SEM Added model granularity at customer segment and regional level
  • 7. © Absolutdata 2014 Proprietary and Confidential 7 In most cases the data is highly noisy The key challenge it to find a statistically sound solution that can help the business The exogenous (independent) time series are highly inter- correlated The business requires an impact estimate for the actions they can take in order to evaluate strategies (what if analysis) The business would prefer not to hear about dimension reduction S-curves
  • 8. © Absolutdata 2014 Proprietary and Confidential 8 Business understanding is introduces through transformation Ad-stock Saturation Effects  Ad stock captures exponential decay effect of GRPs - TV  This depends on the temporal effect of GRPs, can be done for Print, Radio as well  Research suggests that immediate response generated by advertising follows an S-curve  It introduces three additional parameters: saturation rate, point of inflexion and ‘half-life’ parameter (for carryover effect) as unknowns in the model  Together this captures the diminishing return of advertising Ad-stock impact which depends on the temporal effect of GRPs Carry-Over Effect SampleOutput
  • 9. © Absolutdata 2014 Proprietary and Confidential 9 Various research indicated that immediate response generated by advertising is followed exponential decay Ad-stock carry over is estimated and reported in terms of half-life Current Effect Half-Life K= Carry Over Carry-Over Effect
  • 10. © Absolutdata 2014 Proprietary and Confidential 10 0 200 400 600 800 1000 1200 1400 1600 1800 0 50 100 150 200 250 300 350 400 450 RevenueImpact (ExcludingCarry-overeffect) $ Spent per Week Inflexion Point Saturation Level Adstock Equation: Where, X is actual spending, K is decay constant and determined by expression exp(ln(0.5)/t1/2); V as saturation parameter and Xd as diminishing return point (point of inflexion) S-Curve Impact Carry-over ImpactAd-stock Impact At= 1/(1+exp(-V*(X-Xd ))) + K*At-1 Half-Life Research suggests that immediate response generated by advertising can be modeled using an S-curve followed by exponential decay of effect which helps capture the diminishing return of advertising S-Curves transformation reflect the belief that Marketing spend reduces in its effectiveness at a certain point and it impact decays over time
  • 11. © Absolutdata 2014 Proprietary and Confidential 11 Standardize the variables De-trend and account for seasonality effect Fit an s-curve for each variable optimising it for regression to the residuals to the Revenue trend What is more appropriate from a businesswise perspective: Univariate or Multivariate? Standardize the variables Select a subset Apply s-curves Fit a regression model explaining the Revenue Evaluate quality of fit Multivariate – optimize s-curves to work together in a particular setting Manually optimize Univariate – fit a s-curve for each variable on its own Automate
  • 12. © Absolutdata 2014 Proprietary and Confidential 12 Instead of calculating the s-curve from t=0, I concentrate on the last n lags: I explored two sas procedures for fitting the s-curves Fit aggregation model where S is the dependent and the residual to the Revenue trend is the independent (no intercept for now – might need to reconsider that) NLMIXED MCMC
  • 13. © Absolutdata 2014 Proprietary and Confidential 13 My code – if you must %Macro HazSThree(SVar); S_Mue=1/(1+exp(-sV*(&SVar._lag&Nlags.-sXd))) ; %do l=%sysevalf(&Nlags.-1) %to 0 %by -1; S_Mue=1/(1+exp(-sV*(&SVar._lag&l.-sXd))) + sK*S_Mue; %end; %mend proc nlmixed data=HAZ.Detrend MAXITER=4000 maxfunc=4000; ods select ParameterEstimates; parms b1=1 se=1 sV=1 sXd=0 SK=0.25; bounds se>0; bounds sV>0; bounds 0<SK<1; bounds b1>0; %HazSThree(&HazVar.); Mue= b1*S_Mue; model Detrended ~ normal(mue, se); proc mcmc data=HAZ.Detrend &MCMCOptions.; ods select PostSummaries ; parms b1 se sV sXd SK ; prior b1 ~ normal(mean = 1, var = 0.5); prior se ~ igamma(shape = 3/10, scale = 10/3); prior SK ~ uniform(0,1); prior sXd ~ uniform(-2,2); prior sV ~ uniform(0,10); %HazSThree(&HazVar.); Mue=b1*S_Mue; model Detrended ~ n(mue, sd = se); For illustrative purposes the code shown here fits only one s-cure. The solution actually fits a baseline and all the curves
  • 14. © Absolutdata 2014 Proprietary and Confidential 14 Vector Auto Regression (VAR) time Steady state progression Target Sales/Signups time Co dependency associationIndigenous Web search time Marketing impact & decay Exogenous campaigns
  • 15. © Absolutdata 2014 Proprietary and Confidential 15 Phase I: Top down marketing mix modeling Phase III: Reconcile MMM & Cookie Attribution Phase IV: Reporting, Simulatio n and Optimization Phase I: Marketing Mix Modeling Phase II: Cookie- Based Attribution Algorithm Search Clicks Affiliates Display Impressions TV Impacts AffiliatesSecondary Relationships Search Signups Email Signups Print Signups Signups from Other Factors Previous Day’s Baseline Signups +TV GI Signups Display Signups+ + + + + Daily Signups=
  • 16. © Absolutdata 2014 Proprietary and Confidential 16 Secondary attribution provides A refined view of the system Paid Search Clicks Non paid search Cable Total Impact 11.4% 9.0% 2.5% 3.8% -1.0% 2.2% -0.1% 2.6%-3.8% -2.2% Actual TV Attribution taking into account indirect contribution of Search Final Attribution 7.5% 5.7% 11.1% SampleOutput -0.1%
  • 17. © Absolutdata 2014 Proprietary and Confidential 17 Applying VARx to a BIG number of SKUs How to choose priors We have encountered interesting challenges Challenges Modelling short term and long term effect Cookie data allows to not only do attribution but identify key sequences
  • 18. © Absolutdata 2014 Proprietary and Confidential 18 Incomplete sales (Target) data We do not know what is sold when by the distributers We sell to their stock
  • 19. © Absolutdata 2014 Proprietary and Confidential 19 http://www.linkedin.com/groupItem?view=&gid=130238&item=233249172&type=member&commentID=5810144471664320512&trk=hb_ntf_COMMENTED_O N_GROUP_DISCUSSION_YOU_FOLLOWED#commentID_5810144471664320512 Friends do not let friends to use Excel for statistical analysis
  • 20. © Absolutdata 2014 Proprietary and Confidential 20 Absolutdata provide analytics based solutions to address business critical issues 40% increase in profits through Conjoint based Pricing Optimization – A top SaaS company $50MM increase in revenue by Market Mix Modeling across 4 geographies – A leading CPG Company 15% revenue growth through Multi Channel Attribution – A large ecommerce company $23MM increase in Customer Loyalty and CRM marketing revenue – A major Hotel chain $9MM incremental revenue as a result of focused promotional campaigns created – A major Online Retail Discounter Contribution of $78MM over the last few years to their margins – A major Retailer We are decision scientists who help decision makers take better and informed decisions
  • 21. Eli Y. Kling Director - Analytics Phone: +44 (0)7940094976 Email: Eli.Kling@absolutdata.com LinkedIn: Uk.linkedin.com/in/elikling Follow us on: