how market mix modeling can impact cross-channel budget and business planning
TRANSCRIPT
Cross Channel StrategyHow Market Mix Modeling Can Impact Cross-Channel
Budget and Business Planning
Speakers:Dhiraj Rajaram, Mu Sigma
Craig Kronzer, UnitedHealth Group
Session Objectives
• Learn approaches to Market Mix modeling – how it enables measurement of multi-channel activities
• Discover the advanced framework to quantify ‘true’ cost of acquisition, netting out cross channel effects and cannibalization
• Evaluate tools and platforms for budget scenario planning and optimize marketing budget allocation
BACKGROUND
Organization Overview
• Established in 1998 as a AARP/UHG relationship
• Nation's largest supplemental insurance program focusing on people age 50 and over
• Distribution: DTC, Employer, Agent, Web
• Largest provider of pure-play decision sciences and analytics services
• 30 Fortune 500 Clients in 10 Industry Verticals
• Headquartered in Chicago IL with presence all over the US
Insurance Solutions
Business Problem
Background• Insurance Solutions uses
multiple marketing channels to attract members
• Operational constraints result in less than complete attribution of sales to marketing efforts
• Several sales are not attributable to any of the marketing channels
Business Hypotheses• The business wanted to
test the hypothesis that unattributed sales are driven by marketing
• In particular, there was a need to understand the impact of DRTV on sales
• The solution framework required to measure cross-channel impacts
The Challenge of Measurement
Attribution by Channel
Sales
DRTVOther ChannelsUnattributed
• A major portion of sales is unattributed to any advertising channel
• Sales attributed to DRTV are low compared to proportion of investment
• Business wants to measure the true effect of TV advertising by understanding the “halo effect”
The Need for Measurement
Channel 4
DRTV
Channel 2
Channel 1
Cost per Sale
• Due to relatively low attribution of sales to DRTV, the apparent cost of acquisition for the channel is high
• There is a need for improved measurement to calculate the ‘true’ cost of acquisition
• Cost of acquisition is a key component in marketing planning
Cost of Acquisition
SOLUTION APPROACH
Problem Solving Framework# Strand The Why?
The What & How?
1 SCQInitial
XXX YYY ZZZ
XXX YYY ZZZ
2 Factor Network
XXX YYY ZZZ
XXX YYY ZZZ
3 Hypothesis Matrix
XXX YYY ZZZ
XXX YYY ZZZ
4 SCQFinal
XXX YYY ZZZ
XXX YYY ZZZ
SCQInitial SCQFinal
SCQFinal SCQInitial
FactorNetworkFactorNetwork
HypothesisMatrixHypothesisMatrix
The Mu Sigma Problem DNA ensures appropriate emphasis on design and hypothesis leading to right representation
Solution ApproachThe ms Factor Network
Factor 2
Factor 3
Factor 4
Problem
Factor 1
Factor N
Mapping the exhaustive set of factors enables testing of all relevant hypotheses
Sales
Direct Marketing
DRTV
Other DTC
Agent
Seasonal Index
TrendExternal
Economic
Product price changes
Competitor data
Demographic data *
The Market Mix Framework
• The Market Mix Framework decomposes total sales into contributions by advertising vehicles and external factors
• Contributions from different channels enable calculation of ROI
MMX Modeling ApproachesDirect Marketing
Agent
Direct Response TV
Marketing Mix Model
Sales = f(DM, DRTV, Print, Web, Events…)
Contribution
Percentage of enrollments due to each promotional program
Total and Marginal ROI for each program
Cost per Sale
Lifetime Value
Optimization
Promotional spend allocation at aggregate program level taking into account diminishing marginal
Portfolio level optimization for all products
WebPro
mo
tio
nal
Act
ivit
y
Measurement of diminishing returns
Multiplicative
Unattributed Sales
Measurement of individual contributions
AdditiveMeasurement of
cross channel effects
Multi Target
Ad stock – Lagged effectsDM enrollments Effort adstock 0.2 adstock 0.7
Weeks
Enro
llmen
ts
xx
Adstock transformation methodology
At = Tt + λ At-1
Where:• Tt is the value of the marketing variable at time t
• λ is the decay or lag weight parameter• At-1 is the carryover of Advertising at time t-1
“HALO” EFFECTS AND REATTRIBUTION
Multi-target Model
Total Sales
DRTV Sales
Channel 1 Sales
…
Unattributed Phone Sales
Other unattributed
sales
Each of the target sales modeled on all advertising inputs as well as external factor
Reattributed Sales Original Attribution
Post Modeling Reattribution
Sales
DRTVOther ChannelsUnattributed
Sales
DRTVOther ChannelsUnattributed
The Market Mix models are able to measure the contribution of advertising to previously unattributed sales
Improved measurement
Channel 4
DRTV
Channel 2
Channel 1
Cost per Sale
Original CPS
Channel 4
Channel 2
DRTV
Channel 1
Cost per Sale
Reattributed CPS
Due to higher level of attribution in sales, the effective cost per sale reduces significantly
Halo EffectContribution of Media Activities
DRTV Channel 1 Channel 2 Channel 3 Channel 4 Channel 5
Enrollments from channel
DRTV
Channel 1
Channel 2
Channel 3
Channel 4
Channel 5
Unattributed
Halo Effect
Self ContributionThe ‘halo’ effect of advertising channels enables
quantification of cross-channel contribution
Impact of the initiative
• Cost of sale calculated based on direct attribution used in budget planning
• Member lifetime value calculations biased by high cost of acquisition in some channels
• “Dark Test” conducted to verify impact of TV on unattributed sales
• The optimization process for allocating budget across channels refined by using ‘true’ cost of acquisition
• Budget allocation across marketing channels changed significantly
• “Bright Test” conducted to test additional advertising opportunities
Pre-MMX Modeling Post MMX Modeling
SPEAKER BIOS
Speaker Bios
Dhiraj Rajaram
• Founder and CEO of Mu Sigma, an analytics services company that helps clients such as Microsoft and Dell institutionalize data-driven decision making. Prior to founding Mu Sigma, he advised senior executives across a variety of verticals as a strategy and operations consultant at Booz Allen Hamilton and PricewaterhouseCoopers.
Craig Kronzer
• Leads a Data Analytics team for UnitedHealthcare. Team is responsible for enterprise-wide analytics including building predictive models, designing and analyzing marketing tests, and claim data analytics. Previously, was with Carlson Marketing Group and Lands' End. Craig holds an MS in Statistics from the University of Minnesota and BS in Computer Science from the University of Wisconsin.