Multichannel Marketing Mix - PMSA Mix Modeling • the ability to link offline brand advertising to...
Transcript of Multichannel Marketing Mix - PMSA Mix Modeling • the ability to link offline brand advertising to...
Multichannel Marketing MixP. K. Kannan
How multi‐channel marketing mix works
Advertising Analytics 2.0
Credit: Nichols, Wes. “Advertising Analytics 2.0.” Harvard Business Review March 2013.
Marketing Mix Modeling
• the ability to link offline brand advertising to business outcomes and understand how on‐and offline marketing interact
• Steps:– Attribution– Optimization– Allocation
Challenge of Multichannel Mix
• Spillovers across channels and carryover within channel complicate attribution
• Managerial actions upon which outcomes depend on are not randomly set– Outcome is not dependent on a experiment– Endogeneity
• Optimization – how do you handle uncertainty?
Dynamic Optimization
FromFischer, Wagner, and Albers, MSI Working Paper 13‐114 (2013)
Allocation Mechanisms
• Naïve Allocation: – equal distribution across all allocation units –specific marketing activity for specific product
• Percentage‐of‐sales rule:– Percentage of previous year’s sales for each product
• Attractiveness allocation heuristic• Numerical optimization
Modeling Spillovers
• Simultaneous equation systems– Sales in each channel as a function of budget allocation across different marketing instruments
– Modeling direct effect and indirect effects– Multi‐period modeling with aggregate data– Complement with individual level analysis
• Challenges– Endogeneity– Having enough variation in data
Challenges in Modeling
• Iterative development in model– Initial to final involves adding, removing variables, interaction terms, functional forms
• New marketing instruments added – e.g. social media– Drastic changes in parameter estimates– Collinearity problems– Granularity of data
Bayesian Methods to the Rescue
What’s the advantage?
• We can specify priors for estimates based on external information– Experiments– Prior studies– Managerial intuition
• Hierarchical Bayes– Higher level parameters as priors for lower level estimates
– National level estimates used as market level priors
Case Study 1 ‐ Retailer
• Sales through multiple channels• Click‐based attribution model for online sales• Display experiments – spillover to search• Tracking impressions social media, experiments• Integrate offline marketing spend• Develop multichannel marketing mix –Bayesian methods
Case Study 2 – Insurance Industry
• Measuring spillovers using aggregate data• Developing individual level path data – data warehousing capability
• Linking online multichannel spend along TV spend
• Modeling using hidden Markov models• Dynamic optimization methods for allocation
What is the next frontier?
• Modeling the supply side – why?• Recall figure from last presentation
$ $Budget Set by MgmtMonthly/Quarterly
$ $
Consumers’ PurchaseFunnel
Budget Set by MgmtMonthly/Quarterly
Modeling Approaches
1. Hidden Markov Models2. Nested Logit Model3. Generalized Poisson4. VAR Models5. Machine Learning
$ $Budget Set by Mgmt Daily
Another option
• Limited experimentation
Challenges for Pharma
• Direct to physicians – Detailing – Sampling
• Direct to consumers– TV spend– Social Media– Other media
• Linking the path for patient decision making
Questions?
Contact InformationP. K. Kannan
Ralph J. Tyser Professor of Marketing ScienceChair, Department of Marketing
Smith School of BusinessUniversity of MarylandCollege Park, MD 20742
[email protected]: 301‐405‐2188