In advertising it is known that half the money spent is wasted – what is not known is which half!
There are various data science applications to help advertisers decide on the best usage of their marketing budget. Marketing Mix Modeling (MMM) is a method that helps quantifying the impact of various marketing inputs or attributes to sales or market share. The purpose of MMM is to identify how much each marketing input contributes to sales and how much should be spent on each marketing input. MMM uses various statistical methods.
Objectives of Market Mix Modeling
The objectives of MMM are to discover the sales drivers, understand and measure sales KPIs and ROI, predict Future Sales Performance and optimize Marketing Budgets.
MMM uses statistical analyses like multivariate regressions and time-series analysis to estimate the impact of marketing inputs/tactics on sales and then forecast the impact on future marketing inputs/tactics.
Data Requirements in developing Market Mix Models
Various types of data are used while developing a marketing mix model –
- Product Data (no.of units sold of product, price of product, sub-products)
- Promotion Data – Day when promotions/offers was active & offer type (discount, cash back etc.)
- Ad data or ad spend ($) – digital spend, affiliate marketing spend, content marketing spend, social media spend etc.
- Seasonality – summer/winter, holidays, events
- Geographical data – city, state, postal code of store, serviced states
- Macroeconomic data – inflation, GDP (include to understand recession, cyclical effects)
- Sales – vol. of units sold and revenue $
Data Science methods for MMM
When the regression model is built, sales is the dependent variable and the other input parameters are the independent variables. To begin with Univariate statistics like mean, standard deviation, quartiles etc. are analyzed for the variables and different slices as well. Correlation matrix is used to understand the carry over effect of the ads over time. The carry over effect and the concept of diminishing return are used in the regression model.
Apart from multiple regression other methods like SUR model – Seemingly Unrelated Regression, Bayesian Modeling, Partial Least Square regression etc. are also used while developing market mix modeling in data science.
Modern tools are built in data science for market mix models. Proprietary tools are available from marketing and research agencies while models are developed in python and R languages as well. Market mix modeling is a marketing application of analytics and data science for advertisers and marketing professionals. These models and tools help marketers play around with various input parameters and optimize their marketing budget based on the sales they would like to generate.