Skip to main content
Three Techniques for Scientifically Balancing Complex Marketing Decisions


By Shalini Raghavan

Many companies are getting a head start with Big Marketing using predictive models with tried-and-true business rules libraries for common marketing decisions, such as up-selling and cross-selling strategies. Yet, with bigger and bigger data sets and an expanding range of analytic predictions to take into account, the decision process can grow quite complex. Companies that rely on simple business rules alone may soon find it difficult to manage or even fully comprehend how they are making decisions.

Use of data-driven techniques for modeling and optimizing decisions processes is essential for effectively bringing Big Marketing intelligence into operations.  Some key techniques that we are using with our clients today include:

  1. Action-effect models predict how likely the customer is to respond to a particular action within a particular time period (e.g., is this customer likely to purchase if offered a 20 percent discount on lawn mowers valid for 30 days?) and what the effect would be on KPIs (e.g., response rate, sales, interest income).Used in the most sophisticated Big Marketing projects, these simultaneously deployed models are deployed in the dozens, if not thousands. Working with a leading retailer, for example, we used automated techniques to generate propensity models for every offer available in the retailer’s loyalty program. These models can be generated at the sub-category level. They may also be based on custom groupings.  These propensities, along with investment goals and eligibility criteria for each offer, are input into a mathematical optimization to match offers to consumers in a manner that maximizes overall response.
  2. Optimization scientifically balances what the company is trying to accomplish with real-world constraints under which the business and its partners operate.  Let’s say the objective of a campaign is to achieve the highest incremental sales that have a positive margin. Additional constraints might require the campaign to target a maximum of two million customers and stay within the combined budgets of Supplier A, which has a promotional budget of $5,000, and Supplier B, which has a budget of $10,000. Only customers who have not purchased the item in the past 120 days will be targeted, and customers whose local store does not carry the promotional items will be excluded.
  3. Simulation allow marketers to modify one or more constraints to see how the optimal decision strategy shifts, exploring trade-offs and choosing the best operating point for their business. In the case of the retailer loyalty program described above, the result of optimization and simulation is a frequently refreshed selection – from literally trillions of possibilities – of a unique set of relevant offers for each customer in the program. This has driven its response rates as high as 30 percent.

As data becomes more abundant, decisions become more complex and nuanced. By leveraging analytic science to balance these decisions, companies can improve customer service, build loyalty and gain a competitive advantage. For more information on this topic, check out our Insights paper (registration required).

related posts