The first step in the MDS marketing science approach is to capture the impact of marketing efforts on customer and market behaviors in the form of causal, statistical models. The statistical models are built from a mix of data specific to each client situation, but can include survey data, transactional sales data, web visit data, and customer interaction or CRM data. Within our modeling approach, we factor out the impact of non marketing activities that are beyond the marketers' control such as seasonality or one-off events such as merger impacts, in order to reliably estimate the causal impact of controllable marketing efforts on customer or market behavior.
With causal models of customer and / or market behavior in place, we then link the predicted behaviors to their downstream economic impacts on the client company. Although the economic and financial metrics we model are tailored to each client, these metrics typically include:
With the linkages made between marketing efforts and customer behavior, and between customer behavior and financial metrics, we now have a platform for market simulation and optimization and for evaluating alternative marketing decisions and actions.
To provide decision support, we develop market simulators that allow us to explore the economic impact of alternative marketing decisions and actions. Although these simulators are always tailored to each client, the following are typical of the marketing actions that we simulate in our work:
With a market simulator in place, the next step is to optimize the key economic metrics within operational, strategic, and organizational constraints. We have found through experience, that it is the artful use of constraints in optimization that allows marketing science to have significant impact on a real marketing organization.
The simulation and optimization phase, almost always results in recommended decisions that involve some degree of extrapolation into areas of less familiarity. In order to test and refine our decision recommendations, we frequently develop market experiments. Experiments allow us to test our recommendations on a reduced scale, or in a less risky way, and to collect additional data to improve our modeling. We use both in-market experiments and pre-market experiments using the internet as a consumer lab, as mechanisms for testing and learning. The design of effective experiments is driven by a mix of technical statistical design considerations, and practical constraints and limitations. As a practical matter, market experiments must achieve financial results at the same time that new data and learning's are collected.
