To recap: We have covered three of the five steps to optimize your data: data aggregation (Multichannel Campaigns: Getting Your Data Together), data augmentation (Multichannel Campaigns: Filling in the Gaps in Your Data), and data processing (Multichannel Campaigns: Making Your Data Useful).
The next step in the process is data analysis, or making your data meaningful. At the risk of getting kicked out of the statisticians’ secret society, I must confess that this is the easiest step — it is also the most fun. The analysis stage is the easiest one because there are essentially two factors that drive all analyses: data and objectives. Once you have carefully identified and defined each, it is almost like following a recipe.
For example, if your broad objective is to target customers for special treatment or with a particular product, you could use either some form of regression model or CHAID (chi-square automatic interaction detection), in which relationships between dependent variables and predictor variables are examined. CHAID models are typically used (in the predictive sense) when your data contain a high proportion of categorical data and/or in situations in which there is a lot of missing data. (Of course, there is the matter of correctly interpreting the output … but that’s beyond the scope of this article.)
Data analysis is also the most fun part of the analytical process. Here, you get to see up-close what is working – and with with whom – from among all of your carefully designed marketing tactics. You’ll also see the implications and interpretations for program enhancements and tactical manipulations. It’s the feedback – like getting that report card or knowing the score of a game – and seeing where you are exceeding your goals and where you need to improve.
Data analysis makes your data meaningful – in the context of your business and marketing objectives. To simplify analytics, think of analytical models as existing in four general classes – defined by the objectives of your analysis:
- The first class, and the most fundamental, is description – in which you describe who your customers are and how they behave and respond. Descriptive analyses summarize and simplify data – presented in tabular or graphical form – that describe the customer in terms of the central tendencies and variations within and across customer groups. <br>Descriptive summaries serve as excellent baselines of customer “performance” and provide snapshots of what is happening with your programs at any given point. Descriptive analyses, which are often called exploratory data analyses (EDA), should generally precede all other analytical efforts, because it will familiarize you with the general “scene” before you start the next effort. Although descriptive analyses should precede all other analytical efforts, the remaining classes are in no particular order – but are often used together in a comprehensive and meaningful analysis program.
- The second class involves segmentation models. Segmentation modeling entails grouping customers based on their needs, preferences, behaviors, and/or value contributions. The resulting customer segments respond to tactics and/or channels in ways that are useful to marketers in designing effective programs.
- The third class includes prediction and targeting models. Just as it sounds, marketers can use these models to make meaningful generalizations about the future behavior of customers based on past behavior and identify and select customers or groups of customers for differential or special treatment. These models enable marketers to identify channel propensities, content and offer preferences, the likelihood to churn, and more. This insight is extremely useful in designing and executing programs.
- The fourth class includes tracking models, which are employed to assess the impact of communications on consumer behavior over time. We design our programs to align tactics with targeted behaviors; tracking tells us whether we are successful. Tracking models, including various ROI models, provide insight into the impact of specific communications and offers as well as trends in engagement.
These models take the idiosyncrasies of channels into account while helping to determine whether tactics are eliciting the desired response and resulting in sustained profitability. Of course, tracking should not only indicate whether programs are working at an aggregate level, but should also indicate specifically where things are working and where they are not for fine-tuning.
Katie Cole, Ph.D, is vice president of research and analytics for Denver-based Merkle|Quris, the e-mail marketing agency of Merkle.