A common question we hear from clients is whether mathematical models used for target marketing always work, and will it work for my business?
Whether models always work can be a subjective question, unless we define the parameters for success at the beginning of a project. It is true that marketing models vary in performance but they invariably result in successfully predicting the desired outcome. The level of model performance though is primarily due to:
- Uncovering strong patterns that exist in the data
- Data quality
- Data enhancement
- Use of strong analytical techniques.
Ultimately, success is determined by whether a positive Return on Investment (ROI) is generated.
In regards to model performance, some models will show response rates in the top scored names that are two or three times the response rates in the lowest scored names. Whether it is possible to get two or three times lift in response is often down to the available data, and the relationships that exist in the data as potential predictors for outcome measures like the response rate. Even if we are able to lift response by say a modest 10% -20%, it is often meaningful in terms of the ROI. The ROI can easily be computed by looking at the cost of using the model, incremental sales generated, and the lifetime value of these incremental sales. Carefully created control groups to measure business as usual against analytic-driven targeting is recommended.
Another common question is why a model works well in one geography but not as well in another geography.
This is typical and is often due to the different profile of these geographies which generate different results. Usually though, a model “lifts all boats”. So, an underperforming geography can be lifted as well as an already strong performing geography. The result is that each geography is performing very differently but is still lifted from where they were without a model.
A natural next question is then how do I get started to use modeling to enhance the effectiveness of my current target marketing programs.
As long as either customer or campaign data exists, a strong analyst with strong software can analyze the data and use the relationships that exist to help predict a desired outcome.
In general, if the data is collected, cleaned and handled by a strong data analyst, and if there are any data relationships existing, you will undoubtedly be able to develop a predictive model that can predict an outcome of interest. The question is not whether a model will work, but whether it generates enough incremental value to provide a positive return on investment.
Sign up here to subscribe to the blog