Customer Predictive Analytics
Predictively Analyze Your Customers and Score the Best-Matching Prospects
Customer Predictive Analytics involves using statistical regression techniques to determine the most important demographic/behavioral components and their relative weights. This information can be projected onto the U.S. Consumer database to score new prospects based on how well they match your best customer profile. Once the third-party consumer database has been scored, it can be used to pull best prospects for Multi-Channel Acquisition Campaigns.
Different Approaches: Self-Contained Analytic Solutions vs. Custom Analytics
Custom Projects for Customer Predictive Analytics
Custom analytics — Using R, Alteryx or other statistical software tools — which your data scientists and analysts can perform (or which Data Marketing Strategies can help working together with one of our Analytics Partners) will ultimately perform better than self-contained / self-service solutions. If you do undertake this project, we highly recommend Enhancing Your Customer Data with additional third-party data to improve model performance.
A custom analysis will allow you to take advantage of what you already know about your customers such as last purchase date, purchase frequency, purchase amount, life-time value, number of visits to website, etc.
The downside is that custom models will take time to develop. Perhaps the most time-consuming aspect involves “data engineering” which essentially means getting your data into the proper formats required for various types of statistical processes to work properly. Tools such as Alteryx are very good at this…even so this is typically a substantial project. And unless you have a local instance of a U.S. Consumer database, you will not be able to apply scores to individual prospect records, if reaching best prospects is your goal.
Self-Contained/Self-Service Customer Predictive Analytics
By connecting the analytic environment to an instance of the U.S. Consumer database ahead of time and standardizing connections to all relevant data elements (e.g., age, gender, income, marital status, interests, behaviors, etc.), then the data engineering phase is performed once and the process is standardized. Now all that is required is to match uploaded customer name/address records into the secure analytics environment, and match them in order to be ready to analyze the customer data.
A variety of predictive analytic techniques and workflows can also be developed up front and implemented in a standardized way. Different statistical algorithms can even compete with each other for best model results.
The advantages of self-contained solutions in terms of time and effort can be huge. Project turnaround can be reduced from days or weeks to hours or minutes with commensurate cost savings.
But there are also shortcomings associated with various offerings currently available on the market. The greatest is probably that no single solution provider (currently) offers a complete set of reference data about individuals and households. For example, one of the organizations we work with is a Christian non-profit. But the self-contained solution we were investigating from one source did not include the data element for Religion, so the model that would have resulted would not have been accurate.
This is why it is critical to work with a trusted source of advice on how to proceed. We might actually recommend testing solution A vs. B to see which works best.
Self-Contained Customer Predictive Analytics
Because of their relatively low cost, quick turnaround, optimized analytics based on multi-variate regression and use of high-quality third-party data, it is possible to get great results from the self-contained models, especially if the goal is to use the results to score the U.S. Consumer database for prospecting campaigns.
To perform the analysis it is necessary to upload between 5,000 and 50,000 names and addresses and a list of ZIP Codes to be used as the reference (for national brands we would just use the entire U.S.). There is a small fee for the model, some of which can be applied toward the purchase of targeted lists. Hundreds of demographic elements are analyzed and the most predictive ones are identified. Here is an example of a data element found to be highly predictive in a recent customer analysis:
This example shows that males age 55 to 64 (vertical bars) are overly represented in the local market area (red line) and that this is a highly predictive element in the model.
So when selecting prospect lists based on decile 1 or 2, you would expect to see this age and gender range highly represented in the list, along with other highly predictive data elements.