Predictive Analytics for Direct Marketing
How would you like to improve your direct marketing campaign's results for the same budget? Or achieve the same results for less money? What if you could target your best prospects and ignore those who wouldn't be interested in your offer anyway? There is a way to achieve all that, all the while, increasing your ROI and having a predictable, steady stream of new leads every month. It's Predictive Modeling.
If you incorporate direct mail into your marketing program already, you might not be achieving the superior results you could be. This is especially true if you do large mailings. If you're not using direct mail at all, there's a good chance you're missing out on a great opportunity for business growth. The infographic below provides a simple overview of how to apply predictive data modeling to a direct marketing campaign. Whether you hire a statistician or outsource it, this is the general process that's followed.
After discussing expectations and determining goals, look at your respondent data. This can be from a past campaign or you can do a test campaign to gather new data. You'll want to start with a list of 5,000 - 20,000 or more consumers. You can even start with your customer list. This list can consist of responders and non-responders, buyers and non-buyers, donors and non-donors, etc. Amount spent, number of purchases, and last date of purchase are all pieces of data that can come in handy for the modeler.
2. Clean & Append
You'll want to ensure that you have clean (i.e. up-to-date and accurate) data. If you're starting out with a test campaign and you rent a list from a vendor, the list should already be clean. If you're starting with your customer database, you'll want to ensure the addresses are valid and correct. You'll need to do an NCOA/CASS update. If you combine multiple lists, you'll have to merge records and remove duplicates. If you provide your customer database to a vendor, they should update it for you.
Next, the consumer characteristics must be appended, or matched, to the list. These characteristics will be the variables that the modeler uses to develop a statistical model. If you're using a vendor, ask them about the match rate.
3. Model Development
At this stage, a model is developed using the respondent list and the appended characteristics. The modeler will use regression techniques to identify the characteristics of the best prospects, then test and validate the model for quality. The algorithm developed from this process is applied to other prospects in the next stage.
4. Rank and Find
The model is then used to "score" a larger data set. For example, the modeler can score the prospects in a city, state, or the rest of the country. The prospects are then ranked starting with the best at the top.
You focus your direct marketing on the best prospects and ignore those unlikely to respond. You get a higher response rate and save money on printing and postage. As you keep track of responses, you provide that data to the modeler who will compile it over the course of several months. Every six months, the modeler should use this accumulated response data to improve the model and sharpen your targeting. Thus, the repeatable "learning cycle" repeats.