Do’s and Don’ts of Data Mining | npENGAGE

Do’s and Don’ts of Data Mining

By on Jul 10, 2013


I have spoken with a lot of organizational leaders that see the value in leveraging robust analytical solutions to help take their engagement to the next level.  Throughout the conversations, I’ve heard some great stories of success and some experiences we can learn from.  This month, I’m going to go through some of the approaches to take and approaches to avoid.

Do: Start with the Big Questions

Many times, we get caught up in very specific questions.  For instance, instead of asking, “What is my biggest opportunity in my direct marketing campaign?” – we ask, “What is my highest or lowest performing segments?”

Although knowing you highest and lowest performing segments are important in terms of understanding past performance, they may not be indicative of the greatest opportunity.  The opportunity may actually still exist with your highest performing segments.  Or maybe the segmentation variables aren’t yielding the best cohorts for measurement.

Start with the big question  – each time you answer a higher level question, other questions will continue to lead you down the path of the most valuable information.

Don’t:  Limit Yourself to Your Data

Too many organizations resist posing the ‘right’ question because they don’t feel that they have the right data to answer it.  I suggest an approach of starting with the right question and determine the data needed.  If you don’t have all of the information necessary to answer that question, then you have options.  First, you can attempt to acquire the right data – either through purchased appends, or start to collect it through your constituent interactions.  Data can be acquired through vendors, and you can modify donation forms, response devices, or added questions to telemarketing campaigns in order to solicit the relevant information.

Alternately, you can adjust the question while you gather the right information.  This is not the ideal approach, as the answer to the sub-optimal question is not typically as valuable as the answer to the ‘right’ question – but as long as you understand what the information that you are getting indicates, and you are moving toward answering the ‘right’ question, then you can move forward with the partial information knowing that it is partial information.

Without going through this exercise, you limit yourself to your existing data and avoid pursuing the data necessary to answer the most valuable question.

Do:  Say Away from Assumptions

Too many folks assume certain facts about their data and may be prone to starting from a faulty point.  For instance, I assume that the lowest performing segment in my direct marketing campaign is my area of greatest opportunity, so I focus on how to maximize the results from this segment.  Harkening back to the first Do on the list, this may be a faulty assumption.  Based on the make-up and prior performance of even the highest performing segments, they still may be the greatest area of opportunity, as historically they have even higher returns, while the lowest performing segments might have reached their peak and now you should decide whether to continue to include them at all (depending on their ROI).

Every assumption that you make potentially closes the door on a valuable insight.

Don’t: Stop

The poorest analogy that I can think of is this: mining your data is like potato farming.  If you stop digging when you reach the first layer of potatoes, you miss the more mature and larger potatoes buried deeper in the dirt.

I’ve spoken with many analysts that harvest a handful of useful analytical nuggets and stop, running to their strategists with their insights.  While discovering a useful nugget of information is exciting and should be celebrated, keep going.  Some of the most valuable potatoes are buried deep within your data.

I’ve never heard a good analyst ever say “I’ve learned everything I can learn from my data”…don’t be the first!



Comments (1)

Leave a Reply

Your email address will not be published. Required fields are marked *