Are You Doing Business in 107 Countries? Or 7? Data Hygiene Matters in Predictive Analytics

March 29, 2011 - 3 minutes read

On a recent assignment for setting up lead generation in Salesforce, we used an existing customer database to build a statistical model to score a leads database. The client does business in 7 countries, or so they said, and I believed them. But they quickly added a caveat, “no one has looked at our database in awhile.”

First, we looked at their billing country field. This was an open text field in their Salesforce system that could be edited by just about anyone. What we found was amazing. There were so many variations for each country that unique values quickly proliferated to 107. Misspelling, case difference, punctuation and abbreviations all conspired to create many versions of the same country!

Our first task was to identify the obvious countries and group them, followed by corrections to the remaining data. This took 107 countries down to the correct 7. Then the country field was locked. Going forward, a drop-down menu of countries is being used to prevent this proliferation from happening again.

For statistical modelers, why is clean data important for sales intelligence?

First of all, we want to point out that even a simple field can pose a challenge.  In the absence of consistent coding guidance, it becomes hard to create a segmentation or master-filter at the top with which to analyze data. In this case, it is not wise to mix multiple countries within the analysis.

Second, there may be other fields with which data can be cross-referenced. If there is a shipping country and a billing country, chances are that both ought to be the same. So correct the data, then carry over the values into the other field so you have fewer blank and invalid rows.

Third, the state of data quality can give a clue as to what other fields could be a problem. Even if not used in modeling, always keep in mind how this data was created, who and when it is updated, and what procedures are used to correct bad data.

Finally, proactively start addressing quality and instill an ongoing practice of making data cleanliness everyone’s responsibility. Have a method by which to collect feedback and incorporate it. This way, when you are ready to perform analyses, there are fewer surprises with respect to data quality.

Parting thought:  Do not assume anything “should be obvious” or find where to lay blame for bad data. As you can see from the simple example above, all businesses have to approach data hygiene with care, caution and respect. As my friend’s father, an electrician said:  “The day I stop fearing electricity is the day I will stop working.”