Lead Scoring Models: Testing the Model

Once point values have been assigned to important lead attributes, and the dangers of low sample sizes and multicollinearity have been thwarted, it is time to test the model.

Testing the model is a fairly simple process. First, create a spreadsheet with a column for the lead name, a column for conversion (yes/no), and a column for each variable/attribute that has an assigned point value. Then, calculate each lead score by summing up the points (positive or negative) for each lead’s attributes. Next, sort the spreadsheet by lead score.

If the data set has 500 leads, of which 50 converted to opportunities or sales, the top 50 lead scores of a perfect lead scoring model will capture all of the conversions. In reality, building a perfect lead scoring model is nearly impossible, but even if the top 250 lead scores capture all 50 opportunities/sales, you have eliminated 250 leads that sales would have otherwise wasted its time on, which it could now spend interacting with prospects that have a high probability of buying.

A good way to visualize the model is to calculate lead score percentile and cumulative percentage of captured opportunities, and graph cumulative percentage of captured opportunities (y-axis) against lead score percentile (x-axis). For example, if the top 50 (10%) lead scores captures 10 (20%) conversions, (10,20) would be one point on the lead scoring curve. The chart below shows a sample lead score curve compared to an ideal curve (the top 50 (10%) lead scores capture all 50 conversions) and to a random curve (the top 10% of lead scores captures 10% of the conversions, the top 25% of lead scores captures 25% of conversions, the top 50% of lead scores captures 50% of conversions, etc.). The ideal model is represented by the green dashed line, the sample model is the solid orange line, and the random model is represented by the gray dashed line.

Once the model has been tested and visualized, you may want to keep tweaking the point values for various attributes to see if you can get “lift” and move the orange line closer to the ideal green line. Different iterations of the model can be compared visually by graphing the iterations on the same chart. Once you are satisfied with the model, test it against a second, independent data set that was not used to calculate point values and create the model. If the model performs comparatively in the second test, it is time to operationalize the model and use it to optimize your sales and marketing efforts. Next week, I will cover how a model can be operationalized in an expansion stage company.