Lead Scoring Models: Assigning Point Values

In my last post, I covered how to start building a lead scoring model and at what point in a company’s lifecycle it should be done (once the company has reached the expansion stage, is focused on sales and marketing, and has a high inflow of leads). To quickly review, the first step is to put together a data set of lead attributes (search phrase used, number of pages viewed, country of origin, etc.) along with whether the lead converted to an opportunity or sale. After the data set is assembled, the second step is to analyze each attribute’s (independent variable’s) correlation with conversion (the dependent variable). The idea in this step is to find attributes that are either strong positive or negative predictors of whether a lead will convert. For example, if leads that trialed the product convert at a higher rate than average, this would be a strong positive predictor of conversion. Alternatively, if leads that only viewed one page of the web converted at a much lower rate than average, this would be a good negative predictor of conversion.

After each attribute’s correlation with conversion has been analyzed, the third step is to assign point values for each attribute (positive or negative). If you are creating a manual lead scoring model (as opposed to using multiple regression software to get attribute/variable point values), it is important to be consistent. A simple, consistent method of assigning point values is taking the overall conversion rate and subtracting it from the conversion rate of leads with a specific attribute. For example, if the overall lead to opportunity conversion rate is 10%, and the conversion rate for leads that have trialed the product is 25%, add 15 points to every lead that has trialed the product (25 minus 10). If you find that leads that come from Google Adwords campaigns only convert at 3%, deduct 7 points from every Google Adwords lead (3 minus 10).

When manually assigning point values, there are some pitfalls that should be avoided – namely assigning point values to attributes that have low sample sizes and assigning point values to attributes that are highly correlated with one another (multicollinearity). I will cover both in more detail next week.