Performance Reporting of Net Promoter Score data

I find many of my clients are adopting the Net Promoter Score (NPS) method for evaluating their voice of the customer measures.  It is based on questions and surveys of your customers on how well they would promote you to others.  It is a method to evaluate the customer loyalty rather than their satisfaction.

A good premiss, but it has the same numeric problem of all surveys.  Ordinal data!  The overall score is from 0 to 10, while the individual questions are on a 1 to 5 scale.  Larger is better.  So how to you track them.  The NPS system uses a weighted average to track performance, where 0 is a generally neutral customer and it can range from 100% to -100% based on a positive promotor (+) or a negative promotion (-).  Sounds fine, but it only looks at averages.

A student was evaluating the data with averages by week.  It all looks fine, except on weeks with little response counts.  We deal with that by combining adjacent weeks to keep the sample sizes up.  We recommend plotting it with an I-chart.

What is missing from the analysis is any evaluation of extreme values.  You may have a great average, but a key customer had a bad experience and scores you with a zero.  It may be a minor impact on the mean, but you may have lost a customer.  I recommended that the student develop a method to track the existence of bad scores.  They considered a range, but that is not great because of the ordinal nature of the data.  A better choice that one of our clients chose was to plot the % of scores that were bad (less than 3 in the 5 point and less than 6 in the 10 point)  In this way they can easily track the fraction of evaluations that are unacceptable.  If this increases you should evaluate.

It is a lesson from lean six sigma.  For non-attribute data, you should assess the average and the variability to understand the full nature of the process.