BEYOND SENTIMENT; WORKING TOWARDS A SOCIAL NPS John Clarvis John.clarvis@Gmail.comAll content Property of John Clarvis,28/11/11
DEVELOPING A SOCIAL NPS Social NPS is a misnomer – NPS has always been social Recommendations are social acts involving an object (to be recommended) and others (to recommend to) The key questions addressed here to develop a social NPS are: What is the object being recommended? Who is it being recommended to? When is the recommendation effective? How is this recommendation effective?
WHAT DATA (DON’T) WE HAVE? Buzz volume (brandwatch,radian6 et al)Raw metrics Buzz sentiment Spend Derived metrics Cohort Source Reliability Reach Relevance Outcome Footfall Page views variables Buzz growth Page views Seasonality Concurrent campaigns Extraneous/Control Variables Competitor activity
FINDING THE RIGHT WORDSMost Social data is word (string) based This creates too many variables to be manageable Word frequency is measured Data reduction analyses are used to Each word becomes a Process the results variable – with an attached value derived by its frequency Factor and Cluster analysis is then Used to derive manageable numbers of Each word is entered Variables into a factor analysis to create word groups and create variables
WORKING BACKWARDSTo develop an algorithm for a social NPS we need to decide What we arerecommending, and Who We are recommending it too. By entering each factor into alinear regression we can establish the importance of each factor to the outcomevariable.Each variable is assigned a coefficient, which indicates the direction that it influences abehaviour (i.e. positively or negatively) and the amount that it influences in that direction.In a simple example our Social NPS algorithm may look like:OUTCOME VARIABLE Homepage views = Relevance (1.30)+ Sentiment (2.40)+reach(1.10) +mozrank*(0.9 The Social NPS score is calculated for each mention based on this formu The social NPS works the same as a normal NPS and on the same scale
TOOLS USED Brandwatch (paid subscription) R Statistics (open source - www.r-project.org) Twitter Client for R (cran.r- project.org/web/packages/twitteR/vignettes/twitteR.pdf) Google Analytics5 Beta