Sonamine Website

  • Increase font size
  • Default font size
  • Decrease font size

Adding network analytics to lifetime value of customers

AddThis Social Bookmark Button

Lifetime value of customer – adding WOM and network effect

Most customer lifetime value models take into account what a customer will purchase from a business over the course of the entire relationship.   This study calls it the “intrinsic” value of a customer.  Standard DCF and add-on product concepts factor easily into such calculations.  However, recent studies have shown, organic word of mouth marketing was occurring and resulting in a 0.11% uptake among the network neighbors.   How do we add this “network marketing” value to the customer’s lifetime value?

This study models how to add network value to lifetime value.  The basic idea is to first model the intrinsic value using existing methods.  Then calculate the total value of the customer.   The difference is the network value of that customer.

What is interesting is that the study applied the model to a database of movie rankings to estimate the network value of various raters.  They were able to graph the distribution of network values for the raters.  ”The unit of graph is the average revenue that would be obtained by marketing to a customer in isolation… A network value of 20 for a given customer implies that by marketing to her we essentially get free marketing to an additional 20 customer.”

Typical distribution of network customer values

What they found is that there is no “average network value” for a customer.  Instead it seems to follow an exponentially decaying function.  A few raters had extremely high network values.  Their impact on the network is very high.  These might also correspond to the network hubs or media influencers.

The authors then proceeded to model out the best way to market to customers along 2 axes.

The first axis was how to search for the target prospects of the campaign. They proposed 4 ways to identify optimal propects using a combination of intrinsic value and network value –  (a) mass marketing ie blanket everyone (b) single pass (c) greedy search (d) hill climbing search.  b, c, and d are progressively more computationally intensive.

The second axis was based on the type of marketing campaign.  Three different offers were tested – free, discounted and advertising.

In all types of marketing offers, the mass-market scenario resulted in negative profit.  A customer that looked profitable on her own may actually have a negative overall value.  Taking into account network effects, using the single pass or greedy or hill climbing method, resulted in better profits that mass marketing.

So what?
Ignoring network effects can make an unprofitable decision look profitable.


Mining the network value of customers.  Domingos, P., Richardson, M.Proceedings of the seventh ACM SIGKDD international conference on Knowledge discovery and data mining 2001.  57-66.

Unfortunately the eachmovie dataset has been retired by HP.   The University of Minnesota has 10M movie rating data set found here


User Login

               No account yet?