Home / News / #2 How frequently should you model your Churn scores?

#2 How frequently should you model your Churn scores?

Expert's tips / 27 February 2019

Customers sometimes ask us to advise them on the necessary frequency of modeling … The mean the frequency with which the model will have to assign a weight to the discriminating criteria of the behavior that the advertiser seeks to anticipate and which will be translated by a score. Are you still with me?

First, remember that behind these scores is an intention (sometimes still latent) of the consumer to adopt a commercial behavior: the acceptance of an offer, a (lack of) interest for a product or a category of products, a departure of the Company as a customer or subscriber … Some profiles thus release “high” scores, others “low” according to the discriminant characteristics that the Modeling will have determined and weighted.

Well, that little reminder being made …

A technical answer first, in two points

The re-modeling at a given frequency will have interest and legitimacy only if

  1. The explanatory criteria of the model are subject to change and observable in the meantime (a surf behavior, a contact with the company …) and if they are likely to be taken into account in the construction of the model
  2. We are in a position to verify the occurrence of the element we were trying to anticipate (the purchase, the departure, …), what is called the “y” in jargon

There is definitly an interest in modeling regularly (provided that these conditions are met) if the Marketeer wants to be able to base his/her observation on fresh scores and ensure the quality of the model regularly.

A triple business answer now : Insurance use case

  1. Let’s answer with an example… An insurance company decides to build a Churn score (also called an Attrition score) that will reflect the risk of the insured leaving the company. Depending on the risk that the model has calculated it will decide whether or not to send a satisfaction survey (assuming that it is better for the insured to have the opportunity to express his dissatisfaction) to all policyholders with a churn risk above 60%. The model is recalculated every day, because the discriminating variables influencing the model are likely to change every day: surfing behavior, occurrence of a disaster, contact with the Company, etc. Consequently, the individual scores are also likely to change. evolve. An individual who yesterday had a risk of Churn of 58% can today have a risk of 61% and as such receive the investigation. The automated action will be dependent on the calculated score and therefore there will be an interest in remodeling regularly and “re-scoring” the database.
  2. Another case … If the assigned score is intended to customize a response to an incoming request (inbound) there is obviously interest in modeling regularly because obviously this solicitation can occur at any time. If for example the score is used to build the argument of a call center receiving calls, or to condition a personal page on a site, …
  3. Finally, for a campaign initiated by the Marketer, at least as soon as a targeting will be considered, there will be an interest in re-scoring the database in order to select the most relevant profiles … that makes sense.

In conclusion it is the Business interest which, as often, guides the frequency of modeling as long as the technical conditions are fulfilled …


Albert DerasseAlbert Derasse

Managing Director & Partner at Inbox Belgium

Contact us