Home / News / Predictive marketing: there is no “poor” data !

Predictive marketing: there is no " poor " data

Expert's tips / 6 May 2019

As you understand, the creation of predictive scores relies on data. The algorithms will be built on what is called an ABT, Analytical Base Table … In fact it is nothing more than a large database that will contain as many lines as customers / prospects and as much columns of information (fields) on these customers … Simple!

During my meetings with clients, often comes the question … “What data is needed for the algorithms to work properly?”; that means produce “scores” that are close to the reality finally, with a very good power of prediction!

My answer is systematic … “do not deprive us of any information available because we can have a good surprise …”. From experience, however, it is necessary to gather (and this is part of the work done by INBOX) information, various and varied from different operating systems or already collected at the advertiser (we can dream) in a datalake:

 

  • Transactional information: who bought what, where, what payment and in what timeframe? … The Retail organized the harvest of this transactional info … For many advertisers (insurers, banks, B2B players, … This transactional history is available and exploitable for predictive purposes. Convenient!
  • Relational information: who was contacted / solicited by which channel and with what results: mail, e-mail, incoming call, outgoing call, visit, …
  • Operational information: complaint, from the plain, intervention … If there is a DB specific to Customer Care, this information will be interesting …
  • … And these “three small dots” are important! Do not deprive us of information that may be “discriminating” from the commercial behavior we are trying to predict. Our team was working recently on a project and in the Set-Up phase, the client asks us if it is appropriate to entrust us the results of a sponsorship/member gets member system set up recently … Our answer: certainly!

Unsurprisingly, it turned out that sponsoring a friend was clearly a sign of commitment … to the brand and has a negative influence on attrition. It also emerged in the construction of the score that the customers who joined the company by way of its “pairs” also has particular behaviors …

In conclusion …

Let’s forget to have a 360 ° view of the client (how awful) but do not deprive us of any available and exploitable information to build the predictive algorithm. The score can only be better!

Contact us