ALTERYX INSPIRE | Join us this May for for a multi-day virtual analytics + data science experience like no other! Register Now

Alteryx Designer Discussions

Find answers, ask questions, and share expertise about Alteryx Designer.

Engaged vs Non-Engaged Users Analysis

5 - Atom

Hi there,

I have two datasets with a list of engaged and non-engaged users for a marketing campaign. I want to perform a predictive analysis to determine the BEST audience to target int he next campaign for example, what Company Industry we should target, Job Title to target etc. Most of the models on Alteryx deal with Quantitative data, but I am dealing with Qualitative data here.

I want to get the score for each predictor value for each variable that impact whether it is an engaged user and this would help me select that audience in the future for improved marketing. Help needed please! 
See attached for a sample of the dataset. Thank you!


Primarily, you will want to tidy that data up, either split the data in 2 for the different record types, or combine fields in an appropriate way to remove most of the nulls. You won't find much significance in a field with mostly Nulls, and if you do, then you should be skeptical unless you can say why they are Null. 


There is a sample under 'Help > Sample Workflows > Predictive Analytics > 11 New Donor/12 Donor Score' that show how to then choose a model and use it to score the data. That sample has instructions as to why each step was taken and is a great intro into this kind of analysis.



7 - Meteor

Sounds like you are doing something close to "propensity to buy" analysis.  I would suggest logit or probit modeling.  The Scaler in the "Score" algorithm seems to deal with the nastiness of logit and probit coefficients quite well.  


Both Logit and Probit modeling is scaled to deal with Binary variables, a yes, or no, Engaged or Not Engaged.  


Or if you really need the coefficients from the regression you can do a Linear regression where your target variable is the binary variable.  Yeah, you are breaking some of the best practices for regression modeling, but it makes it a lot easier to read and understand.  If you need something quick and dirty, I vote for linear.  If you need something exact and reliable, then go with Logit or Probit.