I am currently working on a project with colleagues at my organization on a "Customer Journey Project", which will be launched shortly after the results of my current work. The objective of this project is to place each customer on our email list on a journey based on our programs. We've developed four main journeys based on the programs they have attended in the past and also based on their behaviors online with reference to wish list data, shopping cart activities, quotes, orders, web login, call log data, page views, clicks, keyword search on our website, etc. The journeys we formulated are based on the programs we offer and each customer are place in each journey base on their "inferred program preferences." The journeys are:
1. Destination journey ( ex: Alaska, California, Central America, South Atlantic, Western Europe, Europe, Asia, etc). The destination journey is further subseted into primary state, primary country, primary region and primary continent).
2. Lifestyle journey (ex. independent travelers, intergenerational travelers, service learning travelers, Living & Learning travelers, Physical Interest Travel (Easy vs. Challenging Programs).
3. Interest journey (Ex. Birding, Bicycling, Crafts, Music, Holidays & Festivals, Sports, Walking/Hiking, etc).
4. Program Line Journey (Ex. Domestic programs interest, International programs interest, Afloat program interests (ships, cruise ships, badges, river boats, etc.).
Thus, using a wide range of datasets my task is to come up with a list of "recommended" journeys for each customers that will be used by the email campaign team to target customers through a specific journey base on the result of my analysis.
The first part of this process is straight forward, that is, using the data available and those derived to compute a score for each customer base on the variables by program. At the end of the day, I will have a spreadsheet like result, which shows the each customer's email address, programs they have attended or might have clicked on or visited based on their page views and wish list and bunch of other stuff, as well as their overall predictive score/ranked score as well as the program line, and the other three journeys listed above.
Thus and given the result, it is more likely that each customer will be in or have activity in more than one journeys listed above. Thus, I want to be able to select the most dominant journey for each customer (base on the data) and in that process recommend which journey journey they should be promoted with via email. Is there an easy way to do this in JMP?
Jenkins: For starters this is a typical, yet pretty complex project that can take many twists and turns. I suggest following a classic data mining/predictive modeling workflow that has the following major components...of which JMP and if you've got JMP Pro, better still, can certainly play a role.
1. Articulate the business problem...it looks like from your original post your team has thought this through.
2. Find/collect relevant data...not just data you have access too...but data that's needed...for responses and factors.
3. Explore the data for relevance, quality, messiness, missing, etc. JMP can help here with the Column and Modeling Utilities platforms. Decide how you'll handle problematic variables/observations. Think about how you'll handle cross validation in modeling...perhaps add a validation construct variable to the data set.
4. Explore the data for correlations, where's the middle, how spread out is it, shape, anything unusual...multiple JMP platforms play in this space from Graph Builder, Distribution, Fit Y by X, Multivariate (and all the subplatforms therein).
5. Start building and evaluating models...OLS, Logistic (for categorical responses), PLS, Generalized Regression subplatforms, GLM, and others could play a role here. If you've got JMP Pro, evaluate your models in the Model Comparison platform.
6. Communicate your findings, maybe using the en masse export to MS applications such as Word or PowerPoint, or the HTML5 with Data interface.
7. Pilot a solution to verify modeling work.
8. Solve practical problem? If yes, celebrate...it no, circle back to find out why.
Good luck...wish I was there to help you
The processes you listed is exactly what I am currently doing. I am wondering if clustering could her group journeys base on the customers' interest preferences.
Jenkins: Clustering is certainly one of the multivariate exploratory data analysis methods that might yield some insight. With projects like your's often throws the kitchen sink (or close to it) at the data and move along as you gain insight. Some paths will bear fruit, others not so much. But as a colleague of mine once was fond of saying, "Sometimes knowing an 'is not' is just as informative as knowing an 'is'."
I think clustering might show some insights and a bunch of others JMP offers. Thanks for your inputs.