Analytics Translators work with business leaders to find and prioritize projects where analytics can help choose the right course of action. They translate the problems into analytical questions to be answered, while taking into account data quality and resource availability. In other words, they help turn a business decision into a question that can be answered by the data.
If you are an analyst in a medium to large size company, you probably encountered this phenomena. Most new programs as tests you assess are not worth doing, and most existing programs have a low ROI. The implication is that you get to deliver bad news to business stakeholders. A lot!
I received my Graduate Certificate in Data Science from Harvard Extension School this January. It took me two years since I was taking one class at a time, and he is the summary of my experience with the program.
In his post he describes the application of a test vs control design to breaking down how different channels of communication perform separately vs when used together. This is a good application of the controlled experiment methods.
SAS has created a special feature in Enterprise Miner called Incremental Response modeling that helps you find relevant input attributes differentiating between customers have higher incremental response rates and those that don’t. A modeling procedure can predict uplift for every customer and then put customers into deciles based on their potential incremental response due to a marketing campaign.
Uplift modeling is generally harder to execute than response or propensity modeling. While response follows known customer traits (demographics, lifestage, transience, change in circumstances), uplift can be dependent on variables not commonly used for analysis. After watching a few failed attempts to create uplift models, I can identify the most common barrier in creating an valid uplift model: marketing programs that are ineffective.
The goal of control group selection is to keep the test (treatment) and control (holdout) groups representative of each other, both in terms of sample composition and in terms of measurement. The only difference between the groups should be where your treatment is applied.
Did you ever wonder why your results always seem to go the same way? Many analysis setups have subtle biases that drive conclusions in the same direction. Learn to identify and eliminate them for high quality results.
It’s much more efficient to retain existing customers than to gain new ones. Learn from my experience in telecom services on how to use analytics to reduce attrition
A/B testing is a special case of experimental design, aimed at determining a better performing option of communication. In practical terms, it means that all of the targets in the A/B test get exposure to some version of marketing communication while in test vs control design only the test targets get exposed.