Patterns and Key Words of Predictive Analytics Questions

Not all questions are “predictive analytics questions”. That is to say, not every question you receive as a data science practitioner requires predictive analytics to answer.

So how do you know if someone is asking a question that’s a good fit for using predictive analytics to provide the answer?

Patterns and Key Words

From a recent read through Predictive Analytics, I jotted down a few patterns that emerged as each application of predictive analytics was presented. Apendix B in the book is an excellent summary of twenty predictive analytics applications across different industries.

First, let’s take a look at a few examples of predictive questions. Then let’s see if we can find the pattern or identify some key words.

  • “Which customers will respond to marketing contact?”
  • “Which ad is a customer most likely to click?”
  • “Which employees will quit?”

When you hear people asking questions like, “Which [customer/ad/employees] will [respond/click/quit]“, it should trigger “predictive analytics” in your mind as a reasonable way to approach getting the answer to the question.

Key words to listen for are:

  • Which [noun] – The questioner is wanting to narrow things down to a particular set… “nouns of interest”, if you will.
  • Will – There’s a future-focus to the question.
  • Likely – When probability enters the vocabulary of the question, there’s a chance that predictive analytics can help with the answer.
  • [Take action] – Fill in [take action] with just about any verb of importance. When you hear someone asking whether someone/something will respond, or click, or quit, or [you name it], predictive analytics could be a tool that helps shed some light on the answer.

Here are a few more: * “Is this e-mail spam or not?” * “Will the stock go up or down?” * “Will the player win or lose?”

Key words to listen for are:

  • “___ or ___” – in each of the questions, the answer is in one of a few categories: [is/is not], [up/down], [win/lose]. There need not be only two categories. The point is that any time the questioner is attempting ot figure out whether [something] fits in one category or another, predictive analytics could lend a hand in sorting things out.
  • [final outcome] – Many times, questioners are wanting to know the probable final outcome of some event that hasn’t occurred yet. When that’s the case, let predictive analytics be a tool you use for giving insight into the answer.
comments powered by Disqus