Talking Numbers: What is Your Quest?


Monty Python and the Holy Grail Bridgekeeper

What is your name?

What is your quest?

What is your favorite color?

Arguably the most difficult skill for any analyst to learn is how to effectively communicate the numbers to the business. This act of translation is the most important moment in any analysis because it is when we, as number crunchers, get the buy-in (I know, corporate-speak) of people who actually do stuff with what we’re telling them (a.k.a. decision makers) … or not.

Tip for Translation: Remember the Dependent Variable

Usually, when doing any form of analysis, the dependent variable is the one that you want to impact using the model that you’ve created. This may take the form of a key performance indicator, lifetime value measurement, customer segment, or other (hopefully influencable) output. This is what the customer or manager cares about.

For most business-minded people and non-analysts, the use of a specific modeling technique or statistical test is meaningless and may well cause a general glazing over of eyes. Instead, focus on how the results can be used to map and/or change behaviors, save money, gain them more revenue, acquire customers, or whatever combination it is they’re looking for.

Example: Uplift Model Explanation

I recently spoke with a former client who had gotten back in touch with my company about a potential project that would need marketing analytics. The client described the initiative as a way to generate goodwill amongst customers before they were hit with a price increase. Their plan was to send out what I refer to as a "cheerleading campaign": a feel-good piece with no real offer but a lot of fluff about how wonderful the brand is. The goal, of course, would be to get customers more excited and engaged with the brand so they would either not notice the increases or, at least, would be less likely to defect to a competitor.

Their objective, and the sensitive timing of the communication, brought to mind uplift modeling. In an uplift model, four main groups of customers are identified with differing responses. Here’s the spin on each (with ear-perking points in bold).

  • Likely to Stay Regardless of Communication (Sure Things) – This group does not need a communication sent to them telling them that the brand is great. They have no intention of leaving. The client can save money by not sending a piece and will not sacrifice their relationships in doing so.
  • Likely to Leave Regardless of Communication (Lost Causes) – These customers are out the door no matter what. Usually, these are the ones that are either already dissatisfied or are very price-sensitive. Sending a communication with no "save" offer will not help retain them. Once these customers are identified, either send an offer in the piece or don’t bother sending it at all. You can save your pennies, even if you can’t save the customer.
  • Likely to Leave Only If Sent a Communication (Do Not Disturbs) – The logic here is that you increase the "piss-off" factor for customers by sending a communication and reminding them that their price is about to increase or juxtaposing the increase against a Ra-Ra Brand message. Sending communications to these folks is sending good communications dollars after bad customers, a.k.a. a really bad idea.
  • Likely to Stay Only If They Receive Communication (Persuadables) – These are the Only group that are really worth communicating with at this stage. You can change behavior to increase retention among these customers but only if you generate the goodwill you are looking for through communication.

By phrasing an analysis in terms of what the business user is looking for, you are more likely to sell your services, gain traction for your analyses within your organization, or at least keep the attention of your audience when discussing the numbers.

One thought on “Talking Numbers: What is Your Quest?

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.