CoolData blog

17 April 2010

Models outside of fundraising

Filed under: Model building, Private sector — Tags: , , , — kevinmacdonell @ 2:13 pm

(Image used under Creative Commons license. Click image for source.)

Today’s post strays well outside the realm of nonprofit fundraising, so feel free to give it a miss. It’s all about predictive modeling, though, so if you’re still with me, please indulge me while I stray. (And toward the end, I will bring the discussion back to nonprofit fundraising.)

I work in university advancement, but most of my working life (so far) was spent in newspapers and magazines, as a writer and editor. When my mind returns to my old work these days, I think more about the business side of things, in particular the subscription department, more properly known as circulation.

It was while working for a magazine publisher that I encountered my first database. I developed and maintained a very simple database (more of a flat file, really) for indexing editorial content for later retrieval. Another database, a more sophisticated one which I had nothing to do with, was used to manage subscriptions and mailings.

Having rediscovered the database in a whole new way via predictive modeling for nonprofits, I can’t help wonder what sorts of things I could do for a business such as my former employer. What about a magazine (or newspaper) subscription predictive model? Why not?

Some things a publisher might want to predict include:

  • Who is most likely to renew. Or most likely to commit to a multi-year subscription.
  • Who is most at risk of letting their subscription lapse.
  • Which lapsed subscribers are most likely to be enticed back.
  • Who is most at risk of cancelling their subscription.
  • Who is most likely to purchase a gift subscription.
  • Who is most likely to be successfully cross-promoted to another magazine put out by the same publisher.
  • Who is most likely to purchase books or other promoted products.

Applied to promotions, a renewal-propensity model would help target dollars in the most cost-effective way. The publisher could cut back on expensive incentives to the most loyal subscribers, who are going to renew anyway. As well, former subscribers who are least likely to return can be proactively identified as a dead loss, and cut loose. That would free up dollars to target current at-risk subscribers, and to lapsed subscribers who are most likely to return.

Potential predictor variables for a subscriber model include:

  • Geography (from address)
  • Rural/urban (from postal code)
  • Payment method
  • Gift subs purchased
  • Source of original sub (newsstand sale, referred by friend, received gift sub)
  • Lapses in subs
  • Phone number present
  • Books purchased
  • Letters to the magazine
  • Surveying data
  • Complaints on file
  • Gender (imputed from first name or name prefix)
  • Subscribes to other magazines
  • Has placed an ad
  • Free-text comment fields
  • Various patterns in subscription behaviour
  • Has ever had an issue returned undeliverable

Let me bring this back to fundraising for non-profits. Think about all the different kinds of magazines you see on the newsstand. Imagine how different the subscriber bases are from each other! The characteristic attributes of those database constituents are incredibly specific to geography, lifestyle, hobbies, political leanings – you name it. It isn’t hard to imagine that having a rural postal code is predictive of renewing a subscription to a gardening magazine, but no one would ever leap to the conclusion that therefore a rural code is predictive for, say, The New Yorker, or any other type of magazine.

So why would we accept any assumptions about what is predictive of giving to our institutions? Are we all alike? Nope!


Blog at