Part of me would prefer this post be about “all politics are local”,  or “to think globally, act locally”,  or some other bumper sticker phraseology.  Alas, it’s about a theoretical and practical problem that won’t neatly fit on your bumper.

What if your best practice, or your best statistical model,  or your best appeal,  or your best training for your canvassers wasn’t the overall best (the global best),  but instead, just the best of the options you’ve tried (the local best)?   If you ponder that rhetorical for an extra moment you’ll likely find that many or even most of our “best” likely lives in the locally optimized world.

This ought to be both our source of angst and motivation.  It ought to be what keeps us up at night.

I’m reminded of a recent conversation where a perfectly competent analyst said their model predicting attrition included 4 predictors/variables.

Statistical models are incomprehensibly fast but also incomprehensibly dumb.  Models and statistical software don’t know what’s missing.  Nor do they have any domain expertise or theory-derived solutions to your problem (e.g. retention) to guide the model beyond what some human provides as parameters.

The “fact” that there are 4 predictors/variables ought to be source of fear and inspiration–triggering  the analyst to question what’s missing.   If only the world of human behavior could be neatly reduced down to four things or variables.  That only happens if you gave the model 5 or 10ish  factors to consider.  Sure, then there’s perhaps only four that the statistical software deemed different enough from each other and impactful enough on the attrition outcome to rise to the top.

Heck, if the 5 or 10 you started with were all of the same data type (e.g. demographics, passive secondary data) then they may be redundant enough with each other that only 2 things “matter”.  This is locally optimized.

It is not only a technical and theoretical problem, it’s a problem of mindset.  The issue of global vs. local plagues those parts of our sector who believe their control appeal represents the ‘best’ when in fact it only represents the middling average.  Unfortunately, trying to beat the average is tough– unless you think beyond the one-size-fits-all world that the control lives in.

Here’s a Thought Experiment

Do your donors all have the exact same motivation and reason for supporting your organization?

If you instinctively or knowingly said “no” to this question then you are instantly on a path to discovering that your “global” best control has to be plural– controls.  At a bare minimum,  but also as a useful starting point, can you imagine one control for Group A with Motivation X and another control for Group B with motivation Y?

You’ve just broken through to a new testing paradigm.  And it frankly is far less complicated than the spreadsheet with 67 different rows of random RFM categories and/or list sources that are used to break out results.  This Group A, Motivation X and Group B, Motivation Y is a 2×2.  4 groups.  Inherently different based on who they are.

Too much of our fundraising world is locally optimized.  That work, built up over decades of experience and learned best practice has value in that it’s predictable.  It shouldn’t be abandoned but it’s showing its age in shrinking donor pools and flat to declining returns with constant upward, cost pressure.

Leave that as your control world.  Stop testing in it.  Test around it.  Cheaply, quickly and with a globally informed re-think.  Find winners in that new world.  Throw them over the fence to the ‘control’ world then rinse and repeat.

Kevin