Catching up on silicon-based fundraising

25 years and one day ago today, a computer program called Deep Blue beat the human chess champion Garry Kasparov for the first time ever.  Since then, AI/machine learning passed many other areas that had been humans’ domains: Jeopardy in 2011!, text-based conversations in 2014, Go in 2017, contract law in 2018, and so on.

But Kasparov vs Deep Blue, perhaps more than any other event, sparked the debate over the advantages of people over AI, and vice versa.

The challenge in the nonprofit sector isn’t that AI is going to take all our jobs (although we’ve talked about how natural language is improving to the point that machine learning can write).

The challenge is that we aren’t adopting it well enough to compete against other ways someone can spend a dollar from the for-profit world.

To oversimplify, humans are better at being human.  We empathize and have emotional intelligence.  We humans also do better at unstructured problem solving and discovering what is relevant in undefined phenomena.

Machine learning is best at pattern matching.  At some point, the resources a machine-learning algorithm passed the point that any one human could throw at a problem.  Then it surpassed what every human could throw at a problem.  You can’t beat something at chess that can see all the way down millions of possible pathways.

And yet we insist on using humans for pattern matching.  As an example, some organizations are still using RFM — recency, frequency, and monetary value — segmentation for their donor pieces.  And some are doing only part of this — mailing 0-24 in donor and everyone else in acquisition.  That’s not even RFM analysis: it’s just R.  This type of segmentation was outdated in years beginning with 1.

To be sure, this type of transactional information is vital for understanding a donor.  But so are:

  • Demographics
  • Capacity to give
  • Whether someone has been to your website
  • Volunteer status
  • Satisfaction with previous donation experiences
  • Whether someone answered your survey
  • Direct experience with your issue (e.g., had the disease, experienced poverty)
  • Event attendance
  • What content they’d interacted with
  • Disaster donor vs non-disaster donor
  • Whether they purchase monthly subscriptions to things

And on and on.  There are literally hundreds of variables that can be tossed into this rich data stew, each of which adds a bit about what or whether to communicate with this person.

No human can take all these factors into account when deciding who should get a mail piece or whether this person’s digital ads should be for one-time or monthly giving.

So we draw simple lines that fit categories we understand: everyone who donates 3+ times a year gets the monthly giving upgrade, 0-24 month donors get the mail newsletter, send every new donor this exacted email series.

To the combination of rich data and strong machine learning, our attempts to predict who will donate must look like watching a toddler trying to cram the red star-shaped block into the blue pentagon hole. 

We can do better by allowing machine learning to do what it does best — pattern recognition.  That frees us up to do wrestle with the many unstructured, strategic challenges that come at us each day.

Garry Kasparov is instructive.  He maintained his human championship for chess until 2005.  Since, he’s become chair of the Human Rights Foundation, founded the Renew Democracy Initiative, and is a powerful voice for the freedom that comes with liberal democracy.  In these realms, he’s unlikely to meet his match from a circuit board.

Sign up for Moore updates