Predictive analytics are generally very good, to the point that people are living in different product bubbles from their neighbors. (In Weapons of Math Destruction, Cathy O’Neil tells the story of an investor saying that their new technology would make sure he would “never have to see another ad for the University of Phoenix.”)

But there are two (at least) areas where algorithms provide poor results in the for-profit and nonprofit worlds.

Repetition. This is exemplified best in this tweet:

Dear Amazon, I bought a toilet seat because I needed one. Necessity, not desire. I do not collect them. I am not a toilet seat addict. No matter how temptingly you email me, I'm not going to think, oh go on then, just one more toilet seat, I'll treat myself.

— Jac Rayner (@GirlFromBlupo) April 6, 2018

Algorithms are not (yet) good at distinguishing when you are buying something to fill a need or a larger desire. That is, am I buying a hammer to hang a picture, to get a set of tools one at a time, or to add to my hammer collection?

All the algorithm sees is “bought a hammer” as the most recent (and thus relevant) transaction. So you will have hammer or toilet seat or whatever ads following you around the Internet. Enough people buy hammer number two to make this a worthwhile proposition.

This same thing happens when nonprofits first try customizing their communications to individual donors. We know the best communication to reacquire a lapsed donor, or prevent their lapsing, is often the same communication that was used to bring them in initially. If someone responds to a match campaign, they are more likely to want another one. The nonprofit is looking at this the same way the algorithm is – I’m getting rewarded for doing this thing; I will do it more.

The trick is that breaking out of transactional data and asking donors can yield better results. For one, you can ask donors what type of issues they are interested in and gear your communications that way. You will likely have a few different topics to talk about in case the first gets stale. Remember that the acquisition communication can sometimes be an accident of history; their real interest might lie elsewhere.

More than that, you can ask people how committed they are to you (or their intent to give) right at acquisition. This is the equivalent of “are you interested in more hammers?” Not only do donors not object to these questions – they like being able to set a relationship with you on their terms.

You just donated because your aunt passed and this is the “in lieu of flowers?” We can take your donation and part as acquaintances.

You donated just for an emergency, but want to learn a little more? Let’s keep talking. Maybe you’ll build an interest; maybe not.

You have been a volunteer for 20 years but just made your first online donation? You, ma’am, are my target donor. You collect hammers.

The algorithm based on transactional data is looking at the same data for all three of these people. You, however, can differentiate among them because you asked.

The Heisenburg effect. No, not from Breaking Bad. Heisenburg found that you can find the position of a particle or the momentum of a particle but not both because the act of observing affects the observed.

I’ll use myself as an example. I’m a frequent audiobook user and, as I’m affectionately (?) known at DonorVoice as “the cheapest bastard on the planet,” I often buy Audible’s Deal of the Day.

I’m clearly not the only one. Most of my “people who liked this book also liked…” recommendations for deal books are other deal books (no longer on sale).

Those aren’t books that are naturally bought together. They shouldn’t show up together. They are ones I rejected even though they were cheap. But the act of marketing affected the marketed-to.

How does this occur for nonprofits? Let’s say you introduce a new technique, a match for the sake of this example. It works. You get another match. It also works. Pretty soon you have four matches running throughout the year.

Over time, two things will happen:
1. Donors who like matches will be more likely to retain and those who don’t will be less likely. And you will be acquiring match-positive donors. You are changing the composition of your donor file, building one that responds to this tactic and driving away others.
2. Those donors who pay attention to your communications will start waiting for matches – their preferred tactic.

Replace “match” with “dog story,” “premium,” or “advocacy appeal” and you can see how relying on any one mass-market strategy can turn your file into a monoculture, driving away people who would have wished you well.

Not every appeal is meant for everyone. If dog stories work better than cat stories for your full list, be aware of the cat people who are looking for a reason to donate to you but possibly not finding it.