Should we explore causation? Well, the few fundraising shops with the resources to properly conduct that kind of experimental research should go for it. But I think they’ll find the underlying causes (such as student experience, and institutional reputation) are well outside their purview. Better that every part of the university — from admissions to the registrar’s office to student services to the faculty — understand that they are all partly responsible for the future financial health of the institution, to the extent that it relies on the support of alumni who found their time in university worthwhile and rewarding.

In the meantime, I see fundraisers obsessing over whether to use postage on return envelopes, obsessing over comma placement in letters, and on and on. Sure, that’s what they control (institutional reputation? not so much), but I think that time would be better spent figuring out who among our alumni have indicators of affinity but may not have expressed that affinity through giving yet.

That’s higher ed. The non-higher ed world, which I cannot speak to, may be very different.

We believe firmly in the need for the fundraising industry to get far more sophisticated in its targeting efforts, using predictive models, external data and yes, affinity markers or proxies. However, we also believe the industry needs to get much more serious about MAKING more good donors to augment more sophisticated targeting.

Kevin, in his comments, is 100% right when he says predictive modelers are not concerned with causation. They deal with the world as it is served up to them and what non-profits need to focus on is changing that world. Good donors are not born, they are created, but behavior based approaches alone will NEVER do a good job at CREATION, hence the dreadful retention rates.

Today’s 0 to 12 month donor is tomorrow’s 13 month plus donor. Using affinity markers or other behavior based proxies (beyond RFM variables) to get better at targeting those in the 13 plus who are “worthy” of being pulled back into 0 to 12 “status” is worthwhile but it is also a losing battle if not seriously bolstered with far more understanding of CAUSE and the confounding variables.

This is being done on a large scale by our commercial brethren who, in addition to spending oodles and oodles of money on data mining (which we support) also dedicate huge sums to brand building, customer service and relationship building.

More specifically, the confounding variable of “engagement” (we prefer to call it Commitment) can not only be defined and measured but managed by the non-profit to MAKE more good donors.

So, where we take issue with the comments to this post is twofold: 1) this is very doable and needs to be done. That it is not yet being done in any meaningful way in this sector only bolsters the argument. 2) we would argue the exact opposite of Kevin’s assertion that as non profits “we overestimate our ability to sway potential donors via our messages”. If the sector truly believes this then it is forever lost and we need to clean house. Seriously.

We know as well as anybody that targeting is the number one “variable” to dictate response on appeals – more so than the message. So yes, better targeting equals better response and more effort should be focused there. However, one CANNOT simply target his/her way to long term fiscal prosperity. The vast majority of donors attrite, they have negative lifetime values and the leaky bucket gets leakier and leakier all the time. This industry has been taking in its own laundry for too long and the tipping point is already upon us.

Let us also be clear, the case for putting resources towards “making more good donors” (i.e. educating, persuading, motivating and relationship building) need not be some soft, fuzzy, just-believe concept. It can be done with financial projections and metrics just as rigorous as the campaign and behavior based ones we are so familiar with in the lead generation business now called fundraising.

Thank you all for your comments and discussion on this critically important subject.

Roger

Some great points, and I agree wholehearedly with your point about mistaking correlation for causation, but like Kevin I disagree with the implication that we shouldn’t use these indications for analysis because there may be confounding factors.

The reality is that in many situations we don’t have access to these confounding variables – such as a independent measurement of engagement – so we look to other variables to stand in for them. Insteading of confounding, the variables are proxies for what we want to measure. To use your ice cream and murder example, if I want to predict murders, and I don’t have temperature records accessible I may then look to ice cream sales as a proxy variable to stand in for it. I could predict murder rates based on ice cream sales (although I’m not saying you should). An example I often use is the Google fluwatch system, where they use searches on flu symptoms to predict outbreaks. The flu itself is the cause, but all we have to measure is people’s behaviour in response.

Thanks for the post to get us talking!