RFM Too Crude For Fundraising
In a recent Agitator post I referred to the practice of using simple RFM in the fundraising segmentation process as ‘crude’. The equivalent of using an axe where a scalpel would be much more productive.
Many Agitator readers emailed me to ask “Why?” Here’s why.
I chose ‘crude’ to describe the way three important variables in predicting future behavior – recency, frequency and monetary amount – are applied in the vast majority of fundraising cases. Arbitrary because the cutoff points are usually based on myth (0 to 12 months for example in the case of ‘recency’), sometimes on experience, but seldom on math. This is really not a way for determining who gets targeted.
For the less math-inclined, conventional RFM practices are analogous to returning again and again to your favorite fishing hole and casting your line into the same place over and over hoping to catch a fish, regardless of other variables like the weather, the lure you’re using or the time of year. Sometimes you’ll hook one; sometimes not. An approach not recommended if you’re fishing for a living.
But, back to fundraising and the absolute necessity to put math to work along with the art of fundraising for best results.
When it comes to RFM, for example, 0 to 12 months is very common for defining an organization’s ‘active’ donor universe. What about a donor with more gifts and a higher HPC and average gift amount than those in this 0 to 12 range whose most recent gift was 13 months? They are excluded, despite 2 of the 3 variables that matter being ‘better’.
There are literally infinite numbers of these examples, and by extension, infinite cases where arbitrary cut-off points lead to inefficiencies akin to using an axe instead of scalpel for donor selection. Some people might even go so far as to call this ‘naïve’, or ‘stupid’ or worse. I don’t go that far, because by and large the myth of ‘RFM’ has been so pervasive that it’s seldom challenged and thus becomes part of ‘general principles’ fundraisers use.
So what is the alternative? For starters, even if one only uses the three RFM variables, there are fairly simple models (e.g. simple linear regression) that should be utilized to understand the relative importance of each (in math terms, their ‘weight’) in order to avoid having, by default, one variable be 100% deterministic and the other two not being used at all, as in my hypothetical example.
This way, every donor gets evaluated on the particulars of their past behavior with some making it into the selection because they really outperform on 1 or 2 variables, some because they are ‘good’ on all of them or various other combinations. But importantly, those who make the cut all fall into a math-based, acceptable range of likelihood to engage in the desired future behavior (e.g. responding, giving certain gift size) and each donor was given a fair shot at making it.
The far superior alternative is to expand this simple model to a more complicated one using other variables on the house file, plus external data, to have a more complete view of the donor.
No question that R, F and M matter … a lot. But to think that three arbitrary variables make a donor is, again, crude. One doesn’t need a lot of ‘exotic’ external data to improve on a 3-variable RFM model. Basic demographics (e.g., age, income) provide incremental insight into the likelihood of responsiveness and giving levels.
For example, consider an appeal where the goal is to get $100 dollars from donors who historically have given in the $10 to $25 range (which probably describes a large number of nonprofits). Including age and income in a model with RFM variables means the final model will, in essence, be using external knowledge about your donors to select among those whose internal behavior on the file (i.e. their RFM) suggests they are good targets, but whose life ‘status’, suggests otherwise. No matter how committed I am to an organization, if I don’t have the financial means to give $100 in one check (or even cumulatively over a year), then I shouldn’t be asked for it.
By simply including basic demographics, the select becomes more efficient. Extend this to a world that is technically and inexpensively within reach — where internal variables (RFM) are properly weighted, and hundreds of external variables or attributes are given a chance to define the final model (rather than arbitrarily selecting which ones to include) and where math is used to determine what matters and what doesn’t, and how to apply the former on an individual donor basis — we get to a sophisticated and very smart approach for achieving the most from our most valuable asset … the house file.
This is one of the reasons that over at our sister company, DonorTrends, we’ve created TrueTags — a sort of GPS model for planning and segmenting for all fundraising programs that enables fundraisers to improve bottom lines and improve long-term donor relationships by focusing on:
- Donors who are likely to respond to a certain type of solicitation;
- Donors who will respond best to upgrading;
- Donors who should not be solicited, or should not be solicited as frequently as other donors;
- Lapsed donors who you should cultivate for reactivation.
This particular series of predictive models uses over 1,500 external attributes and RFM to tag an entire donor file and then clusters the donors into actionable fundraising segments.
Regardless of what system you decide to use, my point in calling mere RFM ‘crude’ is simply this: intelligent, data-driven and math-based segmentation is no longer a ‘research luxury’ … it’s a necessity to meet bottom line goals today.
It’s time to abandon the axe and turn to the scalpel.
Roger
P.S. Every fundraiser should be regularly evaluating where potential exists at any phase of the donor lifecycle – from reactivating lapsed donors to securing legacy gifts.
You can learn more about this by visiting www.donortrends.com and clicking any level of the donor pyramid on the home page. I’ve helped create a free report called the Giving Analysis Profile (GAP) report that allows you to evaluate 4 different predictive models for free.
If you’re ready for your free GAP Report, simply click here, fill out the short form and email will be sent with secure file upload instructions.
AND, in the ‘Special Note’ section please add that you’re an Agitator reader and I’ll personally see that you get your results in 48 hours rather than the customary 7 days.
Great stuff, as always thanks Agitator. But it should come with a warning. Many charities I meet here in Australia, and smaller charities in USA and Canada haven’t even begun to segment their database by RFM. Those that have often don’t have the volume to justify much more than crude targeting, for which RFM, along source and transaction type, is usually fine.
No matter how complex your RFM, source and transaction type should always form part of your targeting. For example, two monthly donors giving $50 a month for the last nine months should be communicated and targeted differently depending on what triggered their initial gift. A face to face (direct dialogue) recruited monthly donor behaves very different to a mail recruited monthly donor even though their transactions look identical.
Whatever targeting is implemented, for all charities the Pareto principle should be applied and a more sophisticated targeting approach for their top donors implemented. Even a charity with just 5,000 donors would benefit from a more personalised approach to it’s top 100 or so donors.
At the top end, even the type of credit card should influence targeting. An AMEX donor who gave $1000 thirteen months ago is probably a better prospect than a VISA donor or cheque donor from eleven months ago.
But unless you have hundreds of thousands of donors at the bottom end, it really doesn’t matter. If you had to choose between a $50 AMEX donor from 13 months or a $50 Visa donor from 11 months ago – mail them both, or neither because it won’t really make much difference.
Apply the Pareto principle early on in targeting.
Sean Triner, unsurprisingly a Vilfredo Pareto fan.
Pareto Fundraising, Australia, NZ, UK, Hong Kong, US and Canada.