RFM Segmentation: First Refuge of the Scoundrel
“Wait!”, I hear you cry. “You rail against segmentations that aren’t predictive. But transactional RFM segmentation is not a bad predictor.”
And I will stipulate that:
- A person will tend to give the amounts that they have given in the past.
- A current donor is more likely to give than a lapsed donor.
- A donor who has given multiple gifts is more likely to give than someone who has given only one gift.
So, by all means, include transactional information when you set your ask string and when you decide whether to send a communication or not.
BUT… somewhere along the line, we confused transactional analysis as merely useful and elevated it to a position of being “primary”. Sometimes, we even confuse being primary with being the only.
This is wrong.
Transactional analysis should absolutely not be the only segmentation you do. Neither should it even be the first segmentation you do.
This type of transactional analysis focuses on the wrong question: “what people does my organization want to receive this communication?” Rather, the first question you should be asking about a communication is “what people will want to get this communication?” or “how can I make this communication something these people want to get?”
This order is important. You want to have the variables that make the most difference at the beginning of your decision tree so you can create different versions or include/exclude audiences based on the most important information.
Let’s take a trivially simple example. The most important factor in determining if someone will respond to a communication is whether they are alive. If they are, they can respond. If they aren’t, they can’t.
If you were segmenting based on RFM + are they dead, it would make no sense to create 64 different variants of RFM, then apply the “are they dead” filter to every one of those 64 segments. Rather, you can first remove the dearly departed, then do the rest of the segmentation.
There are a few of these gatekeeper questions (e.g., “did they opt out of this channel? No”, “are they on the seed list? Yes”) that are the most important things. They come first.
What comes next? What makes the most difference in both who will get a communication and what it contains? It’s not RFM segmentation.
Yes? You in the back of the room? Channel?
Yes, that’s better. If you are sending a mail piece, you are going to send it to a far greater proportion of mail donors than of online donors. Channel is usually very predictive of response and it is a tactic-based answer to “which people will want to get this communication?”
But it doesn’t address why someone gives to an organization. This is the central point on which all messaging pivots (and thus segmentation should pivot). Until you segment by donor identity, you will always have more difference among people in the segmentation bucket than between buckets.
We said that channel beats RFM in predictive power; how does donor identity do versus channel?
Here’s an example from a health charity (some numbers tweaked for anonymity):
| Channel | Event giving | Direct marketing giving | Total giving | Commitment score (out of 10) |
| 10% | 90% | $400 | 7.6 | |
| Event | 80% | 20% | $325 | 7.8 |
| Digital | 5% | 95% | $275 | 7.9 |
Pretty thin gruel here. What you might learn is that mail donors are a bit more valuable and that people tend to stay with their own channel. But when writing the appeal, you have no idea what to say for the mail group that is different from digital.
Now let’s look at this same audience. This time let’s view it from the perspective of whether they were a direct beneficiary of services, indirect beneficiary, or no connection:
| Identity | Event giving | Direct marketing giving | Total giving | Commitment score (out of 10) |
| Direct | 48% | 52% | $500 | 8.6 |
| Indirect | 34% | 66% | $400 | 7.9 |
| No connection | 30% | 70% | $250 | 7.1 |
Now we have some insights that are actionable. Direct beneficiaries are the most valuable and most committed by a significant margin — there’s far more differentiation here than by channel. They are also more likely to attend your events. Each of these donors has a mailbox, email box, and phone number and it’s just a matter of their preference by which they give.
In fact, when we dug into the details behind these groups, we found that by far the most important factor (60%) in creating the donor’s relationship with the organization was patient care services. If they hadn’t been or loved a patient, they didn’t care about patient care services at all.
With identity segmentation you know which group is most valuable and responsive. You also know what to say to them. Then… if you do RFM segmentation after segmentation-by-identity, you can go deeper into the file of those with a direct connection than those with no connection (with entirely different messaging).
This is segmentation at its best – a segmentation focused on the recipient.
Nick
9 responses to “RFM Segmentation: First Refuge of the Scoundrel”
Behavioral Science Q & A
Integrating an individual giving appeal with other communications from a charity can have both positive and negative effects, and the outcome largely depends on how it’s executed. Advantages of Integration Brand Consistency: Maintaining a consistent appearance and messaging across all communications can reinforce the org’s brand identity and strengthen brand recognition and trust among your […]
Read Full Answer
I’m not aware of any in-market tests specifically comparing recurring vs. gift frequency language. I suspect the answer might not be the same with all gift frequencies, nor with all people. It sounds like a great opportunity for you to test and find out what works for your audience. Based on the literature, here’s a couple […]
Read Full Answer
Based on what we know from existing data, those renewal notices can actually be pretty effective in getting people to donate. They tap into our psychology – creating a sense of urgency, reminding us of past support, and using personalization to make the message hit home. They’re playing on our natural tendencies to feel obligated […]
Read Full Answer
Interesting question. I had a quick look at the testing done on this topic. On the positive side, in all cases, over half of donors decide to cover the fee. In some cases, it goes as high as 65%. Not a negligible percentage at all. Here’s another test from iRaiser showing consistent results (see point […]
Read Full Answer
There’s just one thing to consider when designing a supporter journey: the supporter. More specifically, you need to take into account: Who the supporter is i.e. their identity, which is the reason they support this cause, and their personality, which describes the way they “see” and process the world. These will determine the kind of […]
Read Full Answer
I’m not an expert in this but a quick search surfaced this article on the effect of tax reforms on 2019’s charitable giving. The researchers didn’t find a reduction. Actually, they observed an “increase in charitable contributions in 2019, even with the lower tax rates and the dramatically smaller number of taxpayers who itemize their […]
Read Full Answer
One thing I would like to add here. While I know you are not suggesting to throw the baby out with the bath water. I wanted to stick up for transaction based predictive models. Predictive models based on transactions may not be able to capture motivation, the data is available for every donor and can do a good job of identifying who is likely to respond. To re-frame it instead of identifying people “we” want to contact, I think of predictive models as identifying “donors” who are more likely to want to hear from you.
Of course what you say to them can not be shaped by this type of prediction, and here I think identity and motivation need to play a much larger role, which perhaps is your larger point here.
I was actually thinking about a separate blog post around this topic and the different types of segmentation, but you’ve put it succinctly.
In my mind, there are really two different types of segmentation: segmentation for customization and in/out segmentation.
For me, transactional modeling is almost exclusively in/out segmentation. While there are customizations that can be done that are valuable from transactions (e.g., putting the date of last gift can increase response rates in lapsed packages), there are generally few opportunities and minor customizations. Rather, its usefulness in determining who gets the communication.
With an identity/motivation segmentation, you may be creating different sentences, paragraphs, or pieces, depending on how dramatically different they are. I’m a huge fan of predictive models (especially beyond RFM, which is frequently capricious). My argument is that it makes more sense to do the customization segmentation first, then transactional modeling.
For example, if you are a disease nonprofit, there’s a good chance that someone has the disease themselves, they are worth 2X a donor who doesn’t in lifetime value (and likely in terms of likeliness to respond). If I were working with DonorTrends (hypothetically :-)) on modeling that donor files, I’d give it to you in two segments, with the full intent of 1) having different packages for the different groups and 2) communicating deeper into those with the disease than those without.
Well said and look forward to reading more as you dive into this.
Perhaps a more interesting qualitative research finding to have to support or refute this entire dialogue would would be how many donors/constituents are actually reading the content being delivered to them in part or whole? And if they even detect subtle differences in copy or messaging.
If the donors are simply skimming content, or less than skimming by reacting just to brand affinity or images, is all this proposed focus on customization/relevance only ever going to impact the most narrow of populations on the donor file who are actually invested enough to read the content top to bottom?
Much like your alive or dead example, without knowing if donors are truly scrutinizing the content they are receiving, it is hard to place a value on the necessity of customization of message beyond the slim few qualitative respondents you have to create the commitment score.
In which case, determining who is going to respond in a very binary in/out way, would still seem more critically important for the masses than worrying about if we deliver the best custom message to the slim few who are looking for it.
I used to wonder about that myself. Thought it was a bunch of mumbo-jumbo. Donor identities holding sway, people reading the detail of our letters. Crazy thing is… it’s true. Identity, commitment, complex processing… All of it… It’s all true.
(Sorry, couldn’t not do a Han Solo quote with your commenter name :-). That said, please use real names in comments. You raise an excellent point that is only improved by allowing for a real exchange of ideas between real people. Here, we disagree with passion and verve, knowing that steel sharpens steel.
Plus, we don’t want the Russians/Sith coming in and hijacking the comments.)
This question – do people read our stuff – goes to the heart of the value of customization. If they don’t, we shouldn’t bother beyond getting their name right. If they do, then customization is a vital part of our direct marketing strategy.
Fortunately, the research is in that people do read and have their decision influenced at a nuanced level. Not all of these are about identity (as too few organizations have taken the plunge), but some are. And all show that people are reading and picking up on nuance. Annual Fund instead of Annual Fund increased response rate by 30%: https://site.stanford.edu/sites/default/files/kesslermilkman_identityincharitablegiving.pdf
– ASCPA asks for cat v dog preferences and changes subject lines and content as a result. Response rates went up 230%: http://www.theagitator.net/online-fundraising/what-to-listen-for-in-donor-onboarding/
– Another organization looked at their equivalent of cats v dogs and in one phone call, both asking the identity, then playing it back, they increased response rate by 15% and average gift by 15%.
– Framing the act of helping as a noun identity (“be a helper”) rather than a verb (“to help”) increases people giving by 29%.
– More on Roger’s piece from yesterday: allowing for donor preference on the response device increases response rates: http://www.kenburnett.com/Blog64thedonorschoice.html
– Red Cross added date of last gift to their lapsed packages and had a 20% increase in response from five words: https://directtodonor.com/2016/02/17/priming-with-donation-history-and-localization/
– Adding an additional to an identifiable victim drains the effectiveness of a mail piece: https://www.cmu.edu/dietrich/sds/docs/loewenstein/SympathyCallous.pdf
– Similarly, putting in program effectiveness info – a sentence or two – helps response rate for $100+ donors and hurts it for under $100: https://directtodonor.com/2015/12/31/education-versus-emotion-in-direct-marketing-appeals/
– Saying
– You can increase the amount raised for butterflies if they are flying together, instead of randomly. (Really; not making this up: https://directtodonor.com/2016/02/04/scope-insensitivity-and-direct-marketing-why-one-beats-many/ )
– A simple change like whether you present the same symbolic gift or different symbolic gifts on the reply device can increase revenue by 42%. http://www.thedonorvoice.com/what-your-donation-buys-how-to-get-a-43-lift-in-revenue/
– Slight changes in match/lead gift wording can double/half response rate: https://directtodonor.com/2016/02/19/the-power-of-a-lead-gift/
These are simple changes: one word, one sentence, one concept. All with big impacts.
And if narratives help, Katrina Van Huss’s account from the perspective of a donor with a clear donor identity is a good one: http://www.nonprofitpro.com/post/care-cause-dont-donate/all/
Also, I’ll end this with a bold claim. In some cases, even if you didn’t customize at all based on identity (and you should), you should still have it at the front of the line. Take the health charity in the piece. If you only segmented based on one piece of information, would it be that people like this have a lifetime value twice this other group? Or would it be this group last gave us a gift 11 months ago versus this other that gave 13 months ago? I’d go for the former every time – not only is it predictive of what they will do in the appeal, but it is also about which donor do you want to retain more. I’d also go for commitment level over RFM analysis, for reasons articulated here: http://www.theagitator.net/research/good-enough-is-no-longer-good-enough-part-2-commitment/
Great article. And it certainly makes a case for integrated databases throughout a nonprofit organization. One of the reasons I find segmentation to be challenging for small and medium-sized nonprofits is that the development database is totally separate from the client database, and even the volunteer database. So while it may be possible to segment by cat and dog fans based on earmarked giving, it’s not so easy to segment by other identity factors. Just a sad reality.
Funny enough, I just posted about this on the DonorVoice blog at http://www.thedonorvoice.com/are-your-donors-hardcover-or-paperback/. It’s not just small nonprofits, but big nonprofits and big corporations have this same issue. In the above piece, I talk about how when Amazon got into kitchen implements in 1999, they were using the same database as books, so they had to classify every knife as hardcover or paperback. I’ve worked to consolidate 20 different databases, each with their own peculiar bit of info.
I argue, though, that working with the separate databases before they become the One Truth database is often necessary. When the database project comes around, you’ll know what data you need where and how to map databases to each other, rather than being at the mercy of technical requirements. Fortunately, with the Excel VLOOKUP function, you can approximate the SQL JOIN statement and marry two databases even if it’s just from their CSV outputs.
Nick, what’s the source on this one:
– Framing the act of helping as a noun identity (“be a helper”) rather than a verb (“to help”) increases people giving by 29%.
Sorry – made a mistake on that one; it should be increasing people “helping” not “giving”. The reason the distinction is important is that the study was of noun v verb identities among children, so it was helping, not giving, behavior. Full study at https://bingschool.stanford.edu/sites/default/files/publications/walton_master_helping.pdf