Key points

  • RFM sorts donors into segments using recency, frequency and monetary value.
  • It treats every donor in a segment the same, and it only looks backward.
  • Replacing RFM renewal selects with Squark models lifted one program's net revenue 20%.

What RFM is

RFM stands for recency, frequency and monetary value: how recently a donor gave, how often, and how much. Fundraisers have used it for decades to decide who gets the next appeal. It is simple, and anyone can see how it works.

What it misses

  • Everyone in a segment gets the same treatment. Two donors in the same cell can have very different odds of responding.
  • It looks backward. RFM describes what a donor did. It does not estimate what they will do.
  • It uses three fields. The donor record holds many more: channel, package, timing, engagement and history with each campaign.

Side by side

RFM segmentation and predictive AI compared
RFM segmentation Predictive AI
InputsThree: recency, frequency, monetary valueFar more of the donor record
UnitA segmentEach donor
DirectionBackward: what donors didForward: what each donor is likely to do
DecidesWho is in the selectWho, the ask, the package, the channel and the moment
Where results goA list pullAny system: CRM, email, print vendor, ad platforms or the warehouse

What happened when programs replaced it

One direct mail renewal program replaced its RFM-based selects with Squark models that scored each donor on likelihood to respond, expected return and expected lifetime value. Net revenue rose 20% and cost per dollar raised fell 6%. Read the case study.

A large membership organization had been giving every lapsed donor the same treatment. Scoring each one for reactivation cut the cost to reactivate by 10%. Read the case study.

Keep RFM as the control

RFM is a good baseline. For your next renewal mailing, send the RFM select as the control and the model’s select as the test. Compare net revenue and cost per dollar raised, and let the result decide.

See the lapsed donor playbook and ask amount optimization.