RFM Analysis for Nonprofits: A Practical Guide

RFM is the oldest useful idea in donor analysis and it still outperforms most of what has replaced it. Three numbers per donor: how recently they gave, how often, and how much. Score each, combine them, and you have a ranked file without buying anything.

It comes from catalogue retail, which matters both because it is well proven and because nonprofit giving is not shopping. The adaptations are covered below, and they are where most implementations go wrong.

The three variables

Recency dominating is the finding people find counter-intuitive and it holds consistently. Someone who gave $50 last month is a better prospect for the next appeal than someone who gave $500 four years ago.

If you would rather not build the spreadsheet, our free RFM scoring tool does exactly this in your browser: paste a CSV export, get every donor scored and segmented, and download the result. Your file is never uploaded anywhere, which matters more here than in most places.

How to calculate it

You need three columns per donor: last gift date, number of gifts in the period, and total given in the period. Most CRMs export this directly. Use a fixed window, usually five years, so the analysis is repeatable.

Sort the file by each variable in turn and split into five equal groups, scoring 5 for the best fifth down to 1 for the worst. Quintiles are the standard because they divide your actual file rather than imposing thresholds that may not fit it.

Each donor ends with a three digit code such as 555 or 154. Do not add the digits together. A 5-1-1 and a 1-1-5 both sum to 7 and mean completely different things, and collapsing them into one number throws away the entire value of the exercise.

What the codes mean and what to do

The 5-1 group and the lapsing loyalists are where the money is. First year donors retain at roughly 20% to 25% against a sector average around 43%, so converting a first gift into a second changes a donor’s value more than any other single intervention.

The x-5-1 group is the one most often ignored. Frequent small donors are disproportionately represented in bequests, because the behaviour that predicts a legacy is loyalty rather than gift size.

Where RFM misleads in a nonprofit context

It was built for repeat purchasing and charitable giving differs in ways that matter.

Major donors break the model. Someone giving $50,000 every three years by arrangement scores terribly on recency and frequency. Run major donors outside the RFM file entirely, or you will suppress the people funding you.

Recurring donors distort frequency. Twelve monthly gifts of $10 is not more engaged than one annual gift of $120, but the arithmetic says so. Count a recurring commitment as a single relationship, or analyse recurring donors separately.

Restricted and memorial gifts are not preferences. A gift to a memorial fund says something about a bereavement, not about propensity to support your general work.

It is blind to everything qualitative. RFM cannot see that a donor complained, that their circumstances changed, or that they told someone they were reviewing their will. All of that sits in notes and email, and it is frequently more predictive than the numbers.

Running it in practice

Quarterly is the sensible cadence for most organizations, and before each major appeal. Monthly recalculation produces movement that is noise rather than signal.

Give it an owner. RFM is not difficult, and the reason it lapses in most organizations is that it was run once by whoever was interested, saved to a desktop, and never repeated. A named person, a diarised date and a file everyone can find is the whole of what makes it stick.

Keep the history. The valuable output is not this quarter’s scores, it is the direction of travel: a donor moving from 555 to 355 over two quarters is the alert. A static snapshot cannot show you that, and it is the single biggest improvement most organizations can make to how they use RFM.

Beyond that, use it to decide effort rather than only to decide mailing lists. The point of a ranked file is that the top of it gets a person and the bottom of it gets an email.

Two extensions worth adding

Once the basic model is running, two additions give most of the benefit that more complex approaches promise.

Add tenure. Alongside the three scores, record the year of the first gift. It separates a 555 who has given for eleven years from a 555 who started in March, and they are entirely different propositions. The long standing donor is a stewardship and legacy prospect. The new one is a retention risk until a second and third gift are in.

Weight the variables deliberately. The standard model treats recency, frequency and monetary value equally, and the evidence says recency predicts best. If you are selecting for an appeal, sorting primarily on recency and using the other two to break ties usually outperforms treating the three digit code as a flat ranking.

RFM segments the file. What it does not do is tell you whether fundraising overall is improving, which needs a small set of tracked numbers reported the same way each quarter. See fundraising metrics worth tracking, where retention rather than total raised is the headline.

One group RFM systematically under-serves is your best legacy prospects, because loyalty rather than gift size predicts a bequest: see planned giving.

Where software goes further

RFM is rules you wrote. Predictive scoring learns patterns across the whole file, and can weigh dozens of variables including channel, appeal type, seasonality and demographics. On a large file it outperforms RFM, which is why Dataro and similar products exist. It also costs from $15,000 a year, so the gain has to be worth that.

The other direction is the qualitative gap. Tools that read documents, notes and email alongside the CRM can surface the complaint or the life event behind a falling score, which is the thing RFM structurally cannot see. Gratefully works this way at $4,800 a year for five seats.

Neither is a starting point. RFM in a spreadsheet costs an afternoon, captures most of the available value, and tells you whether the more expensive versions would have anything to add. We cover what those cost and who they suit in AI tools for nonprofit fundraising, and the practical warning signs in how to tell a donor is about to lapse.

Questions people ask

What is RFM analysis in fundraising?

A method for ranking donors on three numbers: recency, how long since their last gift; frequency, how many gifts they have made; and monetary value, how much they have given.

Each donor is scored 1 to 5 on each variable by splitting your file into fifths, producing a three digit code such as 555 or 154. The code tells you what kind of donor someone is and what to do about them.

It comes from catalogue retail, it is decades old, and it still outperforms most of what has replaced it because the underlying insight holds: recent behaviour predicts future behaviour better than anything else you can easily measure.

How do you calculate an RFM score?

Export three columns per donor: last gift date, number of gifts in a fixed window, and total given in that window. Five years is the usual window.

Sort by each variable in turn, split the file into five equal groups, and score 5 for the best fifth down to 1 for the worst. Combine into a three digit code.

Use quintiles of your own file rather than fixed thresholds, so the analysis fits your actual donor base. And do not add the digits together: 511 and 115 both sum to 7 and describe completely different donors.

Which matters most in RFM, recency or amount?

Recency, by a considerable margin, and it surprises people every time. A donor who gave $50 last month is a better prospect for your next appeal than one who gave $500 four years ago.

Frequency comes second, because it distinguishes a habit from a one off, and habit is what compounds into long term value.

Monetary value is the weakest predictor of whether someone will give again, though it is the strongest predictor of how much you receive when they do. That is why it belongs in the model and why it should not lead it.

Does RFM work for major donors?

Poorly, and applying it to them causes real damage. A donor giving $50,000 every three years by arrangement scores badly on recency and frequency, and a mechanical application of RFM would suppress them from your file.

Run major donors outside the RFM analysis entirely and manage them through moves management with named relationship owners. RFM is a tool for the broad file, where individual attention is not possible.

Set your threshold deliberately and write it down, so the exclusion is a policy rather than something that depends on whoever runs the export.

How often should you run RFM analysis?

Quarterly for most organizations, plus a run before each major appeal. Monthly recalculation produces movement that is noise rather than signal.

Keep every run rather than overwriting. The valuable output is direction of travel, not the current snapshot. A donor sliding from 555 to 355 across two quarters is an alert you can act on, and no single snapshot will ever show you that.

Review your window and thresholds annually, since giving patterns shift and a rule written three years ago may be sorting people incorrectly.

What are the limits of RFM for nonprofits?

It was designed for repeat purchasing, and charitable giving differs in four ways that matter.

Major donors break it, because irregular large gifts score badly. Recurring donors distort frequency, since twelve monthly gifts is not twelve times the engagement of one annual gift. Memorial and restricted gifts describe a circumstance rather than a preference.

Most importantly it is blind to everything qualitative. RFM cannot see that a donor complained, that their circumstances changed, or that they mentioned their will to a volunteer. That information usually exists somewhere in your notes and email, and it is frequently more predictive than the three numbers.

Is RFM better than AI donor scoring?

For most organizations it is the right place to start, and for many it is enough. RFM costs an afternoon in a spreadsheet and captures a large share of the available value.

Predictive scoring outperforms it on large files, because it learns patterns across dozens of variables rather than applying three rules you chose. That advantage is real and it starts at around $15,000 a year, so it needs a file big enough for the improvement to pay for itself.

Run RFM first regardless. It tells you whether your file has enough structure to be worth modelling, and organizations that skip it often buy prediction to learn things a spreadsheet already knew.

What should we do with each RFM segment?

Four groups deserve most of your attention.

Your 555 group gets personal stewardship rather than more mail. New donors scoring 5 on recency and 1 on frequency get a second gift conversion effort, which is the highest value action available given first year retention around 20% to 25%.

Lapsing loyalists, high frequency and value with poor recency, get a personal contact and a question rather than an appeal. Something changed and you want to know what.

Frequent small givers are upgrade candidates and disproportionately your best legacy prospects, because loyalty predicts bequests better than gift size does. Long lapsed single small donors get suppressed from expensive mail.

This is reference information, not legal or tax advice. Rules vary by state and change over time. For a decision that carries real consequences, check the current text at irs.gov or your state registry, and talk to a nonprofit attorney or CPA.