Methodology
The reasoning behind every number.
This is a walk from the file you hold today to the revenue it can still return. Each method is named, each is stated plainly, and the statistics are linked so you can learn the terms and check the work. The through-line is simple: more revenue from the file you already have.
Revision: July 2026
A donor file is a portfolio read with averages.
A donor file behaves like a portfolio. It holds positions that compound, positions that quietly close, and a small set of holdings that carry most of the return. Read with an annual average, none of that is visible.
An average retention rate cannot say which month a donor left. An average gift cannot say which donors are worth ten times the rest. An average year cannot separate a strong December from a shrinking base underneath it.
The cost of that blindness is not abstract. It is measured in gifts that stop without a flag, and in acquisition dollars spent on donors who were never going to stay. On a typical file, close to a quarter of active donors lapse in a year, most of them without a single warning in the reports a board actually reads.
22%of active donors lapse in a year
Lineage
Every number traces back to a gift.
Before the first model runs, three systems have to become one. Your CRM holds contacts and gifts. Google Analytics and your email platform hold how those donors arrived and what they opened. We land each source as it is, then reconcile all three into a single gift ledger without discarding an original field.
From that ledger, every named model reads the same rows, and every figure it returns can be followed back down the pipeline to the gift that produced it. This is the part a finance team asks about first. A number that reconciles to source is one a board can act on, and that counts for more than a number that only looks polished.
Schematic of the lineage every figure travels. Sample sources for a fictional organization.
1Survival analysisa,b
Why
An annual retention rate tells you how many donors stayed. It cannot tell you when the rest left, and the when is where the money is. Kaplan-Meier estimation reads each donor as a tenure with a status, still giving or lapsed, and returns the month-by-month probability that a donor is still with you. Because it handles right-censoring, donors who are simply still active never drag the curve down.
How
Confidence intervals use Greenwood's formula, so the band widens honestly as later cohorts thin: a 36-month estimate built on fewer surviving donors carries more uncertainty than a 6-month one, and the chart shows it. A companion hazard function view estimates the risk of lapse at each month of tenure, usually sharpest in the window right after a first gift.
What it's worth
The months where the hazard spikes are where a second-gift or win-back move returns the most. Retention spend stops being spread evenly across the calendar and starts landing on the weeks that decide whether a donor stays.
Illustrative — sample data for a fictional organization.
2Lifetime valuec
Why
An undiscounted lifetime value overstates what a donor is worth and misprices what you can spend to acquire one. We fit a Weibull survival tail to each giving tier and discount the projected stream back to today, so a $25-a-month donor and a $1,000-a-year donor are priced like the different assets they are.
How
Each fit reports its shape and scale parameters and an R-squared against the observed curve. The discounted value carries a sensitivity grid across discount rate and horizon, so the figure is never a single point divorced from the assumptions behind it.
What it's worth
Priced correctly, every tier earns a defensible acquisition budget and a defensible upgrade target. You stop overspending on the lowest tier and underspending on the mid-tier donors who sit one ask from the next band.
| Horizon | 6% | 9% | 12% |
|---|---|---|---|
| 3-year | $182 | $174 | $167 |
| 5-year | $241 | $224 | $209 |
| 7-year | $278 | $252 | $231 |
Illustrative — sample data for a fictional organization.
3Seasonal-trend decompositiond
Why
A strong December can hide a file that is shrinking the other eleven months. seasonal-trend decomposition splits a monthly revenue series into three parts, a repeating seasonal pattern, an underlying trend, and a remainder, so a growth claim survives scrutiny.
How
Reading the trend on its own answers the question a board keeps asking: set aside the year-end spike and the summer lull, is the file growing or shrinking? A rising seasonal peak sitting on a falling trend is a common and costly pattern, and the decomposition makes it plain.
What it's worth
You learn whether last year's growth was real or seasonal before you build next year's budget on it, and you protect the revenue a mistaken read would have committed twice.
Illustrative — sample data for a fictional organization.
4Concentration risk
Why
When the top decile of donors carries most of the revenue, one departure is a budget event, not a rounding error. A Lorenz curve and its Gini coefficient measure exactly how exposed the file is, using the same math economists use for income inequality.
How
Donors are ordered smallest to largest and the curve plots the cumulative share of revenue against the cumulative share of donors. The further it bows from the diagonal, the more revenue rides on a few names, and the Gini puts that gap in one number between zero and one.
What it's worth
The review sizes the departure of the top holdings in dollars, so a major-gift pipeline gets funded against a number instead of a hunch. On a concentrated file, more than half of revenue can ride on the top decile.
Illustrative — sample data for a fictional organization.
5Net dollar retentione
Why
Boards fund growth they can audit. Net dollar retention is the operating metric a SaaS finance team lives by, applied to a donor base: whether last year's donors gave more or less this year, before a single new donor is counted.
How
It extends dollar retention to include upgrades and downgrades among the donors you kept, and excludes acquisition entirely. A file can add donors while its net dollar retention sits below 100%, which means the base is eroding under the growth.
What it's worth
One number tells the board whether the existing base is compounding or leaking, and where a dollar of retention work returns more than a dollar of acquisition.
Illustrative — sample data for a fictional organization.
6Acquisition economics
Why
A channel that looks cheap on the first gift is often the most expensive on two-year value. Reading Google Analytics and the email platform next to the gift file prices every channel, campaign, promotion, and appeal by what its donors became worth, not what they first gave. The finance term is customer acquisition cost; the honest version reads it against a cohort's two-year value.
How
Each acquisition source carries a true cost to acquire and the discounted value of the donors it brought, tracked as a cohort across the following two years rather than judged on the first gift.
What it's worth
You move budget toward the channels whose donors stay and give, and away from the ones that only looked efficient at signup. The best channel can return several times what the cheapest-looking one does.
| Channel | CAC | 24-mo value | Return |
|---|---|---|---|
| $18 | $142 | 7.9× | |
| Organic search | $24 | $118 | 4.9× |
| Referral | $31 | $126 | 4.1× |
| Paid social | $52 | $96 | 1.8× |
| Paid search | $68 | $88 | 1.3× |
Illustrative — sample data for a fictional organization.
7Revenue bridge
Why
A board that can reconcile opening revenue to closing revenue, to the dollar, trusts the plan built on it. The revenue bridge is an accounting identity, not a model. It attributes the full year-over-year change to named components, new, reactivated, upgraded, downgraded, and lapsed, and the pieces sum exactly to the difference. It reads like the waterfall a CFO already knows.
How
Because it reconciles, it is auditable. If the components do not add up, the analysis has an error, and that constraint is the point. Every dollar of change has to be explained rather than waved at.
What it's worth
The board sees precisely where last year's revenue was won and lost, and next year's plan is built on a number that ties out.
Illustrative — sample data for a fictional organization.
8AI, with provenance
The language model sits on top of the analysis, not inside it. Every figure, from the survival estimate to the lifetime value to the revenue-bridge components, is computed by the named methods above. The model calculates none of them. It reads the finished tables, states them in plain language, and drafts the narrative a strategist would otherwise write by hand.
Every answer carries the query and the source tables behind it. When the analysis reports which donors slipped last quarter, it shows the query that produced the list and the tables it read, so the figure can be checked against source rather than taken on the model's word.
The limits are stated as plainly as the methods. The model summarizes and drafts. It does not invent figures, and it does not decide what an organization should do. A person reviews the work before it reaches you, and the rule that governs the rest of this page governs the model too: no number appears that cannot be reproduced from source.
The close
What you leave with.
Each method here ends in a move you can price and sequence, in order of what it returns against what it costs to run.
- 01Find the months your donors leave, and put the second-gift and win-back work there.
- 02Price every donor and every channel on two-year value, not the first gift.
- 03Separate real growth from a good December before you budget on it.
- 04Name the concentration risk in dollars, and widen the base that carries it.
- 05Hand the board a revenue bridge that reconciles to the dollar.
It adds up to one thing: more revenue from the file you already have. A first pass usually surfaces close to $0.5M in recoverable revenue a file was quietly losing, and the plan to keep it comes with the query behind every number.
Notes
- aKaplan, E. L. and Meier, P. (1958). Nonparametric estimation from incomplete observations. Journal of the American Statistical Association, 53(282).
- bGreenwood, M. (1926). The natural duration of cancer. Reports on Public Health and Medical Subjects, 33. Basis for the survival-variance estimate.
- cWeibull, W. (1951). A statistical distribution function of wide applicability. Journal of Applied Mechanics, 18(3).
- dCleveland, R. B., Cleveland, W. S., McRae, J. E., and Terpenning, I. (1990). STL: A seasonal-trend decomposition procedure based on loess. Journal of Official Statistics, 6(1).
- eRetention definitions follow the Fundraising Effectiveness Project conventions, with lapse windows stated per file.
The same rigor, run on your file.
Every method here runs on the two files you already export, with its assumptions stated and its numbers traceable to source.