The donor scoring model
Each donor carries seven numbers plus a segment. This is what each one is, how it's computed, and how it changes over time.
Together's scoring model is rules-based RFM (Recency, Frequency, Monetary) extended with two propensity signals (upgrade, lapse) and a segment label. Scores are recomputed on every intelligence pass; you don't tune the model, but understanding it helps you trust the outputs.
The score fields
| Field | Type | Range | Direction | What it measures |
|---|---|---|---|---|
| Recency | integer | 1-5 | higher = better | How recently the donor gave. 5 = within the last quintile of donors by recency; 1 = oldest quintile. |
| Frequency | integer | 1-5 | higher = better | How often the donor has given. 5 = top quintile by gift count (or recurring active); 1 = bottom quintile. |
| Monetary | integer | 1-5 | higher = better | How much the donor has given in total. 5 = top quintile by lifetime value; 1 = bottom quintile. |
| RFM composite | integer | 3-15 | higher = better | Sum of the three 1-5 subscores. The headline number for sorting and ranking. |
| Upgrade signal | float | 0-1 | higher = stronger upgrade case | Propensity to give more if asked. Above 0.6 = strong candidate. Built from recency, recent gift trajectory, and engagement signals. |
| Lapse risk | float | 0-1 | higher = at greater risk | Propensity to stop giving. Above 0.5 = warning band. Built from time since last gift, churn baselines for the segment, and engagement signals. |
| Segment | string | one of | - | Behavioural label. Derived from the score fields. See the built-in segments below. |
Built-in segments
The labels follow a customer-lifecycle pattern adapted for donor behaviour. Each donor falls into exactly one segment per pass.
| Segment | Indicator | Typical action |
|---|---|---|
| Champion | High score on all three RFM dimensions. Strong upgrade signal often paired. | Thank them publicly; consider major-gift conversations. |
| Loyal | High Frequency, decent Recency, moderate Monetary. Steady giver. | Recurring upgrade ask. They're already engaged; raising the cadence or amount is the natural move. |
| Potential Loyalist | High Recency, low-to-moderate Frequency. New-ish donor showing engagement. | Cultivate. A second ask soon converts these to Loyal. |
| Promising | High Recency, low Frequency. First gift only. | Welcome series. Make the first follow-up ask low-friction. |
| New Donor | Brand new. First gift very recently, no history. | Onboarding sequence; thank-you + next-ask. |
| Needs Attention | Was previously engaged, recent gift has slowed. | Re-engagement campaign. They're sliding before they lapse. |
| About to Lapse | High Lapse risk on a previously high-engagement donor. | Retention triage. Direct outreach; understand why they're disengaging. |
| At Risk | Moderate score across the board, declining recency. | Reactivation. Light touch first; escalate only if they re-engage. |
| Lost High Value | Very low Recency with previously high Frequency or Monetary. | Specific major-gift retention play. These were big donors who walked away. |
| Lapsed | Low Recency, low Frequency. No recent activity. | Optional reactivation. Often noise; the win-back rate is low so prioritise other segments first. |
How scoring runs
The full pipeline runs once a day as a scheduled cron job that processes every org with donors. Individual donors are also rescored on the spot when a new donation lands, so a fresh gift is reflected in scores immediately even outside the daily window. Each full pass:
- Recomputes recency, frequency, and monetary quintiles across your full donor base.
- Re-projects each donor into the new quintiles.
- Recomputes upgrade and lapse propensities using the new quintiles.
- Re-assigns segments based on the new scores.
- Updates suggested ask amounts (separate from the scores, but recomputed alongside).
The pass is org-scoped; one org's score distribution does not influence another's.
When scores change
Scores change when:
- A donor gives a new donation (Recency moves to 5, Monetary updates, Frequency may bump).
- A donor's last donation falls outside the recency band as time passes (Recency decays even without new activity).
- The quintile boundaries themselves shift because your donor base changes shape (e.g., a big campaign brings in many new donors, shifting where the top 20% of recency sits).
Scores are not stable over long periods even without donor action. This is by design; the model is comparative, not absolute.
What scores are not
- Not a prediction of any single donor's behaviour. Propensities are population-level rules, not personal forecasts. Treat them as a triage tool, not an oracle.
- Not adjustable. The quintile boundaries are computed from your data; you don't set thresholds. If you want different segmentation logic, build a custom segment via the API.
- Not a substitute for relationship. A Champion who gets a generic ask doesn't stay a Champion. Scores tell you where to focus; the conversation itself is still yours.
What to do next
- Activate intelligence if you haven't yet: Activate intelligence.
- Read how the score decides ask amounts: How scoring decides who gets asked.