Donor Engine · 02
A score that shows its working, so you can disagree with it.
- Signals
- Five, behavioural
- Refresh
- Overnight, with trend history kept
- Flag fires
- On the third consecutive fall
What feeds the score
Five signals, shipped with these default weights.
You can tune the weights per organisation. A hospice with quarterly appeals and a wildlife trust with two field events a year should not be scored identically.
- 0126%
Giving frequency, against their own rhythm
A donor who has given every March for nine years is not lapsing in January. The gap is measured against the pattern that supporter has established, not against a fixed number of days.
- 0222%
Gift-size direction
The last four gifts read as a direction: rising, flat, or shrinking. A donor giving smaller amounts more often and a donor giving less overall look identical on a total, and completely different here.
- 0319%
Appeal response
Opened, clicked, gave, or silent, across the last few appeals. Silence on three consecutive appeals moves the score well before the giving pattern breaks.
- 0418%
Event and field presence
Attendance, and repeat attendance in particular. Supporters who have stood in the programme they fund behave differently afterwards, and stopping is an early signal.
- 0515%
Volunteer activity
Shifts claimed, hours logged, and cancellations. For supporters who give time as well as money, a quiet quarter of shifts often arrives before a quiet quarter of gifts.
Reading the number
Four bands, and each one implies a different move.
The most common mistake is treating the score as a ranking of who deserves attention. It is a ranking of who is about to change.
Deeply engaged. Usually the right group for a cultivation conversation, a named gift, or an invitation to something with your leadership in the room.
Giving and responding. The group most often left alone because nothing looks wrong, and the group where a small deepening step compounds.
Present but not deepening. Often long-standing supporters on autopilot. Worth a reason to care again rather than another ask.
Disengaging now. Surfaces with the triggering signal and a suggested step, ranked so the shortest, most recoverable relationships come first.
Lapse-risk detection
The flag fires on the trend, not on the anniversary.
Waiting for a missed gift is waiting for a decision that has already been made. Three consecutive falling reads is the trigger, and the reason travels with the flag.
- What triggers a flag
- Three consecutive falling score reads, or a single fall greater than 18 points with a signal attached to it.
- What suppresses a flag
- A known payment failure, a recorded bereavement or financial-strain note, and a deliberate frequency preference such as appeal-only. None of these should read as drift.
- How flags are ranked
- By recoverability first: length of relationship, size of the historical pattern, and how recently the fall began. A supporter three months into a decline is ahead of one who left two years ago.
- What arrives with the flag
- The signal that moved, the last four interactions, any unkept promise on the record, and a suggested next step with a named owner.
Eight reads, one supporter
The flag fires at read six, roughly seventy days before the gift that would have been missed. The three reads before it are the window where a phone call still reads as care rather than salvage.
Stated plainly
What the score cannot see.
We would rather you knew the limits than trusted the number too far.
Anything you did not record
A conversation at a wedding, a promise made on a phone call that was never logged, a WhatsApp thread on a personal phone. The score reads your data, and only your data.
Why a person went quiet
Illness, a job loss, a disagreement with a programme decision, or simply moving cities. The score sees the shape of the change, not the cause of it.
Whether the ask is appropriate
A high score on a supporter who has just been bereaved is still a bad time to ask. That call is yours, which is why sensitivities sit on the record.
On accuracy
Retention prediction is a young field and any vendor quoting a single accuracy figure is quoting it from someone else's donor file. What we will commit to is this: the score is reproducible, the signals behind it are visible, the weights are yours to change, and every flag can be traced to the reading that produced it. As we accumulate outcome data across organisations that agree to contribute it, we will publish what the flags actually predicted rather than what we hope they predict.
Three falling reads is a conversation. A missed gift is an apology.
The window between them is where retention work actually happens.