The segmentation prompt that found $40k in lapsed gifts

A donor segmentation prompt is a single, well-built instruction that turns a messy export of giving history into a ranked list of who to call first. In one anonymized case, a small development team used one to surface a quiet pile of lapsed mid-level donors and ran a focused reach-out that brought back roughly $40k they had written off.

Nothing in that story required new data. The donors were already in the database, the gifts were already recorded, and the patterns were already there. What changed was the time it took to see them. Reading a thousand-row export by hand is a week nobody has. Asking the right question of an AI Teammate is an afternoon.

What problem was hiding in the data

Picture a development team of two at a mid-size org. Their CRM held years of giving history, and like most orgs, they spent their attention on the loud signals: active monthly donors, the recent year-end surge, the major gift in the pipeline. The quiet middle never got a second look.

Buried in that middle was a familiar pattern. Donors who used to give two or three times a year, reliably, in the $250 to $1,000 range, and then stopped. Not angry, not gone, just lapsed. They had aged out of every report because no single gift was big enough to flag and no recent activity pulled them forward. Collectively, though, they represented real money sitting one good ask away from coming back.

The team knew these donors existed in the abstract. What they did not have was a fast way to pull them out of the noise, rank them, and decide who to contact this week. That gap is exactly what a donor segmentation prompt closes.

The segmentation prompt that did the work

Here is the kind of prompt that turns a raw export into a working call list. Paste your de-identified giving data, or work inside a tool that already holds it, and adjust the thresholds to your org.

You are a development analyst helping a nonprofit find lapsed donors worth re-engaging.

I will give you a table of donor giving history with these columns:
donor_id, first_gift_date, last_gift_date, total_gifts, lifetime_giving,
largest_gift, average_gift, gift_frequency_per_year.

Find donors who match a "lapsed but recoverable" pattern:
- Gave at least 3 separate gifts historically (a real habit, not a one-off)
- Average gift between $250 and $1,000
- No gift in the last 14 to 30 months (lapsed, but not ancient)
- Previously gave at least once a year (they had a rhythm we broke)

Return a ranked table, most worth contacting first. Rank by a simple
recoverability score that rewards higher lifetime giving, more frequent
past giving, and a more recent lapse. For each donor include:
donor_id, lifetime_giving, average_gift, months_since_last_gift,
recoverability_score, and a one-line reason they are worth a personal ask.

Then summarize: how many donors matched, their combined lifetime giving,
and the realistic re-engagement opportunity if even a quarter of them give again.

The prompt does three jobs at once. It defines the pattern in plain terms a human would recognize, it ranks the results so the team knows where to start, and it sizes the opportunity so the effort has a number attached. That last line is what turned a vague hunch into a Monday-morning plan.

The donors were never lost. They were just unsorted.

From list to $40k

A ranked list is a beginning, not a result. The reason the reach-out worked is that the team did not blast the segment with a generic appeal. They treated the list as a starting point for real conversations.

The recoverability score told them who to call first. The one-line reason gave each conversation a true hook, a reminder of what this person used to support and why now was a natural moment to come back. For the donors worth a letter rather than a call, the team used the same context to write personal notes fast, the way a year-end appeal that still sounds like you gets built: human judgment up front, AI handling the draft.

What used to be a research project nobody had time for became a focused week of asking. The roughly $40k that came back was money the org had already earned years earlier and quietly stopped pursuing. The segmentation prompt did not create those donors. It just made them visible in time to act.

How to adapt the prompt to your own data

The thresholds in that prompt are a starting point, not gospel. The lapse window of 14 to 30 months suits an org with an annual giving rhythm. If your donors give monthly, a three-month gap already signals trouble and your window should be tighter. If you run a single big campaign a year, widen it. The gift range matters too. A $250 to $1,000 band finds mid-level donors, but the same logic surfaces lapsed major donors if you raise the floor, or lapsed grassroots givers if you lower it.

The point is to describe a pattern you would recognize by hand, then let the model do the sorting at a speed you cannot. Run it once with your numbers, look at who surfaces, and adjust. If the list includes people you know are gone for good, tighten the recency rule. If it misses someone you expected, loosen a threshold. Two or three passes and the prompt is tuned to how your donors actually behave, which is the version worth saving and rerunning.

Why this counts as a system, not a one-off

The temptation after a win like this is to file the prompt and move on. The better move is to keep it. Run the same segmentation on a schedule, quarterly or twice a year, and lapsed-donor recovery stops being a heroic one-time dig and becomes a habit your org maintains. Prompt libraries were 2024. A recurring segmentation that an AI Teammate runs against fresh data is closer to where fundraising is headed.

That is the quiet lesson in this story. The afternoon that found $40k was not luck and it was not a special data set. It was one good question, asked of data the org already had, by a team that decided to look.

If you want help turning prompts like this into systems your AI Teammates run on their own, that is the work we do at If Possible.