Campaign Segmentation Optimization

A capital-campaign segment plan is a roster of alumni cells with provenance: named giving and engagement fields, the rules that put people in the cell, and draft recommendations for ask framing, timing, and channel. Gift officers still own the ask. The model does not invent an amount. If a cell is too thin to support a recommendation after exclusions, that cell stays empty.

The plan is briefing material for campaign operations and assigned officers. It is not a mail merge, a pledge book, or a substitute for prospect strategy. Scores from an alumni giving propensity model can help decide who is even a candidate for a cell. They do not replace the field cites that make the cell defensible in a solicitor meeting.

Write the artifact so a new officer can read one row and know why the person is there, what is being proposed, and what is still undecided. If a row cannot answer those three questions, it is not ready.

Pull a dated extract and freeze it. Record the as-of timestamp, the source systems, and the population rules: living alumni, legally contactable, campaign-eligible, exclusion and do-not-solicit flags honored. Do not restream nightly updates into the clustering job while you are still naming cells. A moving file produces moving segments, and then nobody can explain why a person landed where they did.

Giving history, campaign assignments, and engagement records typically live across advancement CRMs and campus systems in the same class as Salesforce, Ellucian, Workday, and Anthology. Export the fields you intend to cite. Do not assume each system stores the same events the same way. Map every field to a definition an officer can read without a data dictionary open: last gift date, last gift amount, lifetime giving, consecutive years, reunion cycle, event attendance, volunteer service, mail or email responsiveness, solicitor assignment, and any campaign-priority flags already on the record.

Lock a do-not-model set before you run anything. Deceased records, current students mixed into an alumni dump, faculty-staff with different gift policies, and legally restricted contacts all belong out of scope in writing, not as a surprise after clusters appear. Re-extract only when campaign operations agrees the freeze is stale, then version the plan so older cells are not silently rewritten.

If two source systems disagree on last gift date or solicitor, resolve the conflict in the freeze notes. Clustering cannot fix a join you have not named.

Every kept cell needs a short provenance note: which fields defined membership, which ranges or categories applied, and how many records met the rule after exclusions. If you cannot point to those fields, you do not have a segment. You have a label.

Keep annual-fund style volume cells apart from cells that belong with major gift prospect identification. A high lifetime total plus an assigned solicitor is not an email-ask cell. A recent reunion attendee with modest consecutive giving is not a leadership-gift cell. When a planned-giving overlay is useful, park it next to a planned giving propensity model instead of dumping those people into a generic high-capacity bucket that cites nothing.

Watch for cells that are retention problems wearing campaign clothes. People whose last gift is old, whose consecutive-year streak has broken, or whose engagement has gone quiet may need a conversation informed by a donor retention risk model, not a blast timed to the public phase. Cite recency and engagement either way so the officer can see why the cell exists.

When a proposed cell collapses to a handful of records after exclusions, do not pad it with lookalikes that fail the cited rules. Leave it empty. Empty is a quality outcome. A mailed thin cell is not.

Reject cells whose only membership rule is a vendor wealth screen, a purchased score, or a cluster ID with no giving or engagement fields attached. Those signals can sit in an appendix. They cannot be the cite.

For each populated cell, write three drafts: ask framing, a timing window, and a channel. Framing is not a dollar figure invented by the model. It is a pointer to evidence already on the record, such as prior gift size, consecutive-year pattern, or an existing solicitor strategy note, plus an explicit instruction that the officer sets the number.

Timing drafts should cite campaign phase, reunion year, fiscal-year close, or event dates present in the extract. Do not invent a best day. Channel drafts should cite demonstrated responsiveness where you have it (mail, email, call, visit) and default to the assigned officer's existing practice when you do not. If the extract has no channel evidence, say so. Do not fill the gap with a preferred mailing workflow.

Inventing an ask amount is the fastest way to poison the plan. A number with no gift history behind it will not survive the first prospect-strategy meeting, and it trains staff to treat the file as if it already knew the donor. Leave amount blank. Leave "officer to set" as the value.

Illustrative example: campaign operations freezes a living-alumni extract on a fixed Monday. One proposed cell is consecutive annual donors, last gift inside two fiscal years, at least one reunion or volunteer event in the extract window, and no major-gift assignment. Provenance lists those four conditions. The draft says ask framing follows last gift and consecutive-year history, and the officer sets the amount; timing is the quiet phase before the class-reunion weekend already on the engagement file; channel is a solicitor call if an officer is assigned, otherwise a personal note from the class volunteer, not a general appeal. A second proposed cell, unassigned alumni with only an uncited wealth screen and no giving or engagement fields, has too few records that survive exclusions and no citable behavior. That cell is left empty. No ask is generated. Nobody is mailed from it.

Route the plan to assigned gift officers and campaign operations together. Officers confirm or reject cell membership, rewrite the ask, change timing, and change channel. They also pull people out for personal qualification. Until that review, the plan is a draft. After review, still treat it as a work plan, not a pledge. A recommended ask is not a commitment. A timing window is not a booked visit. A channel is not permission to drop the cell into the next appeal.

Two other failure modes travel with invented asks. Mailing a thin cell because the label sounded useful produces a tiny, odd list that confuses volunteers and burns the relationship for no campaign reason. Treating the segment plan as a pledge, or booking it as revenue in a campaign tracker, records money that has not been asked for, let alone given.

Keep the quality bar in the artifact itself: cites on every kept cell, blank rows where the extract was too thin, draft language that never states a dollar figure the officer did not set, and a review status that stays unreviewed until a named officer has signed it. That is the product. Commentary about models is optional. The freeze, the cites, the empty cells, and the officer-owned ask are not.