Resource development decision inputs · Lesson 3

Donor segmentation without unsupported inference

6 min read

Lesson objectives

  • Build a reproducible segmentation rule from gift history, stated interests and contact preferences, and identify records requiring correction or a pause.
Reading tools

When a donor database contains hundreds of supporters, one message rarely serves everyone well. Segmentation, however, is not permission to guess someone's personality or wealth. It is a stated rule that groups records for a defined communication purpose using data the organization may use. A ‘reading program interest’ segment could include people who chose that interest on a form or previously designated a gift to reading. It should not include people whose interest was inferred from their names or neighborhoods. Before using the file, the specialist checks access, retention and contact-preference rules with the responsible owner. Permission to analyze a record internally does not automatically authorize a new message to every saved address.

Define the decision the segment will support. Is this a progress update on a funded service, an event invitation, or a new gift request? Relevance and permission can differ across these purposes. Write each segment rule so a second worker could apply it: ‘A verified gift to the library within the past twelve months, agreement to program updates by email, and no later withdrawal of that preference.’ ‘Strong potential donor’ is not a rule unless its criteria are specified and justified. Time matters: a gift made years ago does not prove present interest. Keep ‘unknown’ separate from ‘not interested.’ Missing data is a reason to verify or limit targeting, not a judgment about the person.

Separate recorded facts from interpretations. In a fictional file, the gift-designation field may record the library, while the donor's reason for giving remains blank. An event ticket shows participation; it does not establish readiness for a major gift. An unopened email could reflect an old address or a mail setting rather than rejection of the cause. Give every rule a data source and last-updated date. If two fields conflict, such as a newer stop request and an older opt-in, follow the verified newer preference. If identity or timing is uncertain, quarantine the record for review instead of choosing whichever value increases the mailing list.

Use a three-layer decision sheet. First is contact eligibility: is this channel and purpose allowed under organizational policy and the applicable local rules? Second is content fit: does a recorded action or stated interest connect this person to the library message? Third is relationship stage: would an acknowledgment or progress report be appropriate before another ask? The third layer never overrides the first. An important donor is not exempt from a request to stop contact. One person can belong to more than one interest group, but the system should suppress duplicate sends rather than contact that person repeatedly on the same day. A contact calendar must account for preference and frequency.

Test the rule on a small sample before releasing a campaign list. Inspect boundary cases: a library gift coupled with a marketing refusal; permission for updates but not for appeals; and a general gift with no specified subject. Make sure the outcome can be explained for each record. Reconcile group counts with the eligible population. If a record disappears or occurs twice in a single send, investigate the join or exclusion rule. Do not export a whole personal-data file outside the approved system to fix three cases. Use synthetic record IDs or aggregate counts in training and review. Note the query date too: a segment correctly built yesterday may be wrong after a preference changes today.

The deliverable is more than a list of names. It includes the segment definition, communication purpose, permitted channel, matching message, exclusion criteria and run date. A communications owner can then repeat the query or halt the send if a complaint appears. Do not describe this result as a certain prediction of future giving. It is a disciplined way to organize known evidence. A different team seeking to reuse the list for a new purpose must reassess purpose and permission. Where data is sparse or unreliable, a restrained general message may be more honest than apparent personalization based on intuition. Segment quality lies in defensible, reviewable decisions, not in producing ever smaller groups.

Lesson summary

Segment donor records with reproducible rules based on stated giving, interests and contact preferences. Pause where permission is missing, and do not infer wealth or sensitive traits from weak signals.