Donor Lapse Risk Scoring

Donor lapse risk scoring uses a machine learning model to estimate how likely each active or recently active donor is to miss their next gift or drop out of the annual fund cycle. The model draws on recency, frequency, and engagement history, then ranks constituents so retention and annual-fund staff can focus outreach before silence becomes a lost relationship.

Tools such as DonorSearch AI and Virtuous Insights are built for this kind of predictive scoring inside nonprofit CRMs and fundraising platforms. The output is a risk score or tier per donor, not an automatic ask. Staff still decide who gets a personal call, a tailored email, a stewardship touch, or no action yet.

A usable lapse score depends on three families of evidence. Recency captures how long it has been since the last gift or meaningful interaction. Frequency reflects gift cadence over a defined window, such as the past 12 or 24 months. Engagement covers non-gift signals that often precede renewal or attrition: email opens and clicks, event attendance, volunteer activity, petition or advocacy actions, and soft credits where your CRM records them reliably.

When any of those histories is missing or too thin to trust, the page (and the workflow behind it) should return empty output for that constituent. A score built only on gift date, or only on email clicks, is not a lapse forecast. Suppressing incomplete rows keeps the queue honest and prevents staff from treating a weak estimate as a priority list.

Secondary attributes can sharpen the model once the core signals are present: gift size trend, channel preference, campaign response history, and tenure. They should not substitute for missing recency, frequency, or engagement data.

Start from the CRM or insights product that already holds gift and activity history. Configure or refresh the risk model on a schedule that matches your renewal calendar, weekly for large annual funds, or monthly for smaller portfolios. Export or sync a scored list that includes donor ID, risk tier or probability, and the top contributing factors when the tool exposes them.

Filter to donors who are still in the active or at-risk window you care about, for example annual-fund renewals due in 30–90 days, or multi-year givers whose last gift is approaching your lapse threshold. Exclude already-lapsed records if a separate win-back process owns them. Route high-risk names to a human review queue: gift officers, annual-fund associates, or retention specialists who know relationship context the model cannot see.

Staff then choose the intervention. A high score might mean a phone call from a relationship owner, a handwritten note, or a pause on another solicit until stewardship is done. A mid-tier score might mean a personalized renewal email rather than a mass blast. Low-risk donors stay in the standard cadence. The model ranks urgency; people decide contact strategy, tone, and ask amount.

Lapse scoring sits between segmentation and campaign execution. Behavioral segments explain who donors are; risk scores explain who is about to leave the renewal path. Those scores should feed sequence builders and campaign calendars so high-risk cohorts get earlier or denser touches, not louder generic volume.

Tie scores to clear operational definitions. Agree on what “lapse” means in your shop (missed anniversary gift, no gift in 12 months, dropped from monthly giving) and keep that definition stable so year-over-year comparisons mean something. Align risk refresh timing with appeal drops so lists do not go stale between model run and mail or call day.

Governance matters. Limit who can override or bulk-export scores. Document that predictions are advisory. When a gift officer marks a donor as “do not contact” or notes a life event, that human judgment should win over the model for that cycle.

Treat scores as relative priority inside your file, not as absolute probabilities you quote to the board. Compare donors within the same program and gift band. A “high risk” major-donor prospect and a “high risk” $50 annual giver need different staff time even if the label matches.

Inspect drivers when the product shows them. A spike driven by long gift gap plus falling email engagement suggests a different play than a spike driven only by one missed mid-level ask. If drivers are opaque, spot-check a sample of high-risk names against known relationships before you scale calling.

Expect false positives and false negatives. Some donors go quiet and still renew; some look healthy and still leave. Track outcomes: of donors scored high risk last quarter, how many renewed after intervention, how many lapsed anyway, and how that compares with an unscored or lower-touch group. Use those results to adjust thresholds and staffing, not to claim the model is infallible.

Empty or suppressed scores are a feature. They tell you data hygiene work is unfinished for that record, which is often more valuable than a fabricated number.

Confirm recency, frequency, and engagement fields are populated for the portfolio you want to score. Define lapse and active windows with development leadership. Schedule model refresh against the renewal calendar. Build a human review queue with capacity for the top risk tier only, so the list stays actionable. Decide default interventions by tier before the first scored export. Suppress incomplete records rather than forcing a score. Review a sample with relationship owners each cycle. Measure renewal and re-engagement rates for scored cohorts over time, and revise thresholds when the queue is either too large to work or too small to matter.

Done well, lapse risk scoring shortens the gap between early warning signals and a deliberate human touch. The model surfaces who is drifting; retention staff still decide who to call and what to say.