Upgrade Ask ID at Renewal

At renewal, many donors are asked for the same amount they gave last year. Some of those donors could give more; others will churn if the ask jumps without a clear signal. Upgrade ask identification uses machine learning to score which renewing donors look statistically ready for a higher gift, and to propose a suggested ask amount for the officer to review.

The model does not send appeals or lock in dollar figures. It surfaces a readiness flag and a suggested amount so a gift officer can decide whether to upgrade, hold steady, or soft-ask with a range. That human decision stays with the person who knows the relationship, the campaign context, and any constraints the data cannot see.

Related reading: Renewal Timing Optimization.

Upgrade readiness is usually inferred from patterns in giving history and engagement, not from a single field. Useful inputs often include consecutive renewal years, gift size trajectory, recency and frequency of gifts, response to prior asks, and soft indicators such as event attendance, volunteer activity, or multi-channel engagement. Capacity proxies (wealth screens, employer matching eligibility) can inform the ceiling, but they should not alone decide that someone is “ready” for an upgrade at this renewal.

The model learns which combinations of these signals historically preceded successful upgrades versus flat renewals or lapses. A donor with steady mid-level gifts and rising engagement may score differently from a one-time large donor whose subsequent gifts declined. The score is comparative: relative to peers and to that donor’s own baseline, not an absolute wealth label.

When the available history is too thin to score (new donors, sparse gift records, or incomplete engagement data), the system should return empty output rather than a guess. Thin history is a known failure mode for ask models; forcing a number creates false confidence and can push an ask that the relationship cannot support.

Pair scoring with appeal design: Personalized Renewal Appeal.

A readiness flag alone is not enough for renewal operations. Officers need a concrete suggestion: stay at last gift, step up by a modest increment, or move into a defined band. Suggested amounts are typically derived from the donor’s recent gift distribution, peer cohorts with similar histories, and historical acceptance rates for upgrade steps of different sizes.

Common patterns include a modest step from last gift (for example, a percentage or fixed increment within a band), a range rather than a single point, and an explicit “hold” recommendation when readiness is low. The suggestion is a starting point for conversation and letter copy, not a mandate. Officers should be able to override downward when relationship knowledge contradicts the score, and upward when they have evidence the model lacks.

Document how suggestions are calculated in enough detail that officers trust them: which gifts count, how outliers are handled, and whether matching gifts or tribute gifts are excluded. Opaque recommendations get ignored; ignored models waste the scoring investment.

Human-in-the-loop is the operating rule. The model flags upgrade readiness and a suggested amount; the officer still sets the ask. That split protects donor relationships and keeps institutional judgment where it belongs.

Practical control points include: review queues sorted by readiness and gift size, one-click accept/adjust/hold on the suggestion, and notes fields for why an officer overrode the model. Overrides should feed back into later evaluation so the team can see where relationship knowledge consistently beats the score (good) versus where overrides look random (process noise).

Do not auto-mail upgrade asks from the score alone. Renewal packages, phone scripts, and personal notes should reflect the officer’s approved ask. Automation can draft language around an approved amount; it should not invent the amount after the packet has left review.

A workable renewal loop looks like this. First, define the renewal cohort and the window in which asks will be finalized. Second, score only donors with enough history; leave thin cases unscored and route them to standard renewal treatment. Third, present readiness, suggested ask, last gift, and key supporting signals in the officer’s worklist. Fourth, capture the final ask and the rationale when it differs from the suggestion. Fifth, measure outcomes after the campaign closes: upgrade acceptance, average gift change, renewal rate among upgraded vs. held donors, and officer override patterns.

Separate “ready for upgrade” from “at risk of exit.” A donor who scores high for upgrade may still need careful timing or softer language if other signals suggest fatigue. Exit-oriented analysis belongs in a companion workflow so upgrade pressure and retention care do not collide in the same blind push.

See also: Lapsed Donor Exit Signal Analysis.

Upgrade models fail in predictable ways. They over-score recent large gifts that were one-time events. They treat matching gifts as personal capacity. They recommend steps that are too large relative to the donor’s history, which can suppress response even among willing supporters. They also under-score loyal mid-level donors whose capacity grew quietly while their ask stayed flat for years.

Quality checks before each renewal cycle: confirm empty output for thin history, audit suggestion distributions against last-gift baselines, sample high-score cases with officers who know the donors, and track whether accepted upgrades actually convert at rates that justify the extra ask risk. If conversion among model-suggested upgrades is no better than random step-ups, pause and recalibrate before the next cycle.

Keep the outcome focused on ask quality: better targeting of who is invited to give more, clearer suggested amounts for human review, and fewer speculative upgrades that trade short-term dollars for long-term trust.