Executive summary · September 2026
Recurring-Gift Recovery Agent
A decline-aware recovery policy and an agent that triages each failed monthly gift, proposed as a six-week backtest on de-identified BBIP retry history.
Recovered per year
$1.04M
vs $766k with fixed retries
Uplift
+15.4 pts
43.8% → 59.2% of failed dollars
Retries avoided
64%
87,947 fewer authorisation attempts
Do-not-retry violations
27,926 → 0
retries on do-not-retry codes or past the cap
Donors kept
+1,358
$737k of 12-month value protected
Illustrative model on a synthetic portfolio of 40,000 monthly donors at five fictional nonprofits, twelve months, 35,992 failed charges ($1.75M). The real number comes from a backtest on de-identified BBIP retry history.
The problem
Failed monthly gifts are retried on a fixed daily schedule with a monthly card-account update. Every decline is treated the same: hard declines are retried although they cannot succeed, insufficient-funds retries land before payday, expired cards wait for month end, and donors who need to update their details get one generic notice. Each unrecovered gift risks the whole recurring relationship.
The method
- ▸ Read the decline: never retry do-not-retry or hard codes.
- ▸ Query the card updater immediately for expired and reissued cards.
- ▸ Time insufficient-funds retries to the donor’s payday; at most three.
- ▸ An agent triages each gift and drafts a warm email and text with a self-serve link.
- ▸ Both policies replay the same failures with the same random draws.
Recovered dollars, cumulative
- Smart policy + agent
- Fixed daily retries
Where the dollars came from
- Retries
- Updater
- Outreach
By organisation (fictional)
| Organisation (fictional) | Failed | Baseline | Smart | Uplift |
|---|---|---|---|---|
| Riverbend Food Collective | $352k | $155k 44% | $212k 60% | +16.2 pts |
| Northgate University Foundation | $524k | $234k 45% | $310k 59% | +14.4 pts |
| Harbor Light Children’s Hospital Foundation | $434k | $187k 43% | $253k 58% | +15.2 pts |
| Cedar Valley Animal Rescue | $210k | $94k 45% | $128k 61% | +16.2 pts |
| Lumen Arts Alliance | $230k | $96k 42% | $133k 58% | +16.1 pts |
What the six-week backtest would do
Weeks 1–2
Offline backtest
Replay de-identified BBIP retry history through both policies inside Blackbaud’s environment. Output: the real uplift, by decline code and by customer.
Weeks 3–4
Agent
Tune the policy on the backtest; run the triage agent on a sample of failed gifts and review its decisions and drafts with the payments team.
Weeks 5–6
Shadow pilot
Three to five customers. The agent recommends, people approve; measure recovered dollars, retries avoided and donor responses against the current schedule.
Controls
Human approval
The agent proposes a plan. A person approves it before any retry or message goes out.
AI disclosure
Every donor email and text says AI helped write it and a person approved it. Enforced in code.
Network rules in code
Do-not-retry codes are never retried and reattempts stay under the cap: 0 violations in the backtest.
Impact ledger
Every decision and outcome is recorded in a hash-chained ledger that anyone can re-check.
Data handling
No donor or card data leaves Blackbaud. The backtest runs on de-identified retry history, in Blackbaud’s environment, under Blackbaud’s controls. Nothing on this page uses Blackbaud, donor or card data.
Commercial model
Outcome-based: priced on recovered dollars measured in the pilot. Build and hand over — the policy, the agent and the evaluation harness become Blackbaud’s to run.