AI Copilot
The AI proposes. A human approves. Nothing writes itself.
Copilot reads your live donor and campaign data and can propose changes — segmenting donors, scheduling a campaign, voiding a donation. Every proposal sits in an approval queue until someone on your team says yes, and every approved action can be undone. Nothing executes on its own.
- Propose → approve → execute → undo
- Every Copilot action, before anything changes
- Zero LLM calls
- Donor engagement scoring is plain SQL, not a model
- 0 credits
- Approved actions never cost a credit
- Every plan
- Engagement scoring runs nightly, Free included
Copilot chat
It reads your org's numbers, not the internet
Ask it about donation stats, donor stats, or campaign performance and it answers from your own data through a set of read tools — not a generic chat box bolted onto your dashboard. On the Free plan, Copilot is read-only: the tools that could change anything are stripped from the request before it ever reaches the model, so there is nothing to approve or undo because there is nothing it can do.
- Reads donation stats, donor stats, and campaign performance from your org
- Free plan is read-only by construction — action tools are removed before the request is sent
- Metered by credits, not tokens: one credit per conversation turn
Ring 3 — propose, approve, execute, undo
Every write sits in a queue until a human says yes
Copilot can propose eight kinds of action — segmenting donors, assigning a receipt number, flagging a compliance review, scheduling a campaign, dismissing an insight, tagging a donor, updating a campaign goal, or voiding a donation. None of them execute on proposal. Each lands in an approval queue for an admin or finance role to accept or reject, and approved actions can be undone.
- Eight proposable action types, each with its own approval and undo
- Approval queue is a separate screen your team reviews before anything changes
- Approved actions cost 0 credits — you're not charged for saying yes
Donor engagement score
A churn-risk score you can actually audit
Every night, for every organization, DonorCARE computes a 0–100 engagement score from six components — recency, monetary value, momentum, frequency, tenure, and diversity of giving — and classifies churn risk as low, medium, high, or critical. There is no model and no prediction here: it's a documented rule ladder, and every risk flag comes with a plain-language reason, such as giving at 2.3 times a donor's normal cadence gap.
- 0–100 composite score from six measured components, recomputed nightly
- Churn risk (low / medium / high / critical) with a written reason, not just a label
- An active, healthy recurring donor is never scored above low risk
Frequently asked questions
Can the AI change anything without someone approving it?
No. Every proposable action — from tagging a donor to voiding a donation — creates an approval request. It only takes effect once a human with the right role approves it, and approved actions can be undone afterward.
Is the donor engagement score an AI prediction?
No. It's a rule-based composite built from recency, monetary value, momentum, frequency, tenure, and diversity of giving, computed by plain SQL — not a machine-learning model and not a forecast. It describes what a donor has done, with a written reason for every risk flag.
What can Copilot do on the Free plan?
Read only. It can answer questions about your donation, donor, and campaign data. The tools that propose changes are removed from what the model is given, not just instructed to avoid — so there is nothing for it to write, and nothing to approve.
Do approved actions cost extra credits?
No. Conversation turns use credits; approving and executing a proposed action does not.
Does Copilot send our donor data to a third-party AI model?
To answer a question, yes — Copilot calls Anthropic's Claude Sonnet 5 model, and the data its read tools return (donation stats, donor stats, campaign performance) is part of that request. What stays fixed: every tool call is scoped to the requesting organization and re-checks that scope before returning anything, so one org's Copilot conversation cannot pull another org's donor data into the request in the first place.
Can we undo an action after it's approved and executed?
Yes. Each of Copilot's eight proposable action types — from tagging a donor to voiding a donation — has its own undo function alongside its executor, so an approved action that turns out to be wrong can be reversed, not just flagged for next time.
See your own donor data talk back
Start on the free plan and ask Copilot about your first campaign today.