Not-for-Profit /Retrieval (RAG) interactive demo field guide · 8 min

Donor Memory

Gives the relationship manager a 30-second briefing they used to spend an hour pulling together.

theater/demos/nfp_donor-memory.html · sandbox · read-only

Open

FIG. 1

The live demo, running on fabricated data. Open it to step through the full flow — every output is shown for a person to approve before anything happens.

How it would work

Reads only your own donor records and approved policies, answers the fundraiser's question with the source quoted beside it, and refuses — routing to a person — whenever the answer isn't grounded in your documents.

Input 01

Your records and policies

Your own corpus only: donor CRM export, gift acceptance policy, fundraising compliance handbook, and signed grant or funding agreements. Nothing from the open web.

Agent 02

Searches, grounds, cites

Retrieves the passages that match the question, drafts a plain-language answer, and attaches the exact record or policy clause it drew from with a grounding-confidence score.

Output 03

A cited briefing, or a refusal

A short briefing with every claim traceable to a source, for the relationship manager to read and act on — or, when nothing in the corpus covers the question, an honest refusal that escalates to a named person rather than guessing.

Where it works well

It reassembles a donor's whole history in seconds, with every claim traceable to the record it came from.

Done by hand it is most of an hour per donor — and the file note that says "email only, no phone" is exactly what gets missed under time pressure.
Best for a relationship manager or fundraising lead briefing before a call, an ask, or a stewardship touchpoint.
At a portfolio of dozens of major donors, the recaptured time goes back into the relationship and the ask, not the file-digging.

The slow, invisible cost before a major-donor conversation is the reassembly — pulling the pledge amount from the CRM, the contact rules from the file note, the recognition tier from the gift acceptance policy, and the tied conditions from the grant agreement, scattered across four systems.

Where it works badly

It is confidently wrong when the CRM is stale — and a cited answer reads as more authoritative than the data underneath it.

Weak when the answer depends on context that was never written down — a verbal commitment, a relationship nuance, a soft "not yet" the manager holds in their head.
If a donor's give is split across restricted and unrestricted gifts and your records don't separate them, it can quietly conflate tied funds with general revenue.

The honest test

If you cannot say, right now, when this donor's record was last reconciled against what actually landed in the bank — this tool makes your out-of-date answer faster, not more accurate.

Point it at a donor record that hasn't been updated since the pledge was revised, or a gift policy that's a version behind, and it produces a clean, sourced briefing built on a figure that no longer holds. The citation makes the wrong number look checked.

What it doesn't do — and shouldn't

It briefs. A person decides what to do with the donor. That boundary is deliberate.

WHAT IT DOES

Quotes the donor record, file note or policy clause behind each claim

Shows a grounding-confidence score against the matched sources

Refuses, and routes to a named person, when the corpus doesn't cover the question

WHAT IT WON’T

Send anything to the donor or solicit a gift

Decide whether a gift may be accepted under your gift acceptance policy

Treat a question it can't ground as one worth guessing at

A donor briefing touches sensitive personal information governed by the Privacy Act, money that may be tied to a grant agreement's conditions, and a relationship the charity is accountable for to the ACNC and its board. A wrong contact method or a misread tied-fund condition damages the donor's trust and the charity's standing — so the accountable fundraiser stays on the decision, because the consequence lands on them, not the tool.

What your data has to look like

Your own donor records and approved documents — current, and clear about what each gift is tied to.

Typical readiness

across orgs we see, before the first job

A donor CRM export with structured gift records

Usual weak point

Restricted vs unrestricted clearly marked

Needs shaping

A current gift acceptance policy

Usual weak point

Signed grant and funding agreements in the corpus

Needs shaping

Contact preferences and consent recorded against each donor

Needs shaping

The real first job

The donor CRM is usually the weak point — gift records half-structured, tied and untied money blurred, contact preferences in a notes field nobody maintains. Fixing how donor and gift data is captured and kept current is usually the real first job, larger and more valuable than the AI layer on top. Once the records are clean and current, every briefing after that is faster and right by default.

Right fit if…

You hold a portfolio of major donors and brief before calls, asks or stewardship touchpoints

Your donor history is scattered across a CRM, file notes and grant agreements

Your gift records distinguish restricted from unrestricted money

You can point to the current gift acceptance policy and signed agreements today

Walk away if…

Your donor "record" is one fundraiser's memory and a spreadsheet they own

Restricted and general gifts are mixed together with no way to tell them apart

The relationships are few enough that reassembly was never the bottleneck

You want a tool that decides whether to accept a gift or how to approach a donor

Open questions

The worried-buyer questions, answered straight

It can surface a wrong detail — which is exactly why nothing is sent or decided on its say-so. Every claim in the briefing is quoted from the donor record, file note or policy clause it came from, with a grounding-confidence score, and when the corpus doesn’t cover the question it refuses and routes to a named person rather than guessing. The fundraiser reads the sources and decides how to approach the donor. The tool briefs; the person stands behind the approach.

It answers from whatever is in the corpus, so a half-maintained CRM gives half-maintained briefings, and it can’t tell a stale pledge figure from a current one. The biggest gain is usually getting gift records into clean, structured form — amount, date, fund, contact preference — with restricted and unrestricted money clearly separated. That data work is the first piece, and it pays off across every briefing afterwards.

No. It removes the reassembly — the half-hour of pulling a donor’s history together from four systems — so the relationship manager spends their time on judgement: how to steward this donor, when and how large the ask should be, what the relationship needs. The donor relationship is still theirs. The capacity it frees goes back into donor-facing work, not out the door.

Current to the last reconciliation. A pledge can be revised, a gift can land, a donor can change how they want to be contacted, and a gift acceptance policy gets reissued — point it at a stale CRM and it briefs confidently on figures that no longer hold. The honest test: do you know, today, when this donor’s record was last reconciled against what actually came in?

Donor records are sensitive personal information under the Privacy Act, and most ACNC-registered charities above the turnover threshold are bound by it directly. Any deployment runs against your own systems and data handling, not a shared pool, and the corpus is your documents only — it never reaches out to the open web. We scope where the data sits, who can see it, and how consent and opt-outs are respected as part of the build. The demo here runs entirely on fabricated data; Margaret Whitlam is not a real donor.

It can tell you what your records say about a donor, but using donor data to solicit gifts is direct marketing, which is governed by the Privacy Act’s direct-marketing principle and the contact preferences and opt-outs each donor has set. The briefing reflects those preferences so a fundraiser doesn’t approach someone who has opted out — but the decision to contact, and compliance with fundraising rules, stays with your team. It informs the call; it doesn’t make it.

What it takes to build

4–5 weeks · 4 phases

Reused from template~65%

Bespoke to this skin~35%

stack · Claude · retrieval index · review UI

What it would cost

Fixed scope, fixed price, fixed dates.

Bite-sized first piece

One contained change, low risk

Pilot build

Most builds land here

Embedded support

Scale on proof

Considering this for your charity?

The honest place to start is a bite-sized first piece — one contained change, low risk. Tell us where the donor history is hardest to reassemble; we'll play it back, scope it, and show you what's possible.

Book a call How we work