David Herse | May 19, 2026

Mapping the Geography of Giving: a free starter kit for fundraisers

Mapping the Geography of Giving was my talk at digi.raise 2026 on where Australian donors actually live. If you were in the room, this is the take-home: four LLM prompts you can run on your own donor file.

The premise

Most charities have a donor file shaped by their historical marketing reach. The postcodes you’ve targeted are the postcodes you’ve raised from. The postcodes you’ve raised from then drive next year’s targeting. The feedback loop is tidy and quietly expensive. The longer it runs, the more invisible the donors you’ve never asked become.

The kit’s central prompt is Example 02, Spot the bias. You feed it your donor file and your campaign history. It returns a map with three layers: where your donors actually live, where you’ve actually campaigned, and the unasked postcodes that look demographically just like the people already giving you money. Those unasked look-alikes are the headline value of the whole kit.

What’s in the kit

Four self-contained examples. Each is a folder with a prompt.md, a CLAUDE.md for Claude Code users, a README.md, an expected_output.md, and the sample donor CSVs needed to run it end-to-end. Every prompt produces two things side-by-side: a written analysis (answer.md) and an interactive MapLibre map (answer.html) coloured by the result. Open the HTML, share the markdown.

01. Donor fingerprint

What does my donor file actually look like, geographically? The prompt produces a choropleth of donor postcodes coloured by donor count, plus a written profile of the demographics those postcodes share. It’s the “who do we already have” baseline that every later analysis is measured against.

02. Spot the bias

The keystone prompt. Compares your donor file against your historical campaign list and surfaces the postcodes that look demographically like donors but were never asked. The map has three layers (donors, campaigned, unasked look-alikes) so the gap is visually unambiguous. The analysis recommends one cheap pilot postcode to test the hypothesis.

03. Score next campaign

You’ve got a list of postcodes you’re about to target. The prompt ranks them 0–100 by similarity to your existing donor base, draws a recommended cut-line, and outputs a map coloured by score with the cut-line outlined in gold. Use it as a second opinion before a campaign locks.

04. Donor cluster locator

Two maps side-by-side: the top postcodes by raw donor count, and the top postcodes by donor rate per 1,000 adults. The delta between the two, where you’re big in absolute terms versus where you’re disproportionately concentrated, is usually where the strategic story lives.

How it works

Every prompt fetches its reference data live from public ABS endpoints. Postcode boundary geometry comes from https://geo.abs.gov.au/arcgis/rest/services/ASGS2021/POA/, and Census features like income, age, religion, and housing tenure come from https://data.api.abs.gov.au/rest/. No auth, no registration, no Mapulus account required. The prompts are also model-agnostic. They were designed in Claude Code but they work in any LLM that can fetch a URL. Paste the prompt into Claude or ChatGPT and it will produce the same map and the same analysis.

The mapping stack is fixed across all four examples so the outputs feel consistent: MapLibre GL JS, CARTO light_all basemap, and GeoJSON inlined into the HTML so the file works after a single page-load. You can open the result on a plane.

A note on privacy

Your donor file goes to whichever LLM you run the prompt in: Claude, ChatGPT, or another. De-identify it before you paste. Names, emails, exact addresses, and anything else that could identify a person should come off the file first. Mapulus Transform is built for exactly this. It scans a CSV for PII locally and lets you redact, drop, or synthesise sensitive columns before the data ever leaves your machine.

And the disclaimer worth reading

Mapulus is not responsible for the accuracy, completeness, or fitness-for-purpose of any output generated from these prompts. LLMs are confident in the same voice whether they’re right or wrong. Treat every result the kit produces as a hypothesis worth testing. A pilot to run, a question to ask your board. Not advice to act on without validation. The Census features are real; the inference layered on top of them is not always.

---

If you came to the digi.raise talk, the kit is the practical follow-up. If you didn’t, the headline still stands: a fundraising team’s geographical confidence is usually a record of where it has been, not a map of where it would be welcomed. The prompts are an attempt to put a cheap, repeatable second opinion in your hands. Run them, argue with them, share the maps with your team.