Case study - Good Church

Good Church: governance answers in seconds, straight from your own documents

Good Church is an AI governance and policy advisor for New Zealand churches. Staff and volunteers ask a question in plain English and get an answer drawn from their own church's documents, with the source cited. We took it from a working MVP to a multi-tenant platform churches can sign up to themselves.

The Good Church website showing an example question about youth events answered from a church's Child Protection Policy, with the clause cited
Client
Good Church Ltd, New Zealand
Product
AI governance and policy advisor for churches
Our role
Phase 2 platform build, hosting and ongoing development
Live
goodchurch.co.nz, public sign-up since July 2026

The brief

Clare Pattison and Dan Cooper started Good Church in late 2025 to get what a church board has already decided into its people's hands fast. Anyone on a church team - committee members, leaders, volunteers - should be able to ask "can a new volunteer drive the youth van?" and get the church's own answer.

They came to us with a working MVP built by another developer. The question-answering pipeline was there, but it served one organisation. Phase 2 was about turning it into a product: true multi-tenancy was a hard requirement, along with per-church document upload, Charities Register data, legislation monitoring, safeguarding alerts and billing.

What we built

Phase 2 went from handover to public sign-up in about four weeks, in July 2026.

Platform

Multi-tenant from the database up

  • Every church is isolated with row-level security in Postgres, not just in the app
  • Three roles: members, church admins and platform admins
  • Self-serve sign-up, with each church found on the Charities Register
  • In-app document upload, processed on the server
  • Stripe subscriptions with plan tiers and member caps
Answers

Cited answers, tuned to who is asking

  • Answers adapt to the asker's role - governance depth for the board, practical steps for volunteers
  • Every answer points back to the clause it came from
  • A benchmark harness to test answer quality as documents change
  • A backup route to the AI model, so answers keep flowing if the main provider is down
Compliance

Staying ahead of law changes

  • Monthly monitoring of legislation.govt.nz, the Privacy Commissioner and WorkSafe
  • Claude summarises each change and drafts suggested policy wording for each church size
  • Nothing reaches a church until an admin has approved it
  • Conflict detection flags where a church's own documents contradict each other
Safety

Safeguarding alerts

  • Questions that touch care, safety or protection alert the church's safeguarding contacts by email
  • Leaders review them in an anonymised queue, so a real person can follow up

How the AI is used

The rule for every AI feature: stay inside the documents the board has already signed off, and keep a person in the loop wherever it matters.

Search that finds the clause

Hybrid search over each church's own documents, with OpenAI embeddings, finds the right section before Claude writes an answer.

Claude for the writing

Anthropic's Claude writes the legislation summaries and suggested policy wording, and powers the conflict checks.

Small model first, big model to confirm

For legislation relevance and conflict checks, a fast model screens everything and a stronger model confirms what it flags.

People approve, AI drafts

Law-change recommendations wait for admin approval, and safeguarding questions go to a person, not just a chatbot.

Built to be trusted

Churches are handing over their policies and their people's questions, so the platform had to earn that.

  • Each church's data is walled off at the database level.
  • Answers are drawn from the documents the church has uploaded, with the source cited.
  • Safeguarding questions are flagged to the church's leaders for review.
  • Backed by more than 350 automated tests.

Got a document-heavy compliance problem?

Policies, procedures and legislation that keeps moving. If your people need fast answers from your own documents, we can build it, host it and keep it running.

Last updated 19 September 2026