Light through a stained glass window falling in colour across a stone wall

AI, included in the $19/mo plan

An assistant that only knows your church.

Every church-software company added AI this year. Almost none of them will tell you what it can see. SundayBridge will: it reads the records your congregation already keeps, it cannot address another church's data, it is forbidden from inventing a number, and it never sees the inside of a pastoral care note.

Your records onlyNo invented numbersCare notes never readIncluded · no add-on

Why you can trust it

Four promises, and the reason each one is more than a promise.

Trust in software is not a feeling you talk someone into. It is a set of things that are structurally true, or not.

01Scoped

It sees your church. Full stop.

Most software asks an AI politely to stay in its lane. Ours cannot leave it. The boundary is enforced by the database itself, beneath every line of code — the assistant reads through an account that is allowed to look at your congregation and literally nothing else. If the scoping ever failed, it would return nothing. It fails closed, not open.

02Grounded

It can't invent a number.

The model is never asked to calculate. It receives figures that our own code computed — the same figures on your dashboard — and its only job is to put them into a sentence. On the insight panels, a number it wrote that we didn't compute causes the sentence to be discarded before you ever see it.

03Discreet

The confidential stays confidential.

A pastoral care note is the most sensitive thing a church keeps. The assistant can be told that three care cases are open. It is never shown a word of what is inside them. Giving works the same way: totals and trends, never an individual giver's name attached to an amount.

04Honest

It tells you when it doesn't know.

If your records don't hold the answer, it says the record is empty and points you to the screen where you'd enter it. It does not guess, soften, or fill a silence with something plausible. This is the single most important property, and the hardest one to fake.

What it gives you back

Three questions, three Tuesdays saved.

The point of the assistant is not that it is clever. It is that the thing you were going to do on Tuesday night, in a spreadsheet, is already done.

You ask

Which regulars have quietly stopped coming?

It answers

Four people who attended at least twice a month have missed the last three Sundays. None of them has an open follow-up.

What that saved: Twenty minutes of scrolling an attendance sheet, on a Tuesday you didn't have twenty minutes.

You ask

How did giving do this quarter against last year?

It answers

Up 6% on the same quarter last year. The giver count is flat — the increase is in average gift, not in more people giving.

What that saved: A board answer that took an evening in a spreadsheet, and was one pivot table away from being wrong.

You ask

Who is serving every single Sunday?

It answers

Two volunteers served all four Sundays this month. Nine people have never been assigned a role.

What that saved: The burnout you would have found out about when someone quietly quit.

Ask AI

Ask in plain English. Get a sentence.

No report to build first, no export, no pivot table. Ask the way you would ask a person who had read every record, and get an answer with the summaries it consulted written underneath, so you can check its work.

The only six things it can read

Not your database. These, and nothing else.

Congregation summary

Head counts by role, household totals, and who is new.

Attendance trend

Weekly attendance across its own history.

Giving summary

Totals and trends — never which individual gave what.

Follow-up status

What's open, what's owed, and how long it has been.

Groups and serving

Membership and assignments across teams.

Care cases

Counts and status only — never the contents of a private note.

Ask AI is switched off entirely in the public demo — the demo church's records are not yours to ask about.

app.sundaybridge.com/ask

Which regulars have quietly stopped coming?

Four people who attended at least twice a month have missed the last three Sundays: Marcus Bell, Priya Nair, and the two adults of the Okafor household. None of them has an open follow-up.

Read attendance trend, congregation summary.

How did giving do this quarter against last year?

Giving this quarter is $71,480 across 138 givers, up 6% on the same quarter last year ($67,410). The giver count is essentially flat — the increase is in average gift, not in more people giving.

Read giving summary.

Is the Alvarez family planning to become members?

Your records don't say. I can see the Alvarez family first visited on July 5 and has attended twice since, but membership intent isn't something SundayBridge stores. The Follow-up tab is where a conversation about it would be recorded.

No tool returned an answer.

7 of 10 questions left this month · answers come from your records, not the open internet

app.sundaybridge.com/dashboard

What changed this week

Attendance averaged 88 over the last four Sundays, the highest four-week stretch this year. Three follow-ups have been open more than a week, and one of them — a first-time guest from July 5 — still has nobody's name against it. Two volunteers served every Sunday this month.

AI-generated · estimates, not advice

  • 01

    Detected. Deterministic code finds what's unusual — no model involved.

  • 02

    Scrubbed. Emails, phone numbers, and birthdates are stripped before it's sent.

  • 03

    Narrated. The model writes prose. It is not allowed to calculate.

  • 04

    Checked. Any number it wrote that isn't a real figure is thrown out.

Insights, on 12 pages

The paragraph you'd have written, if you'd had time to read the numbers.

You never ask for it. Open a tab and it is already there: the follow-up that has sat a week, the attendance stretch, the volunteer serving every Sunday.

Ordinary code — not a model — decides what is worth pointing at. The model only writes the sentence. And if it writes a number that is not one of your real figures, that sentence is thrown away and a plain template runs instead. It is structurally unable to make a number up.

DetectScrubNarrateCheck

Emails, phone numbers, and dates of birth are stripped out before anything is sent. Every panel is labelled AI-generated · estimates, not advice.

What it will not do

The limits, said out loud.

A vendor who lists only capabilities is selling you a checkmark. Here is where ours stops, on purpose.

  • It gives estimates and summaries, not advice. It is not your treasurer, your accountant, or a pastoral conversation.
  • It will not write to your records. It can only read.
  • It will not tell you why someone stopped coming. It can only tell you that they did.
  • It cannot see a private care note, and it cannot tell you what an individual gave.
  • It is capped at a monthly question limit, so it can never quietly run up a bill.

What a board will ask you

Can the AI see another church's data?
No, and not because we promise to be careful. The assistant is only allowed to call six read-only summaries of your own records, and none of those summaries has a field for naming a church — so another congregation cannot be requested, because it cannot be addressed. Underneath, the query runs as a restricted database account that returns no rows at all unless your church is set. If that were ever left unset, the answer would be empty rather than someone else's.
Can it read our pastoral care notes?
No. The care summary can count how many cases are open and how long they have been open. The contents of a note — what a family told you in confidence — are never returned to the assistant. Likewise, giving summaries report totals and trends but never identify who gave what.
Could it make up a number?
It is not allowed to do arithmetic. Every figure it quotes is calculated by the same code that draws your dashboard. For the insight panels we go further: any number the model writes that is not one of your real figures causes the whole paragraph to be thrown away and replaced by a plain template. A wrong number cannot reach the page.
Is our congregation's data used to train an AI model?
No. Your question and the summaries needed to answer it are sent to our AI provider to produce the response, and they do not use it to train their models. We keep the question, the answer, and a record of which summaries were consulted, so an elder can audit later exactly what the assistant was told.
Does it cost extra?
No. 10 questions a month are included in the one $19 plan, and the insight panels are always on. There is no AI add-on to buy and no per-question billing.
What happens when it doesn't know?
It says so, and tells you which tab would show the answer. It will not speculate about a family's giving, guess a person's reason for being absent, or invent a policy. A tool that admits ignorance is the only kind worth putting near a congregation's records.

Still uneasy? Write to [email protected] and a person will answer. Our privacy policy spells out exactly what leaves the database.

Judge it against your own records.

Walk the demo to see the insight panels on a real church's week. Then subscribe, and ask the assistant something only your records could answer.

10 questions a month included · no AI add-on · cancel anytime