How AI works here
Every AI output is cited, grounded in your own records, and waits for a person to approve it. What that means in practice, and where the limits are.
ManageStrata uses AI in several places — answering questions, drafting minutes, proposing a budget, triaging repairs, watching deadlines. All of it follows the same three rules.
The three rules
It is grounded in your data
Answers come from your corporation's own records and your own uploaded documents — not from the internet, not from general knowledge about strata law, not from another corporation. If your documents don't say, it says it doesn't know rather than filling the gap.
It cites its source
An answer drawn from a document names the file and the page. You can open it and check. For anything you're going to act on, do.
A person approves before anything happens
AI drafts, proposes and flags. It never files, sends, spends or approves. Minutes wait for a council member. A budget proposal waits for the council and then the owners. An agent's suggestion is added or dismissed by a human.
Nothing is auto-filed
There is no path by which the AI posts a transaction, sends an email to owners, issues a levy, closes a work order, or finalizes a vote. Those are all human actions, and the audit trail records which human.
Where AI appears
| Feature | What it does | Who approves |
|---|---|---|
| Ask the documents | Answers from your library, page-cited | You read it |
| Red flags | Scans for risks in your documents | Council reviews each |
| The Assistant | Answers across records and documents | You read it |
| Minutes drafting | Drafts from agenda and results | A council member approves |
| Budget proposal | Drafts next year's budget | Council, then owners |
| AI agents | Watch records, propose actions | Each suggestion added or dismissed |
| AI vendor match / research | Suggests and researches contractors | You verify, always |
Why grounding matters more than cleverness
A general-purpose AI can tell you what strata legislation usually says. That's close to worthless when the question is whether your bylaws allow a dog, because your bylaws are a specific document that may say something unusual.
Grounding is the constraint that makes the answers useful: the model is only allowed to answer from what your corporation actually holds. It's also why the most valuable thing you can do for AI quality here has nothing to do with AI — it's uploading your documents.
What it still gets wrong
Be a careful reader. Realistically:
It can misread a badly scanned page. A photographed document is read less reliably than a digital original.
It can miss context it was never given. A budget draft doesn't know your insurance renewal came in 22% high, or that you're changing landscapers.
It can summarise a nuanced passage too confidently. The citation is there so you can check the actual words.
It doesn't know what happened in the room. Minutes drafted from the agenda and the vote counts won't contain the discussion. That's your job to add.
Proposed is not decided
Worth its own heading, because it was a real failure and the fix is a rule you can hold the answer to.
Asked about an EV-charging resolution, the assistant read the AGM notice of call — the agenda — and reported that the resolution "passed, authorizing up to $20,000 from the Contingency Reserve Fund". The minutes of that meeting were not in the library, and the resolution had in fact been defeated. A notice reads exactly like an outcome — "BE IT RESOLVED THAT the Owners approve…" — and nothing in the excerpt itself says which it is.
Both grounded surfaces — the Assistant and Ask the documents — now hold the same three rules:
- a notice, an agenda or a proposed resolution records what was put forward. A proposed budget or levy is not an approved one;
- only minutes, or an excerpt stating the vote or the result, establish that something was approved, passed, defeated or authorized;
- where a resolution was proposed and nothing records the vote, it says so, names the minutes of that meeting as where the outcome would be, and does not soften that into "was approved" or "appears to have passed".
This is the argument for uploading your minutes
The rule turns a confidently wrong answer into an honest gap, which is the right trade — but it is still a gap. Telling owners the corporation holds a spending authority it never granted is the worst thing this assistant can do; being unable to answer at all is merely unhelpful. Get the minutes into the library and it can answer properly.
Treat it as a well-briefed assistant
Fast, thorough, and reads everything you gave it. Not present at your meetings, not a lawyer, and not accountable for the decision. Check what matters; the citations make that quick.
What it will never do
- Give legal advice. It reports what your documents say; whether that's enforceable is a lawyer's question.
- Move money, or change anyone's balance.
- Send anything to owners.
- Approve minutes, finalize a vote, or close a period.
- See data your login can't see. An owner asking the Assistant about arrears gets their own account, not the building's.
Privacy
Your data is stored in a Canadian region under PIPA/PIPEDA. Some AI processing occurs outside Canada under contract — that's a real limitation and worth knowing rather than glossing. Details are in privacy and data and in the privacy policy.
Think about what you upload. Legal advice, personal information about an owner, an unresolved dispute file — all need care about who can reach them, quite apart from AI.
Turning it off
If your council would rather not use the AI features, don't. Every core function — compliance tracking, meetings, voting, the books, documents, maintenance — works without touching an AI feature. The compliance calendar isn't AI; neither is the ledger, the voting, or the reserve projection.
Cost
AI usage is metered in AI actions, with a monthly allowance on every plan. Nothing is charged per question unexpectedly — see plans and billing.
Still stuck? Open Support in the top bar of the app, ask the Assistant, or contact us.