AI Assurance — how TownSuite Intelligence is built to be trustworthy

TownSuite Intelligence is the AI assistant built into TownSuite mERP®. Municipal IT leaders, CISOs and procurement teams are right to ask hard questions of any AI: where does it run, whose law governs it, what does it do with our data, and can we trust what it says? This page sets out, in plain language, how TownSuite Intelligence is engineered, hosted and reviewed so that its answers can be trusted — and is candid about the one case where the sovereignty guarantee does not apply.

Our own AI, in Canada

TownSuite Intelligence is TownSuite's own AI inference — not a bolted-on foreign cloud service. It runs on your municipality's own data inside a Canadian-operated datacentre (Cirrus, Micrologic's cloud) under contractual operational sovereignty.

• Your data is not sent to any foreign AI service.
• Encryption keys are held in Canada.
• Your data is never used to train AI models.
• Answers are generated only from your connected records — not from a general model's memory of the wider internet.

This is the same operational-sovereignty model that governs the rest of the TownSuite platform: the systems and the company operating them answer to Canadian law alone.

Designed not to fabricate

The biggest worry about AI in a financial or regulatory setting is that it will state something with confidence that simply isn't true. No AI can be guaranteed perfect — so rather than promise that, we have engineered TownSuite Intelligence specifically to reduce that risk as far as we can. The guardrails are built into how it produces every answer:

• The model's job is to shape, aggregate and narrate your data. A separate deterministic engine does all of the formatting — the model never hand-draws a number or a chart.
• It must pass raw numbers through as they are, never inventing precision that the underlying data does not have.
• Missing data is shown as a gap, never silently filled in as a fake zero.
• Every figure carries its source and as-of date (provenance), so you can trace where it came from.
• A validator checks each output before it is rendered.
• If an output is bad, it degrades to a safe placeholder rather than presenting a confident-but-wrong chart.

Put plainly: it is built to say “I don't have that” rather than make something up.

Reviewed by people and AI

The reporting engine behind TownSuite Intelligence went through a documented internal review process combining human and AI review. A multi-reviewer process examined it through QA, completeness, critic and accessibility lenses. QA and accessibility were then re-run in a loop until both signed off, and malformed-input testing was carried out across every chart type to confirm the safe-degradation behaviour holds up under bad data.

Verified behaviour

In independent testing, the assistant behaved as designed:

• It gave grounded answers drawn from the connected data.
• It declined to invent figures or customer names it did not have.
• It refused prompt-injection attempts.
• It answered data-sovereignty questions consistently.

The assistant also shows a standing disclaimer that it can make mistakes — verify important figures. That is by design: it is meant to assist judgment, not replace it.

We are candid about this: no AI is ever 100% accurate. We have made — and keep making — many optimisations to keep its answers accurate and well-grounded, but we cannot guarantee that every output is perfect. Because every figure carries its source and as-of date, important numbers can and should be checked against the underlying records before you rely on them.

Bring-your-own-model (optional)

By default, TownSuite Intelligence runs only on models TownSuite hosts in its Canadian datacentres inside the sovereign environment described above. A municipality may optionally connect its own external model. We are transparent about the trade-off: in that configuration, that traffic leaves the sovereign environment, and sovereignty no longer applies to it. If keeping everything under Canadian operational sovereignty matters to you, the default configuration is the one to choose.

For procurement & security teams

We know AI assurance is now part of municipal due diligence. Our security and AI assurance documentation — including architecture, data-flow detail and an independent assessment — is available on request, under NDA. For the broader sovereignty, hosting and security picture, see our resources for procurement & IT and our Privacy Policy.

Frequently asked

What AI model powers TownSuite Intelligence?

TownSuite Intelligence runs on TownSuite's own AI inference — running one or more models we host inside a Canadian datacentre (Cirrus, Micrologic's cloud) under contractual operational sovereignty. It is not a bolted-on foreign cloud service, and your data is not sent to any foreign AI service.

Does TownSuite's AI make up numbers?

It is engineered to avoid fabricating, and these guardrails make it rare — though, like any AI, it cannot be guaranteed perfect. A deterministic engine does all formatting while the model only shapes, aggregates and narrates your real data. It must pass raw numbers without inventing precision, missing data is shown as a gap rather than a fake zero, every figure carries its source and as-of date, and a validator checks each output before it renders. A bad output degrades to a safe placeholder rather than a confident-but-wrong chart. Put plainly, it is built to say “I don't have that” rather than make something up — but important figures should still be verified against the source.

Is TownSuite Intelligence always 100% accurate?

No — no AI system is. We have made many optimisations to keep its answers accurate and well-grounded, and in testing it behaves as designed, but we cannot guarantee every output is perfect. It carries a standing disclaimer to that effect, and every figure shows its source and as-of date so important numbers can be verified against your real records.

Is my data used to train AI models?

No. Your municipality's data is never used to train AI models. Answers are generated only from your connected records, and encryption keys are held in Canada.

Does my data leave Canada or go to a third-party AI service?

No. By default TownSuite Intelligence runs only on models TownSuite hosts inside its Canadian datacentre, and data is not sent to any foreign AI service. The only exception is if your municipality opts into connecting its own external model (bring-your-own-model), in which case that traffic leaves the sovereign environment and sovereignty no longer applies to it.

Has the AI been tested and reviewed?

Yes. The reporting engine went through a documented internal review process using multiple reviewers across QA, completeness, critic and accessibility lenses, with QA and accessibility re-run in a loop until both signed off and malformed-input testing across every chart type. In independent testing the assistant gave grounded answers, declined to invent figures or customer names, refused prompt-injection attempts, and answered data-sovereignty questions consistently. It also shows a standing disclaimer that it can make mistakes and that important figures should be verified.