How myLot keeps planners in charge of planning

The planner decides

Every finding myLot produces arrives with its reasoning, the planning control it relied on and the page of the application it came from, hyperlinked to both. The planner can accept it, edit it, dismiss it with a reason, add it to an RFI, set it as a condition or refer it.

Accepted findings flow into reports and correspondence. Dismissed ones are recorded with the officer’s explanation. Some judgements are never delegated. Whether an application contains sufficient information to proceed is a statutory call, and myLot presents the evidence without making it. The same applies to every recommendation in an assessment report: Assess drafts, the officer finalises, and council makes the decision.

A record you can rely on

All platform activity is written to an append-only event log. Records cannot be overwritten, so council holds a complete history of what the AI proposed, what information it relied on and what the planner decided. Planning controls are held in a versioned library, so you can establish which version of a control was applied to any application at any time.

If a decision is challenged, we can reconstruct the output, the reasoning, the configuration and the provisions in force at the time, and who reviewed it. The full record is available to council on request, in open formats.

Governance you can read

Our AI Governance Policy and AI Privacy Policy are published in the myLot trust centre alongside our security policies. They are consistent with state and national privacy law.

Frameworks we map to

Both policies are mapped to the frameworks Australian councils are asked to work within:

• NSW AI Assessment Framework
• Victorian Government guidance on generative AI
• Australia’s AI Ethics Principles
• National Framework for the Assurance of AI in Government.

Alignment, not certification

We describe this as alignment and assessment, not certification, because that is what it is.

How we keep the AI honest

Accuracy starts with what the model is not asked to do. myLot’s AI never uses a language model’s general knowledge of planning. It reasons only over the specific controls for the site and the specific documents in the application, so it cannot introduce planning advice with no source.

Deterministic where it counts

Each task is handled by a narrowly scoped AI role that sees only what it needs. Calculations and spatial tests run in deterministic code, not in the AI. And every process is measured continuously against expected outputs by an automated evaluation layer, reviewed weekly by our planners and engineers.

Feedback that changes the product

When an officer overrides or flags an AI output, the platform records the override and asks for a short explanation. Every flag becomes a logged issue our planning and AI engineering team works through.

Officer accept and dismiss decisions are recorded, which gives a continuous measure of the platform’s accuracy in your council, not just in ours.