Planning and zoning are one of the most data-intensive, high-stakes areas of local government work. Yesterday’s AI In Action was built around that reality. Erica Olsen, CEO of Madison AI, and Kristine Richter, Head of Client Success, walked through the planning model, showed real use cases using live data, and shared what six months of work with planning teams has taught us about what it takes to get this right.
Most AI tools weren't built for planning work. This one was. Share this session with your community development team. It was built with them in mind.
AI for planning and zoning is only as good as the data behind it. The planning model pulls from GIS, permitting systems, case files, master plans, archival documents, and more - some going back 50 years. Unlike other areas of government work, planning requires a complete record. An 80% answer is not acceptable - the stakes are too high.
This section covers how the model is structured, what data sources it draws from, and why completeness is the standard, not a stretch goal.
Erica Olsen, CEO, Madison AI
A parcel lookup sounds simple. What's happening underneath it is not. This use case pulls from GIS, permitting records, zoning case files, and master plans, and surfaces the results in a structured, skimmable format with clearly cited sources. Planners can see exactly where data came from, flag inconsistencies, and assess coverage before acting on the output. The goal is not to replace judgments. It is to give staff the complete picture, so their judgment is grounded in full record.
This section also covers the GIS summary view - a feature that started as an internal tool and turned out to be especially useful for non-planners like city managers, admins, and council members who need quick property context without navigating ESRI/GIS layers.
Erica Olsen, CEO, Madison AI
A zoning verification letter looks like a simple, one-page document. Producing an accurate ZVL requires pulling from every relevant data source in the index. This section walks through how Madison AI takes the same parcel research and structures it into a letter format that meets a department's template requirements - with a human-in-the-loop review step built in.
The session also introduces a completeness-scoring feature in development. Before a planner finalizes the letter, AI evaluates how complete the output is against the required template - flagging gaps before the document leaves the desk.
Erica Olsen, CEO, Madison AI
In Golden, Colorado, every building project requires a full planning review and a report to the council. Site plan reviews were taking staff several days to complete - and the review itself could stretch to two weeks.
This section demonstrates how Madison AI processes an uploaded site plan against the development code, produces a structured compliance report, and flags whether each requirement is met or not.
The AI analysis is constrained; it only draws from the specific code applicable to that development type, not the whole dataset. That constraint is what makes the output defensible. The result is full review in under a day, with the staff member reviewing and confirming rather than building from scratch.
Kristine Richter, Head of Customer Success, Madison AI
