This is a typical scenario. It shows how Remolda would approach AI support for building permit intake in a mid-size Canadian municipality, what we would build and what we would measure.
The situation
Picture a municipality where building permit applicants wait weeks before anyone looks at the technical content of their file. A residential application arrives as a package of drawings, site plans, structural calculations and supporting documents. A small team checks each package for completeness against the municipal checklist before engineers and inspectors see it.
Many packages come back with a deficiency list. The applicant fixes the gaps and resubmits, and the check starts again. Some files loop two or three times. Meanwhile applicants phone the counter to ask where their file stands, and staff stop reviewing to answer.
The bottleneck is the completeness check. It follows a checklist, so most of it can be supported by AI.
The approach
Audit (3 weeks). We review past permit applications across residential, commercial and heritage categories, sit with the permit staff and map every step from submission to first technical review. Two numbers become the baseline: time from submission to a complete file, and the number of resubmission rounds per file.
The audit also checks whether reviewers read ambiguous checklist items differently. If they do, writing one agreed version of the checklist comes first, and it helps consistency even before any AI is involved.
Strategy (4 weeks). The design has three parts:
- Document classification and extraction. Each document in a package is identified by type, and key values such as dimensions, lot coverage and setbacks are extracted.
- Checklist validation. Extracted values are compared with the checklist for the permit category. The output is a structured report: requirements met, requirements missing, and items that need a person to decide.
- Applicant status assistant. Answers routine questions about status, deficiency items and next steps through the permit portal, and hands anything else to staff.
The permits manager, the planning director and IT security review the design. The system works alongside the existing permit software; replacing it is out of scope.
Implement, phase 1 (2 months): extraction and checklist reports. The engine is configured on the municipality's own past applications. A package that the report shows as complete goes to technical review after a staff member confirms it. A package with gaps gets a draft deficiency notice in plain language, citing the checklist item, which staff edit and send.
For the first month the AI runs in shadow mode. Staff review each file on their own, then compare with the AI report. Disagreements are analyzed and used to tune the configuration. Go-live needs an accuracy level agreed with the permits manager in advance.
Implement, phase 2 (2 months): status assistant. Applicants ask about their file in plain language and get the current status from the permit system. Questions that need professional interpretation, unusual cases and signs of frustration go to a staff member.
Implement, phase 3 (1 month): integration. Status changes, deficiency notices and approvals trigger automatic notifications to applicants.
Empower (parallel). Permit staff learn to read the report's confidence flags, handle ambiguous items and mark wrong assessments. Counter and phone staff learn the assistant's scope and escalation rules.
What we would measure
Each item below is a target we would set with the municipality and measure against the baseline from the audit. The size of each target is set only after the baseline is measured.
- Time from submission to a complete file.
- Resubmission rounds per application, before and after checklist-referenced deficiency notices.
- Staff hours on completeness checks versus complex files and applicant consultations.
- Agreement rate between the AI report and the staff decision, tracked every week.
- Status calls and emails to the permits counter.
Rules that frame this scenario
A permit decision is made by the chief building official and staff; the AI prepares reports and drafts. Ontario municipalities handle applicant information under the Municipal Freedom of Information and Protection of Privacy Act, and a privacy review of the portal and assistant is part of the plan. The government sector page lists the rules and sources we check.
Key lessons
1. The checklist step is where the time goes. Permit software usually tracks files and routes them. The content check that decides whether a file can move forward is the part AI can support.
2. Specific deficiency notices shorten the loop. A notice that names the missing dimension and the bylaw section lets the applicant fix it in one pass. A vague notice leads to another round.
3. Status visibility matters as much as speed. Applicants accept a wait more readily when they can see where their file is. The status assistant and automatic notifications address that directly.
For related work, see AI for provincial and municipal government, document processing and customer support assistants.