Skip to main content
PermitAssure — Permits Made Simple
Strategy & Business Case Framework

A Target Operating Model for an AI-Assisted Building Department

An operating model answers who does what, with what authority, on what evidence. Software cannot answer any of those questions for a department.

Editorial owner
PermitAssure Editorial Team
Published
3 Aug 2026
Last reviewed
3 Aug 2026
Reading time
11 minutes
Audience
Government & AHJs, Building Officials, Technology & Innovation Leaders
Three construction professionals in hard hats reviewing project documents on a building site.
The operating model is mostly about people: roles, authority and the evidence each role works from. Photo: licensed stock imagery (iStock); licence confirmation pending.

Executive summary

Introducing compliance intelligence changes work allocation more than it changes job titles. Administrative checking moves earlier and becomes systematic; reviewer attention concentrates on interpretation, exceptions and decisions; two new capabilities appear — rule stewardship and quality assurance.

This article sets out a target operating model with five layers, the decision rights that belong to each, and the assurance activities that keep the model defensible.

Why an operating model comes before configuration

Configuration encodes decisions about who decides. If those decisions are implicit, the system will make them by default and the department will discover its new operating model after go-live. Guidance on automated decision-making assumes the opposite order: define oversight, then deploy [1][2].

The model also determines whether accountability remains locatable. A finding produced by a rule authored by nobody in particular, applied by nobody in particular, is not usable in a regulatory context.

Five layers of the operating model

Operating model for an AI-assisted building departmentFive layers: statutory authority held by the chief building official and delegates; professional review by plans examiners and specialists; rule stewardship by code specialists who author and validate rules; operations covering intake, triage and coordination; and platform and assurance covering configuration, security, privacy, evaluation and logging.Target operating model layers1. Statutory authorityChief building official and delegates: approval, refusal, conditions, overrides2. Professional reviewPlans examiners and specialists: interpretation, exceptions, findings dispositions3. Rule stewardshipCode specialists: author, test, validate, version and retire rules per edition4. OperationsIntake, completeness, triage, coordination, applicant communication5. Platform and assuranceConfiguration, integration, security, privacy, evaluation, logging
Original PermitAssure diagram. Text description: five layers from statutory authority (chief building official and delegates) through professional review (plans examiners and specialists), rule stewardship (code specialists authoring, testing and versioning rules), operations (intake, completeness, triage, coordination, communication) to platform and assurance (configuration, integration, security, privacy, evaluation, logging).

Rule stewardship is the layer most often missing. Without a named owner for rule content, rules drift out of alignment with the adopted edition and no one is accountable for their correctness.

Decision rights

Table 1 — Decision rights in an AI-assisted building department.
DecisionOwnerConstraint
Issue, refuse or condition a permitChief building official or delegateStatutory; never delegated to software
Disposition of a findingAssigned reviewerMust record accept, modify, reject or escalate with reason
Whether a provision is encoded as a ruleRule steward with professional validationDocumented interpretation plus test cases
Rule publication and retirementRule steward, approved by a designated officialVersioned with effective dates
Triage route and reassignmentOperations, reversible by reviewersReason recorded
Model or platform change acceptancePlatform and assurance with reviewer sign-offRe-evaluation before release

Implications for authorities having jurisdiction

Staffing changes modestly but responsibilities shift. Rule stewardship can often be assigned to an existing code specialist with protected time rather than a new post, but the time has to be real and scheduled, because rule maintenance follows the code cycle.

Assurance activities — evaluation against representative files, accessibility testing, privacy review, log review — belong in the annual operating calendar, not in the project plan only.

Implications for applicants and professionals

A clear operating model gives applicants a route for disagreement: which finding, which reviewer, which escalation path. That clarity matters more to professional applicants than the underlying technology.

It also sets expectations about what a finding is. In this model a finding is an evidenced observation carrying a rule reference; its disposition is a human act recorded against a named reviewer.

Risks, limitations and safeguards

  • Unassigned rule stewardship is the most common structural failure; it produces silently stale rules.
  • Reviewers under time pressure may accept findings without inspection; workload planning is part of oversight design.
  • Concentrating platform knowledge in one person creates a continuity risk; document configuration.
  • Assurance activities are the first thing cut under pressure and the last thing that should be.
  • An operating model on paper that differs from practice is worse than none; audit against actual records.

PermitAssure perspective

PermitAssure supports this separation directly: rule content is versioned with an owner and effective dates, reviewer dispositions and overrides are recorded with reasons, and configuration is exportable so the jurisdiction retains its own definitions. The platform provides evidence; the department retains authority.

Five key takeaways

  • Configuration encodes decisions about authority, so the operating model comes first.
  • Rule stewardship is a distinct, frequently missing role with scheduled time.
  • Every automated contribution needs a named owner of content and a named owner of use.
  • Assurance is an operating calendar activity, not a project phase.
  • Statutory approval and finding dispositions remain human acts, recorded with reasons.

References

  1. Directive on Automated Decision-Making. Treasury Board of Canada Secretariat. www.tbs-sct.canada.ca. Accessed 3 August 2026.
  2. Algorithmic Impact Assessment. Government of Canada. www.canada.ca. Accessed 3 August 2026.
  3. AI Risk Management Framework. National Institute of Standards and Technology. www.nist.gov. Accessed 3 August 2026.
  4. National Building Code of Canada 2025. National Research Council Canada. nrc.canada.ca. Accessed 3 August 2026.
  5. AI RMF Playbook. NIST AI Resource Center. airc.nist.gov. Accessed 3 August 2026.

Cited statements follow the sources above. Frameworks, diagrams and interpretation in this article are PermitAssure's own.

Related resources

Next step

Assign rule stewardship and document the six decision rights above before configuring a single rule.

See the governance approach

PermitAssure provides digital review, workflow and decision-support capabilities. This resource is educational and does not constitute regulatory, legal, architectural or engineering advice. Final interpretations, approvals and regulatory decisions remain the responsibility of the applicable Authority Having Jurisdiction and its authorized professionals.