How to Set Up an Architecture Review Board (ARB) with Sparx EA

An Architecture Review Board is a governance forum that reviews and approves technology decisions above a defined threshold of scope, cost or strategic impact. Its job is to enforce standards, reduce risk and keep new investment aligned with the target state. The most common failure is becoming a bottleneck: when everything goes to the ARB, delivery slows and teams route around it. Effective boards are selective, well structured and quick to return a decision.

Much of what slows a board down is preparation: reading submissions to find what is missing, tracing what a change affects, and writing up the decision afterward. Working AI-augmented, AI does those mechanics in the model and the board’s time goes to the judgment only people can make. That only works if the mandate, the rules and the model are clear, so start with the mandate.

Stand it up in four moves

Why EA teams need a board at all

An ARB’s mandate answers questions that, left open, leave the board powerless:

• What decisions require ARB review? Technology investment above a cost threshold, new vendor or platform introductions, changes to the integration architecture, deviations from approved standards.
• Who has authority to decide? An ARB vote, the chief architect alone, or delegated authority for specific domains.
• What does approval mean? What happens to projects that bypass review, and to unapproved deviations discovered later?

The discipline that makes an ARB valuable is institutional. It needs sponsorship from a CTO, CIO or equivalent with the authority to enforce the mandate. Without that, governance is the first thing dropped when project pressure rises.

Keeping the tiers in place

The tiered model works only if the boundaries hold. Tier 3 is where most of the value lives: projects move faster inside pre-approved guardrails, and the board’s time is kept for decisions that need strategic judgment. When minor tool selections start landing on the Tier 1 agenda, the board slows everything and teams work around it. Police the boundary deliberately.

How Sparx EA holds the governance

Sparx EA holds the board’s governance artifacts alongside the architecture they govern.

Architecture Decision Records. Decisions are modeled as elements with a custom «ArchitectureDecision» stereotype and tagged values for status (Proposed, Under Review, Approved, Superseded, Rejected), date, owner and linked alternatives. That gives a searchable, versioned record you can query, for example to find which decisions affect Application X. (See our primer on the Architecture Decision Record pattern.)

Architecture Principles. The board’s governing standards, such as the approved technology list, integration standards and security requirements, live as architecture principle elements. Submissions reference them, and deviation requests link to the specific principle with the justification documented.

Review status tracking. Each project creates a submission package in the repository, with review status tracked on the submission element. Approved submissions link to the design elements they govern, and rejected ones carry the rationale. Over time this becomes a record of what the board reviewed and decided, visible to every architect.

Running reviews AI-augmented

Preparing the agenda. Today the chair or a domain architect reads each submission to find missing sections, unstated alternatives and conflicts with principles, and sends it back. Working AI-augmented, AI checks each submission against the template and the principle elements in the model before it reaches the agenda, and lists the gaps and the principles it appears to deviate from. The skill is writing the template and principles clearly enough to check against, and judging whether a flagged deviation is justified. Automation checks completeness, people own correctness.

Tracing impact. Today an architect traverses relationships by hand to answer what is affected if the integration middleware is replaced, and the answer depends on how much time they have. Working AI-augmented, the architect asks the question and AI traverses the model and returns the affected applications, interfaces and decisions with the paths it followed. The skill is asking the right question and knowing where the model is incomplete, because AI can only trace relationships that exist.

Recording the decision. Today the decision is written up after the meeting, when someone has time. Working AI-augmented, AI drafts the decision record from the meeting notes, sets its status and links it to the submission and the affected elements. The chair confirms the decision and rationale are stated as the board meant them.

Tools that help: Claude or GitHub Copilot with AI Power Tools for Sparx EA, and Kernaro Assist in the EA client. Which fits depends on who runs the preparation and where they work.

Five anti-patterns to design out

1. Universal scope. When every decision goes to the board, it becomes a bottleneck and teams work around it. Enforce the tiers.

2. Retrospective review. When projects present decisions already implemented, the board becomes a rubber stamp or a source of conflict. Review happens before commitment. If that is hard to enforce, the submission process is too slow.

3. No enforcement mechanism. Advisory-only authority depends on goodwill, which evaporates under pressure. Name an executive sponsor with the authority to enforce.

4. Process over judgment. When the board spends more time on template compliance than on architectural risk, the process is blocking the purpose. Let AI handle completeness and calibrate the template regularly.

5. No feedback loop. When decisions are made but never revisited, the board cannot learn. A brief post-implementation review on significant decisions closes the loop.

FAQ

A well-designed ARB is one of the strongest levers for making architecture matter in an organization. We help architecture leaders design the board and set up the repository structure behind it: decision records, principles and the model that impact analysis depends on. See also AI-augmented governance.