Guide · Seuil Digital

Design System Governance Should Make Change Easier, Not Add a Committee

A checkout button is fixed in the product, but three marketing sites still use copied versions. Nobody knows who can retire them or whether the replacement supports every state. Design system governance should answer those questions with named owners, evidence and a release path, not another recurring meeting.

7 min read
The governance test

A team should be able to propose, approve, release, adopt and retire a component without guessing who decides next.

Design system governance begins with decision rights

Name one accountable owner for each decision: component acceptance, design tokens, code release, accessibility verification, documentation and deprecation. Contributors can supply evidence and challenge a choice, but one role must close it. Shared ownership often means nobody can say yes before a deadline.

Imagine product design proposes a compact input while engineering already maintains two similar variants. The component owner checks the use cases, the accessibility owner tests labels and errors, and the release owner plans compatibility. The product lead should not need to assemble a new committee for that one input.

W3C guidance recommends assigning accessibility responsibilities to specific people and spreading relevant tasks across design, development, content and quality assurance. Apply the same clarity to the whole system. Ownership is a delivery responsibility, not a decorative name in a slide.

Start with a design system audit when the team cannot yet map Figma, code and live product usage. Governance cannot repair a system whose actual components and forks remain unknown.

Components need visible states before they need more documentation

Use a short lifecycle that everyone can see: proposed, supported, deprecated and retired. Each state should change what teams may use, what maintainers promise and what evidence is required next. A catalogue label that changes nothing in delivery is merely stationery.

A proposed date picker might be safe for one internal pilot but not for every product. Mark its tested contexts, missing browser coverage and owner. When it becomes supported, publish the intended use, states, code version and support route. When deprecated, name the replacement and migration trigger.

The GOV.UK Design System contribution criteria require proposed patterns to be useful and unique, then review whether they are usable, consistent and versatile before publication. They ask for representative user research, including people with disabilities, plus testing across browsers, assistive technologies and devices. Your evidence can be smaller, but it should still match the risk and reach of the component.

Each component state should unlock a clear action
StateEvidence to showTeam action
ProposedUser need, existing alternatives and named ownerTest in a bounded context
SupportedApproved design, code, states, tests and guidanceAdopt for documented uses
DeprecatedReason, replacement, affected products and migration pathStop new adoption and plan migration
RetiredConsumers migrated or explicitly accepted as exceptionsRemove support and archive guidance

A contribution path must reject duplicates early

Ask for the problem, affected users, existing workarounds and proof that the system does not already solve it. Rejecting a duplicate before design and code are polished saves more time than speeding up the final review.

Suppose two teams request a new status badge. One needs a visual label; the other needs an interactive filter. A short proposal exposes that these are different patterns before both teams build incompatible components called `Badge`.

The U.S. Web Design System begins new component suggestions with a discussion and asks contributors to check existing proposals first. GOV.UK also separates proposal evidence from publication criteria. Use that sequence: confirm the gap, test the pattern, review implementation, then release it with guidance.

Set review checkpoints rather than invented universal deadlines. A copy change and a new date input do not deserve the same evidence or calendar. The owner should publish what is missing, who decides and what happens if the contribution pauses.

Exceptions should expire or become supported variants

Local exceptions are sometimes correct. A regulated flow, embedded partner surface or unusual device can need behavior the shared component does not support. The risk begins when the exception has no owner, reason or return date.

Record the product, component, reason, user consequence, owner and review trigger. If three products request the same exception, evaluate a supported variant. If the original constraint disappears, remove the fork instead of preserving it as product folklore.

For example, a regional checkout may need additional consent copy. That can be a documented content slot rather than a cloned checkout component. The decision keeps the shared behavior testable while respecting the real constraint.

When exceptions repeatedly involve focus, error handling or assistive technology, connect them to a website accessibility remediation plan. Fix the shared failure before each product pays for the same repair.

Releases need compatibility and migration evidence

A design system release is a product release. Publish what changed, who is affected, whether adoption is automatic and how to verify the result. A new package version without a migration note pushes the decision onto every consuming team.

Semantic Versioning distinguishes incompatible changes, backward-compatible features and backward-compatible fixes through major, minor and patch versions. That convention only works when the public interface is defined and the release notes tell teams what to do.

Imagine changing the default spacing inside every dialog. The code API may remain compatible while screenshots, scroll behavior and translation layouts change. List those product effects, test representative consumers and provide a rollback point before broad adoption.

Separate governance from routine operational care. A website maintenance retainer may ship upgrades and monitor regressions, while governance still decides which change belongs in the system and who accepts its consequences.

Small teams need less ceremony, not less governance

A five-person product team does not need a design system council. It still needs one owner, one proposal template, one release record and one place for exceptions. Keep the mechanism proportional to how many products and contributors depend on it.

The serious objection is speed. Review can become a queue that encourages teams to copy components. Publish the smallest safe path for low-risk fixes, let maintainers reject incomplete proposals quickly and reserve deeper review for changes with wide or irreversible effects.

If the system work expands into new journeys, research and interface architecture, clarify the engagement with the product design agency buying guide. Governance cannot substitute for product discovery, and product design should not quietly invent a second system.

Design system governance buying checklist

  • Every decision has one accountable owner and visible contributors.
  • Components move through proposed, supported, deprecated and retired states.
  • Contribution criteria ask for need, uniqueness, tested states and support responsibility.
  • Accessibility, content and engineering evidence appear before broad support.
  • Exceptions record a reason, owner and review trigger.
  • Releases explain compatibility, product effects, migration and rollback.
  • The engagement includes adoption support and a handover, not only new documentation.

Send Seuil one recurring component conflict, the teams involved and the decision that keeps stalling. Email bonjour@seuil.digital to scope the smallest governance model that can unblock real delivery.

Sources and further reading

Continue readingDesign System Audit: Scope, Deliverables and Buying Criteria