If your Agency of Record relationship is starting to feel like a production treadmill, you’re not alone. The model wasn’t designed for a world where output can be generated quickly, but consistency, governance, and adoption are the constraints.

This guide is a practical path from AOR (outsourced delivery) to a fractional / embedded Center of Excellence (CoE) — without a “big bang” reorg, and without freezing the roadmap.

Dave Greene (Head of Client Relations):

“Most transformations fail because they try to change the org chart before they change the operating conditions.”

What changes in the migration

AOR mode looks like

• Work arrives as requests
• Output leaves as files
• Quality is inspected at the end
• The “system” lives in people’s heads
• Success is measured in effort (hours) and deliverables

CoE mode looks like

• Work arrives through an intake system
• Output leaves through governed pipelines
• Quality is engineered into templates, tokens, and checks
• The system lives in a source of truth
• Success is measured in repeatability, throughput, and adoption

Figma describes design tokens as providing a “single source of truth” to maintain consistency between design and code.

Step 0: The 10-minute “Are we CoE-ready?” test

You’re ready to migrate if at least 5 of these are true:

1. People regularly ask: “Where’s the latest version?”
2. You have a design system… but teams don’t trust it (or don’t use it).
3. Campaign rollouts trigger reformatting marathons and regional drift.
4. Approvals are slow because ownership is unclear.
5. Your best people spend time on production instead of direction.
6. Work gets redone because the “brief” changes after the first draft.
7. Your asset library is a pile, not a system.
8. AI is showing up — but you don’t have guardrails to prevent brand drift.
9. Your agency relationship is “fine,” but you can feel dependency risk.
10. You’re entering change (growth, rebrand, acquisition) and creative becomes the bottleneck.

Rebecca Clarke (Head of Production):

“If delivery depends on a few people remembering how things work, you’re already in a risk state — you just haven’t priced it in yet.”

The migration in 3 phases

Phase 1: Stabilise

Goal: keep delivery moving, reduce thrash, create a minimum viable operating layer.

What you do

• Create a single intake lane (even if it’s simple): one request form, one queue, one triage owner
• Define a “Definition of Done” for your highest-volume asset types (banner, email, deck slide, social, landing page section)
• Start a lightweight asset taxonomy (naming + folders + metadata rules)
• Stand up the first “truth layer” (even if imperfect): approved logos, core colour + type, current templates, what not to use
  

What you stop doing

• Ad-hoc DMs as the real intake system
• Net-new custom builds for things that should be templated
  

Outputs that show you’ve stabilised

• A working intake system
• A basic source-of-truth space
• 3–5 high-volume templates actually in production

Walter Wynne (Head of Strategy):

“The first win isn’t prettier work. It’s fewer reinventions.”

Phase 2: Codify

Goal: turn the brand from “guidance” into infrastructure.

This is the phase where “design system” stops being a library and becomes a control plane.

A strong external reference for the “why”: Material Design explains that tokens make it possible for a design system to have a “single source of truth” — a repository where style choices are recorded and changes can be tracked.

What you build

• A source of truth that teams can trust (and that has an owner)
  • Tokens, components, templates, patterns, rules
    
• A starter token set (start small, scale fast)
  • semantic colour tokens (Primary/Accent/Surface/Border)
  • type scale + usage rules
  • spacing rhythm
    
• Templates that bake in constraints
  • layout rules, safe areas, content limits, accessibility defaults
    
• A governance loop that produces decisions
  • exception review → decision → system update → release notes
    

Outputs that show you’ve codified

• A credible source of truth
• Token starter set
• Core templates/components that people actually reuse
• A governance cadence that creates system releases (not just meetings)

Eduardo Mira (Head of Creative):

“Tokenisation isn’t design theatre. It’s how the brand becomes repeatable at speed.”

Phase 3: Scale

Goal: increase throughput without increasing chaos.

This is where you earn operational sovereignty: the work keeps shipping, even as teams, markets, and priorities change.

What you introduce

• Workflow automation (routing + QA + approvals)
• Role clarity (who owns what, who approves what, who maintains what)
• Production lanes
  • “fast lane” templated work
  • “custom lane” for high-stakes creative
• Enablement
  • onboarding for any team touching brand assets
  • “how we work” playbook

Outputs that show you’ve scaled

• Multiple teams self-serving safely
• Cycle time drops, revisions drop, reuse rises
• Exceptions decrease over time (because the system is getting stronger)

Dave Greene (Head of Client Relations):

“The goal isn’t to deliver faster once. It’s to build a system that keeps delivering fast when everything changes.”

A design-specific point about “operational sovereignty”

If you want to justify the “sovereignty” language in a design-native way: the industry is literally standardising the token layer so teams aren’t trapped by a single tool or vendor.

• The W3C Design Tokens Community Group exists to create standards that design tools and products can rely on for sharing design-system styling “at scale.”
  
• In late 2025, they announced the specification reaching a first stable version, explicitly emphasising that an open standard means no single vendor controls the format, giving teams freedom of tooling without compatibility concerns.

That’s your “operational sovereignty” bridge: your brand rules become portable infrastructure, not a set of fragile files.

Why this matters even more with AI

If you’re adopting AI inside creative workflows, a source of truth stops being “nice” and becomes safety infrastructure.

Atlassian put it bluntly in the context of AI-driven workflows: a single source of truth for tokens, components, and patterns accelerates adoption and reduces repetitive generations/compositions.

Rebecca Clarke (Head of Production):

“Automation without governance is just faster confusion.”

The KPIs that prove the migration is working

Measure the shift away from effort and toward operational value:

• Cycle time: request → delivered
• Revision load: average rounds per asset
• Reuse rate: % produced via templates/components/tokens
• Adoption rate: active users of the system / truth layer
• Exception rate: how often work breaks the system
• Backlog burn: how quickly design debt is reduced

In other words: time-to-deliver as a managed asset.

Where Perpetual fits

Perpetual’s role in this migration is not “more hands.”

It’s building and running the operating layer: source of truth, tokenisation, workflow automation, governance, and high-frequency production — so your teams (and your AI) can move fast without drift.

Walter Wynne (Head of Strategy):

“This is the junction. The wrong path is more output with the same chaos. The right path is creative infrastructure — operational sovereignty at scale.”

If you’re currently in AOR mode and you can feel the strain, the upgrade isn’t a bigger agency.

It’s a better operating model.