Your governance programme probably works.
Until Friday afternoon, when the brief is overdue and the client is calling.
That's not a people problem. It's a timing problem. And it's where most agency AI governance fails — not because the rules are wrong, but because the rules arrive at the wrong moment.
The velocity objection is half right
When agency owners tell me governance will slow their teams down, I don't argue with them. Bolt-on governance absolutely does slow teams down. That's documented, not theoretical.
Here's what happens. New process is introduced. Behaviour changes briefly. Deadline pressure builds. The new behaviour erodes — and it erodes first. Not because creative teams are resistant. Because governance layered on top of existing workflows is structurally incompatible with deadline pressure.
Researchers call this compliance decay. A study tracking adherence to a defined protocol over 14 weeks (Howard & Lamb, 2024, Assessment) found compliance falling from 88.9% to 70% by the study's end — with financial incentives in place. The problem wasn't motivation. It was design.
There's a specific mechanism behind this. Research on how healthcare professionals apply clinical guidelines under time pressure (Tsiga et al., 2013, BMJ Open) found something instructive: when physicians were short on time, they didn't drop the most complex guidance first. They dropped whichever guidance required retrieving a separate document.
Your agency's AI policy, sitting in a shared drive, is that separate document.
Where you put it matters more than what's in it
This is the timing insight. It changes the conversation from "is our governance good enough" to "is our governance in the right place."
Grimshaw and colleagues analysed 235 studies comparing how healthcare professionals applied clinical guidelines (Grimshaw et al., 2004, Health Technology Assessment). Passively distributed guidance produced an average improvement of 8.1%. The same guidance, delivered as a prompt at the exact moment of decision, produced improvements of 14.1% to 20%.
Same content. Same professionals. Different placement. A performance difference of 1.7x to 2.5x.
The agency translation is direct. A three-second classification decision before a copywriter opens a tool is one thing. A 30-minute incident review after client data has already entered an unsanctioned system is another. Both are governance. One is pre-decision. One is post-decision. The first costs seconds. The second costs relationships.
Pre-decision clarity. Post-decision bureaucracy. The distinction is where the governance tax gets paid — before the work starts, or after something goes wrong.
What embedded looks like in practice
The Three Simple Rules — Data Traffic Light, Human Wrapper, Prompt Dividend — are decision-point tools, not review frameworks. The difference is when they appear.
For a copywriter, the governance happens before anything enters a tool. What's in this brief? The Data Traffic Light answers in three seconds: Green (public information, any tool), Amber (client-sensitive, enterprise tools with signed DPAs only), Red (personally identifying data, never in any AI tool). Classification is part of the brief intake — a field on the same project card where the brief lives. Not a separate checklist.
The Human Wrapper isn't a review stage added after creative work is complete. It's a commitment made before the draft goes anywhere: reviewed by a named person, on a named date, before it left. The ICO is explicit that rubber-stamping AI outputs doesn't constitute meaningful human oversight. "A human reviewed it" and "a human reviewed, revised, and signed off" are not the same claim.
The Prompt Dividend is the only rule that happens at the end — and it's a capture moment, not a governance overhead. The prompt sequence that matched the client's tone first time. The structure that turned a one-line brief into a usable outline. Added to the shared library because it's faster next time. And the agency owns it, not just the individual.
Three decision points. Under two minutes per piece of work. Governance that doesn't feel like governance because it's already in the workflow.
The tender proof
A 72-hour RFP arrived late on a Thursday afternoon. Major pharmaceutical client. Significant contract. Response required by Monday morning, first thing.
Eight hours.
That's how long the draft took. Not because we were faster than the other agencies. Not because we cut corners. Because the governance was already done. Data handling protocols — written. Case studies with client permissions — secured. Approval chains — mapped and agreed. What should have been 72 hours of scramble became 64 hours of strategy.
Our competitors submitted something rushed. You can tell. Procurement teams can tell.
Early research on AI adoption in UK agencies (Twisted Loop, 2025, n=53) found that agencies with a defined AI lead are significantly more likely to expect accelerating adoption: 79% versus 58% for those without. Structure enables adoption. It doesn't restrict it.
Embedded governance isn't felt as governance. It's just how the work runs.
About the book
This newsletter comes from Shadow AI Governance: The UK Agency Playbook — a book I'm writing in public about making agency AI usage visible, accountable, and commercially defensible. Chapter 10 sits at the heart of the Solution section. The problem is named. The framework is built. This is where the rubber meets the road: governance embedded inside the workflows that already exist. Chapter 11 takes the next step — and it's a harder one.
Want the full chapter?
Chapter 10 lays out the compliance decay curve, the clinical evidence behind pre-decision clarity, and what the Three Simple Rules look like embedded across copywriter, designer, and strategist workflows — role by role.
Ready to build this into your agency's workflows?
The Done-With-You AI Workflow Build is a four-week programme — your tools, your team, your briefs. We embed the Three Simple Rules where your governance actually needs to live: at the decision point, before the tool opens. £3,500. If you're further along and want ongoing AI leadership without the full-time hire, the Fractional AI Leadership retainer starts at £2,500/month.