// Playbook 03 — Governance & PMO design
Building a Continuous Improvement Centre of Excellence.
A six-week playbook for turning scattered improvement ideas into a working portfolio of 50+ prioritised initiatives — without adding meetings, dashboards, or overhead. Worked example throughout: Halcyon Financial, a 700-person regulated fintech.
the system
the portfolio
· 14 steps
or status meetings
Meet Halcyon Financial.
Worked exampleA 700-person fintech with more ideas than process.
Halcyon Financial is a regulated digital financial services platform — the kind of mid-cap fintech that has scaled past the start-up phase but hasn't yet built the operating muscle of a mature enterprise. Eighteen months into a transformation programme in the AI era, and improvement ideas are surfacing faster than anyone can track them.
The Customer Experience team has a recurring problem. KYC documents quietly expire and the platform blocks transactions without warning. Twenty-plus clients a week. Two to three days of held funds per case. Twenty-five minutes of CX time per ticket.
Everyone in CX knows. No one has a way to raise it that goes anywhere.
That is the kind of idea this playbook is built to capture, assess, prioritise, and ship — alongside fifty others that look just like it.
the business
programme
surfaced in 60 days
to triage them
// And the people you'll meet
The Transformation Lead
You, the readerAccountable for nothing. Responsible for the programme glue. The role this playbook is written for.
Maya Chen
CX Officer · 2.5 yrsSubmits the KYC-notification idea from a tram on a Wednesday morning. The protagonist of the intake.
Tom Nguyen
CX Team LeadMaya's manager. Pairs with the Transformation Lead to score the intake in Phase 2.
Priya Nair
Head of CXSits between Maya's frontline reality and the executive. Owns the CX seat at the monthly forum.
Sam Patel
Engineering DirectorThe forum's reality check on what can actually get built, when, and by whom.
Anna Petrović
Head of ComplianceHolds the regulator-tolerance line. The forum's veto on anything customer-facing.
Four phases. Capture, assess, prioritise, sequence.
The framework// The framework is deliberately small. Most CoEs fail not because they lack process but because they have too much of it.
Capture
Weeks 1–2One lightweight intake. Anyone can submit. Standard structure so ideas can be compared.
Assess
Weeks 2–3Score every idea on effort and impact using a published rubric. Triage out the low-value noise.
Prioritise
Week 4A standing forum where leaders make calls together. Transparent criteria, recorded decisions.
Sequence
Weeks 5–6A rolling roadmap with resource limits in plain sight. The portfolio becomes visible to everyone.
Phase one · Capture.
Weeks 1–2The goal of the first phase is volume with structure. Make submission easy. Make the shape of a good submission obvious. Worry about quality control in Phase 2. Before you fix prioritisation, you have to see what you're prioritising.
Publish the intake form.
Five fields, no more — problem, proposed change, beneficiary, rough effort, expected impact. A submitter should be able to complete it in under five minutes. Anything longer breaks volume.
Make it open to everyone.
Open intake signals trust. Restricted intake signals the opposite — and ideas just route around it. The cost of a low-quality submission is a one-line "thanks, not now"; the cost of a closed door is the idea you never hear.
Set the response SLA publicly.
Commit to acknowledgement within 5 business days and a triage decision within 15. Publish it. The biggest reason intake systems die is silence — people stop submitting because they assume nothing happens.
Seed it with real ideas.
Don't launch into emptiness. Hand-collect 10–15 ideas from existing retros, surveys, and 1:1 notes. Submit them yourself. An intake board with momentum invites momentum.
Twenty-three ideas in the first week.
The Transformation Lead publishes the intake form on a Monday. Five fields. By Friday: 23 ideas in the tracker — twelve seeded from CX retros and 1:1 notes, eleven from cold submissions across CX, Compliance, and Engineering.
Maya Chen, a CX Officer at Halcyon, submits the KYC-notification idea from her tram one Wednesday morning. Idea #28. Six weeks later, the tracker holds 67 ideas across five teams. All triaged. All visible.
- The idea"KYC documents expire and we don't tell anyone until a deposit fails. Same thing four to five times a week."
- BeneficiaryClients · CX team · Compliance
- Rough effortNotification logic + customer comms · ~4–6 weeks
- Expected impact20+ tickets/week eliminated · NPS lift · faster funds
At Halcyon Financial, the breakthrough came when the intake was made fully visible — not just the form, but every submission, including those that had been declined. Transparency turned submitters into reviewers. People stopped re-raising rejected ideas because they could see why those ideas had been rejected.
Phase two · Assess.
Weeks 2–3A simple effort × impact rubric removes most of the politics from prioritisation. Score on a 1–5 scale across both axes, published rubric, applied by the same small team for consistency. The goal isn't precision — it's a defensible relative ranking.
Define the rubric.
Effort: hours, weeks, months. Impact: tied to a strategic objective with a measurable proxy. Write the descriptions for each 1–5 level. The descriptions matter more than the scale.
Score in pairs.
Single scorers drift. Two scorers per item, compare, discuss disagreements. Disagreements are where the real conversation lives.
Plot on the 2×2.
High impact / low effort → do now. High impact / high effort → plan. Low impact / low effort → maybe. Low impact / high effort → decline, with the reasoning recorded.
Maya's idea scores two and four.
The Transformation Lead pairs with Tom Nguyen (Maya's CX Team Lead) to score the KYC-notification idea.
Effort: 2 of 5. The platform already tracks expiry dates — the work is notification logic and a customer comms flow. Engineering estimates four to six weeks of build, one week of CX rollout.
Impact: 4 of 5. Twenty-plus tickets a week × twenty-five minutes × NPS lift. A reasonable proxy on time-back per CX officer, and a measurable reduction in failed-deposit complaints within sixty days of launch.
Plotted: top-left. Quick wins quadrant. Recorded in the decline log: "Bespoke regional reporting" — declined with reasoning ("high build cost; addressable through existing self-serve BI").
Scoring drift toward the middle. When everything ends up scored 3/5, the rubric isn't doing its job. Force the scorers to defend a 1 or a 5 — discomfort there is a sign the rubric is working.
Before an idea is scored on effort and impact, sort it by what kind of problem it is. Obvious — known cause, known fix, just do it. Complicated — analysis required, but a right answer exists. Complex — cause and effect only visible in retrospect; probe with a small pilot first. Chaotic — stabilise before improving. Most AI-era improvement work is genuinely complex, which means best-practice playbooks will mislead. Maya's KYC-notification idea is complicated, not complex — the fix is known, the question is build effort. Most "AI tool for X" ideas are complex. Sorting up front protects the rubric from being asked to score things it can't fairly score. From Dave Snowden's Cynefin framework — covered in the methodology page, Section 03.
Resulting. When the forum re-scores at the quarterly reset, the temptation is to downgrade initiatives that turned out badly and promote ones that worked. Don't. A good call can still have a bad result. Score the reasoning available when the call was made, not the outcome. Otherwise the rubric quietly becomes a record of which initiatives got lucky — and the next call gets worse. From Annie Duke's Thinking in Bets — also threaded through Phase 3 and Phase 4.
Phase three · Prioritise.
Week 4The prioritisation forum is the system's heart. Standing cadence, fixed attendees, fixed agenda, transparent decisions. Without it, scoring is just commentary.
Stand up the forum.
60 minutes, monthly. 5–8 decision-makers — never more. Mix functions: an engineering lead, a commercial lead, an ops lead, a customer lead. Diversity beats hierarchy.
Use a fixed agenda.
15 min: new submissions reviewed. 25 min: re-prioritise the top 20. 15 min: confirm what's starting this month. 5 min: parking lot. Predictability lets attendees prep — and lets them leave.
Publish decisions within 48 hours.
Decisions, not minutes. Three lines per item: what was decided, why, who owns the next step. If the forum can't be summarised this way, it didn't make decisions.
November forum · seven seats, one hour.
Seven attendees. Transformation Lead in the chair. Priya Nair (Head of CX) brings the KYC-notification idea forward. Anna Petrović (Head of Compliance) presses on regulator-tolerance. Sam Patel (Engineering Director) commits the build slot. The heads of Customer Success, Partner Operations, and People & Culture round out the room.
The KYC-notification idea moves from the "quick wins" pile to "starts this month." A larger strategic bet — AI-assisted ticket triage across CX — gets debated for fifteen minutes and lands in "Plan & Fund" with a budget request tabled at the quarterly reset.
The decision log entry, published within 48 hours, reads in three lines:
- WhatKYC expiry notification flow · approved to start November
- WhyQuick win, measurable on CX time-back and failed-deposit complaints
- WhoEngineering owns build · CX owns rollout · Compliance signs off comms
Resist the urge to add observers. Every additional person in the room reduces decision velocity. If a stakeholder wants visibility, give them the decision log — not a seat.
The forum's job is accuracy, not agreement. Three small habits do most of the work. One — state confidence levels. When someone advocates for an idea, ask how confident they are on a 1–5 scale. "I'm a 4 on impact, a 2 on effort" reveals where the real debate is. Two — reward dissent. The most useful person in the room is the one prepared to argue against the popular call. Name that explicitly in the charter; otherwise the room defaults to consensus. Three — separate the call from the outcome. Decisions that turned out badly aren't re-litigated. Decisions that turned out well aren't celebrated. The system being judged is the reasoning process, not a streak of luck. Adapted from Annie Duke's truth-seeking groups — covered in the methodology page, Section 03.
Phase four · Sequence.
Weeks 5–6Prioritisation without visible sequencing produces a list. Sequencing turns a list into a portfolio — with capacity limits that make trade-offs honest. Everyone should be able to see what's running, what's queued, and what's been declined, without asking anyone.
Set WIP limits by team.
A team can run two initiatives at once. Maybe three. Never six. WIP limits aren't bureaucracy — they're the only thing that protects throughput.
Build a rolling 90-day view.
Quarterly horizon, monthly updates. Anything further than 90 days is direction, not commitment — and should be labelled as such.
Show the decline pile.
Make rejected and deferred initiatives visible alongside active ones. It shows the system has standards. It also lets submitters see how their idea compares to what made the cut.
Hold a quarterly reset.
Every 90 days, the forum re-scores the top 20. Strategy shifts. New information arrives. The portfolio should breathe.
A 90-day view, visible to all.
WIP limits at Halcyon: CX runs two initiatives at a time. Engineering runs three. Compliance bottlenecks at one. The limits are published alongside the roadmap, not hidden in a planning doc.
The KYC-notification flow runs through November and December alongside a CX onboarding-email rewrite. The decline pile is visible: seventeen ideas, each with a one-line reason. Maya can see why her idea was chosen, and why a colleague's idea — a complete onboarding re-architecture — wasn't.
Quarterly reset is scheduled for January. By then, the KYC initiative will be in delivery, with baseline metrics captured in October and after-state metrics due February.
- Now (Nov–Dec)5 initiatives in flight · 2 per team max
- Next (Jan–Feb)4 queued · re-scored at quarterly reset
- Declined17 ideas · visible, with reasoning
Before approving the next quarter's top three initiatives, spend twenty minutes on a premortem. The prompt: "It's ninety days from now. This initiative failed. What killed it?" Round the room. Capture everything — political risk, technical risk, capacity risk, the things no one wanted to name. The premortem is the cheapest insurance the system buys. It surfaces the failure modes that didn't come up at the prioritisation forum because the energy was on the case for. The premortem is the case against, structured so it can be heard. Most of what surfaces won't kill the initiative — but it'll change how it's run. From Annie Duke's Thinking in Bets (Gary Klein's technique), applied at the portfolio level.
The hardest part of running a CoE isn't deciding what to do — it's deciding what to stop.
Three cadences. Weekly, monthly, quarterly.
Operating rhythm// Once the four phases are operating, the system runs on three rhythms. Each has a different job. Mixing them is the most common failure mode.
The triage.
- Acknowledge new submissions
- Initial categorisation
- Route obvious quick wins
- Flag anything for the monthly forum
The forum.
- Score new submissions
- Re-rank the top 20
- Confirm what starts this month
- Publish decisions within 48h
The reset.
- Re-score against current strategy
- Retire stale items
- Refresh WIP limits
- Review the rubric itself
Four ways this fails.
Common pitfalls// Every CoE that didn't survive failed in one of these four ways. Watch for them in the first 90 days — they're easier to fix early.
The theatre trap.
The forum runs, the matrix gets updated, the roadmap looks beautiful — but nothing actually changes. Effort flows to the loudest stakeholder regardless of score.
Tie a clear capacity number to the roadmap and refuse work above it. Visibility without limits is theatre.
Death by rubric.
The scoring system gets more elaborate with each cycle — weighted criteria, sub-criteria, formulas. Scoring takes longer than doing.
Cap rubric complexity at six fields. If a new criterion is needed, retire an existing one. Complexity is the enemy of consistency.
The silent intake.
Submissions arrive, but no one hears back. Within a month, the form is dead. Within two, people stop telling you about ideas at all.
Acknowledgement is non-negotiable. A "thanks, here's where it sits" within five days is more valuable than a perfect decision in thirty.
Sponsor drift.
The exec sponsor was engaged at launch but quietly disengaged by month three. The forum becomes advisory. Decisions get re-litigated upstream.
Lock the sponsor into the quarterly reset. Re-secure their endorsement of WIP limits each quarter — make the trade-offs theirs to own.
Three disciplines underneath.
What the methodology page covers in full// The four phases are the mechanics. These three are the thinking habits that keep the mechanics honest. They're surfaced in full on the methodology page; named here so the playbook is honest about what it rests on.
Sort the problem.
- Cynefin domain at assess
- Best-practice for the obvious
- Pilots for the complex
- Don't score what can't be scored
Judge decisions, not outcomes.
- A good call can fail
- A bad call can land
- Score the reasoning, not the result
- Premortem before each reset
Make the forum truth-seeking.
- State confidence levels
- Reward dissent
- Accuracy over agreement
- Decision log, not minutes
By week six, you should have.
Starter checklist// If you can tick all fifteen, the system is operational. Anything missing at week six is debt — it'll surface in month three.
Using this in practice.
CloserThe framework is the floor, not the ceiling.
This playbook is a starting point, not a prescription. Every organisation has its own gravity — political, technical, cultural — that bends the framework in different ways. Halcyon Financial is one shape. Your shape will be different.
What travels is the discipline, not the artefacts. The five-field intake form will look different in your tool stack. The 2×2 matrix may be a different shape. The forum will have your meeting room and your faces around the table. The shape of the work — capture, assess, prioritise, sequence — is the thing that doesn't move.
If you're standing one of these up and want to talk through where it's getting stuck, I'm happy to.