Kartini Cooper
← Back to portfolio
// Playbook 06 · 2026
Playbook · Rollout for individual champions

How to spread what you've built.

You've found something that works. Now you want your team to use it too. This is the part most "AI adoption" advice waves at and walks past — and it's the part that actually decides whether the work makes a difference. Six weeks. Four phases. Bottom-up, not big-bang. The discipline of spreading capability without becoming the help desk.

// Who it's for
The person who figured it out
// Time investment
~2 hrs/wk for 6 weeks
// You'll need
Something that works already
// You'll walk away with
A network, not a help desk
01 —
// The framework

Four phases. Find, show, lower, embed.

Adoption doesn't happen through launches. It happens through people who try a thing, find it useful, and tell two other people. The framework below is how to set those conditions deliberately — without becoming the bottleneck for the network you're trying to build.

1

Find your five

Weeks 1–2

Five named humans, not "the team". The earliest adopters are the ones with a real problem and the curiosity to try something new.

2

Make it visible

Weeks 2–4

Work with one or two of them in public. Show your work in channels they already read. Demonstrate the saving in their language.

3

Lower the floor

Weeks 4–6

Turn what you built into onboarding kit. Templates, Projects, Styles, prompts. The next person should start where you finished.

4

Make it routine

Ongoing

Embed it in existing rhythms — standups, 1:1s, demos — so it stops being a special thing and starts being how the team works.

1 FIND WK 1–2 2 SHOW WK 2–4 3 LOWER WK 4–6 4 EMBED ONGOING
02 —
// Phase one

Five people, by name.

The first move isn't a launch. It's a list. Five named humans who have a real problem and the disposition to try something new. Everything else in this playbook depends on getting this part right.

1
Find · Phase 1

Pick the first five.

Weeks 1–2

"Roll it out to the team" is not a strategy. The team is a hundred different people with a hundred different problems, levels of curiosity, and tolerance for new tools. The first five aren't a representative sample — they're the ones most likely to find the thing useful first, become advocates, and pull the rest along.

  1. 01

    Write down five names.

    Not roles. Not "the analyst team". Five actual humans. The ones who you can already imagine sending you a slightly excited message after they try it.

  2. 02

    Match each name to a real problem.

    For each of the five, write the specific problem this would solve for them. Not the generic benefit. The thing they complain about on a Tuesday afternoon. If you can't name it, you don't know them well enough yet.

  3. 03

    Pick the two easiest first.

    Of the five, two will be the easiest sells — already curious, already frustrated with the current way, already trying things. Start with those. The other three watch what happens to the first two.

// Working example · the prompting playbook to a finance team
Names
Five names, written down: Priya (FP&A lead, drowning in board paper prep), Marco (financial controller, hates writing quarterly commentary), Sam (treasury analyst, already using ChatGPT on the side), Jen (CFO's chief of staff, exposed to every kind of writing task), Dan (financial systems lead, naturally curious about new tools).
Problems
Priya — turning 60 pages of board paper drafts into a 2-page CFO brief, every month. Marco — writing the variance commentary nobody reads but everyone needs. Sam — summarising rate analyst notes from three providers. Jen — turning meeting notes into actions and follow-ups. Dan — writing user guides for the new ERP module.
First two
Start with Sam (already trying tools on his own — needs guidance, not convincing) and Jen (most exposed surface area — every win shows up in five different meetings).
i
Insight from the field

The five-names exercise is the moment most rollouts succeed or fail. If you struggle to name five real people with five real problems, you're not ready to roll out — you're ready to do more discovery. Go back, listen harder, build trust. The names come later, not first.

// Phase 1 deliverables
Names list · five real humans
Problems list · one per name, specific
Start order · which two first
2
Show · Phase 2

Work in public.

Weeks 2–4

People don't adopt what they're told about. They adopt what they've seen working on a problem like theirs. Public-by-default working sessions, public-by-default wins, public-by-default mistakes. Visibility is the leverage.

  1. 04

    Run a shoulder session with each of the first two.

    30 minutes. Their problem, their workflow, your laptop or theirs. Walk through the prompting patterns from Playbook 01 on a real task they brought. Don't lecture. Co-build.

  2. 05

    Post the win publicly — in their words.

    After the session, get them to share what changed in their existing team channel — not yours. Their language, their framing, their numbers. "I used to spend 90 minutes on variance commentary. Took me 20 today." Champions rolling out via other champions is the entire mechanism.

  3. 06

    Show the misses too.

    When it doesn't work, post that as well. The output was wrong. The prompt needed three goes. The reference doc was stale. Visibility of the failures is what makes the wins credible — and what stops the rollout looking like marketing.

  4. 07

    Let the next three ask first.

    Don't push to the other three names. Let them come to you after seeing the first two posting. The pull is stronger than the push — and the people who self-select are easier to help.

// Working example · public visibility, week 3
Session
30-min screen share with Sam on his rate-analyst summaries. Walked through Pattern 02 (the exemplar) — pasted two of his old summaries he was proud of, then ran a new one. Output was 80% there in one go. He fixed the last 20% in three minutes.
Post
Sam in #treasury-team: "Spent 30 min with Kartini on this — turning analyst notes into our weekly brief now takes me 15 minutes instead of an hour. Happy to show anyone who's curious. Disclaimer: it gets it wrong sometimes and I still read every line." In his words. In his channel. With the honest caveat.
Miss
Day later, posted in same channel: "Update — tried it on the macro brief and the output was confidently wrong about two rate calls. Stick to the analyst summary use case for now. Working on why." The miss made the wins more credible, not less.
Pull
By end of week: Priya and Dan asked Sam (not me) if they could borrow his approach. Marco emailed me directly. Jen scheduled a Friday session. The pull worked.
!
Watch for

The temptation to do the work yourself and hand over a finished thing. Don't. Co-building beats handover every time — the person who built it owns it, and the person who watched you build it didn't learn anything they can repeat. The session takes longer, the rollout takes less time.

// Phase 2 deliverables
2 shoulder sessions · one per first-two
2 public wins · in their channels, their words
1+ public miss · honest, useful
Inbound requests · from the next three
3
Lower · Phase 3

Onboarding kit, not a course.

Weeks 4–6

Once a handful of people are using it, the next move is to turn what you built into onboarding kit — so the sixth, seventh and eighth people don't need a shoulder session from you. Templates, Projects, Styles, prompt patterns. The next person should start where the first five finished.

  1. 08

    Package the three things that worked best.

    From the first five, two or three patterns will dominate. The summarising prompt. The variance commentary chain. The meeting-notes-to-actions Skill. Package those three — not all twelve things you tried. Fewer, better, finished.

  2. 09

    Make a shared Project in Claude.ai.

    Per Playbook 02: instructions for the team's voice, knowledge base with your team's style guide and 3–4 examples of good output, the prompt templates from Playbook 01 pinned in the instructions. Share it with the first five. Let them clone or use directly.

  3. 10

    Write a one-pager, not a manual.

    A single page: what this is, who it's for, the three things it does well, the things it doesn't do, and a "if you want to try it, do these three steps" section. Link to the playbooks for depth. The one-pager is the doorway, not the room.

  4. 11

    Set a limit on yourself.

    Two hours a week max on this, beyond your own work. If you're doing more, you're becoming the bottleneck — which is the failure mode of every rollout that dies at month three. If you don't have time for the eighth person, the kit has to do the work.

// Working example · onboarding kit, end of week 5
Three patterns
From the five: (1) summarising long docs (Priya + Sam), (2) writing variance/quarterly commentary using two-example prompting (Marco), (3) turning meeting notes into structured actions (Jen). Dan's user-guide work didn't generalise — kept it as a one-off.
Shared Project
Project called finance-team-claude. Instructions: team's house style, Australian English, "lead with the number, then the why, then the so-what". Knowledge base: the team style guide (already existed), three good summaries from Priya, two variance commentaries Marco was proud of. Three prompt templates pinned in the project instructions.
One-pager
Single Confluence page. Title: "Claude for finance — start here." 300 words. Three sections — "Three things this is good at", "Two things it isn't", "Try it in 15 minutes". Three links: shared Project, prompting playbook, memory playbook. No screenshots. No "frequently asked questions". The page that earns the eighth person 80% of the way without a shoulder session.
i
Why this works

Most onboarding decks try to teach people everything. The good ones teach people enough to start. The kit doesn't need to be comprehensive — it needs to be sufficient. Twenty minutes of curiosity, three working examples, a shared Project they can use immediately. That's the whole job.

// Phase 3 deliverables
3 packaged patterns · the best of the five
Shared Project · instructions + knowledge
One-pager · doorway, not manual
Time cap · 2 hrs/wk on rollout
4
Embed · Phase 4

From special thing to default.

Ongoing

A rollout is "finished" when nobody talks about it anymore — because using it is just how the team works. The job in the final phase is to embed the practice in existing rhythms: the standup, the 1:1, the demo Friday, the quarterly review. Stop running it as a programme. Let it dissolve into the operating model.

  1. 12

    Add a standing item to one existing forum.

    One. Not five. Pick the forum your team already shows up to — the Monday standup, the Thursday catch-up — and add a 5-minute "what did Claude help with this week" slot. One forum, lightweight, no decks. If the slot dies, kill it gracefully and pick another.

  2. 13

    Hand specific patterns to specific people.

    Sam owns the analyst-summary pattern. Marco owns variance commentary. Jen owns meeting-notes-to-actions. Each of the five becomes the local expert for one thing — and the person new joiners go to first. You're no longer the only door.

  3. 14

    Quietly step back.

    Stop attending the standing item. Stop posting wins in the team channel. Stop being the named contact in the one-pager. The rollout is mature when you can be on leave for two weeks and the practice doesn't visibly slow down. Successful rollouts make their champion redundant.

// Working example · embedding into the operating model
Standing item
Added a 5-minute "what Claude did this week" slot to the team's existing Monday standup. Week 1 — three people mentioned something. Week 4 — everyone had something. Week 8 — people stopped framing it as "what Claude did" and started just describing how they got the work done.
Ownership handed over
Sam owns the analyst-summary pattern (he wrote a one-page guide). Marco owns the variance commentary (he updated the team's quarterly process). Jen owns meeting-actions (her Skill is now in the shared Project). When a new joiner asks about Claude, they're routed to one of those three, not to me.
The step-back signal
Took two weeks of leave. Came back to: a new joiner had been onboarded by Sam; two new patterns had been added to the Project by Marco; Jen had run a session with another team. None of it needed me. That's the finish line.
// Phase 4 deliverables
1 standing item · in an existing forum
Pattern owners · one per major use
Routing · new joiners go to them, not you
The step-back · graceful exit from the centre
Adoption happens through people who tried it, found it useful, and told two other people.
03 —
// Operating rhythm

Three cadences. Weekly, monthly, quarterly.

Once the four phases are running, the rollout sustains itself on three rhythms — light, repeatable, low-burden. Each has a different job. Mixing them is the failure mode that turns champions into help desks.

Weekly · 15 min

The scan.

  • Skim the team's channels for wins and misses
  • Reply with one comment, not five
  • Note who's stuck — but don't intervene yet
  • Update the shared Project if a pattern needs sharpening
Monthly · 30 min

The round.

  • 15 min in the standing item, listening
  • One coffee with someone who's stuck
  • Add new patterns to the shared Project
  • Retire patterns nobody uses
Quarterly · 1 hr

The review.

  • Audit the one-pager — still accurate?
  • Check pattern owners are still owning
  • Refresh examples in the knowledge base
  • Decide whether to spread further or hold
04 —
// Common pitfalls

Four ways the rollout stalls.

Every rollout that didn't survive failed in one of these four ways. Watch for them in the first 90 days — easier to correct early, before they ossify into "that thing we tried last year".

// Pitfall 01

The keynote launch.

Big-bang reveal. All-hands meeting. Demo with slides. A flurry of enthusiasm, three weeks of nothing, then quiet retirement. Launches don't drive adoption — peer evidence does.

The fix Refuse the all-hands. Start with the five names. Let visibility build through the wins of people the team already trusts. The keynote launch and a closed-door rollout look identical 60 days later.
// Pitfall 02

The mandate.

"Everyone must use AI by Q3." Compliance theatre. People log in once, post a screenshot, mark themselves complete, and never touch it again. Mandates produce activity, not adoption.

The fix Resist the urge to formalise too early. The mandate often comes from above and feels supportive. Convert it into "we're tracking who's tried it and what they've learned" — a softer signal that protects the bottom-up motion.
// Pitfall 03

The champion's burnout.

You become the help desk. Every prompt question, every error, every "can you just show me?" comes to you. Within two months you're rolling your eyes when Slack pings. Within three, the rollout stalls because you've stopped responding.

The fix The 2-hour-a-week cap from Phase 3 isn't optional. Hand pattern ownership to the first five explicitly. "For this kind of question, ask Sam — he's the one who wrote it" is a complete sentence and a complete answer.
// Pitfall 04

The shadow rollout.

Adoption is happening, but invisibly. People are using it but not telling each other, not telling their manager, not putting it in their performance review. Management doesn't see it, doesn't resource it, doesn't reward it. The work happens and then stops happening when the champion leaves.

The fix Visibility upward matters as much as visibility sideways. Once a month, send your manager (or skip-level) a 3-line note: "5 people now using regularly, 2 new patterns added this month, ~15 hours/week saved across the team." Make the rollout legible to the people who fund it.
05 —
// Your starter checklist

By week six, you should have.

Twelve items. If you can tick them all, the rollout is operational and you're on the way to redundant. Anything missing is debt — it'll surface at month three, the moment most rollouts quietly die.

✓
Five names written down
Phase 1
✓
A real problem matched to each name
Phase 1
✓
First two picked, in order
Phase 1
✓
Two shoulder sessions done, in their workflow
Phase 2
✓
Two public wins posted by them, in their words
Phase 2
✓
At least one public miss shared honestly
Phase 2
✓
Three packaged patterns, the best of the work
Phase 3
✓
A shared Project · instructions + knowledge
Phase 3
✓
A one-page doorway document
Phase 3
✓
A standing item in one existing forum
Phase 4
✓
Pattern owners · one per major use case
Phase 4
✓
An upward signal — your manager knows
Phase 4

Using this in practice?

Rollout is the part most "AI adoption" advice waves at and walks past. "Now share it with your team!" is not a plan. The plan is: find five named humans, work with two of them in public, package what worked into onboarding kit, then embed the practice in existing rhythms and step back.

The hardest part of this playbook isn't any of the steps. It's the stepping back. Champions who can't let go become bottlenecks; bottlenecks become the reason the rollout stalls. The work is done when nobody talks about the rollout anymore — because it's just how the team works.

This playbook closes the individual track in the series. The earlier ones — Prompting, Memory and context — are what you teach in the shoulder sessions. Skills, MCP, and Evals are what you eventually package into the shared kit. Playbook 07 picks up the advanced thread: agents.