Most mid-sized companies have between 5 and 20 active AI initiatives. 70% lack a defined owner. 80% have no business metric. 90% have no criteria for closure.
The result: AI sprawl. Every week, someone proposes something new, nothing gets closed, the executive team wastes energy on demos, and real value isn’t measured because nobody knows what to measure.
This playbook solves that in 14 days. It’s not a workshop. It’s not consulting. It’s a minimal operating system you can implement with internal resources, with or without external support.
Before starting: what this playbook is not
It’s not an academic governance framework. It doesn’t require buying new software. It doesn’t produce an 80-page document that nobody reads.
It produces three things: a clean inventory, a decision system, and a cadence that works without needing to be redesigned every quarter.
If you’re not willing to close at least one initiative during these 14 days, the playbook won’t work. The goal isn’t to organize chaos: it’s to reduce it.
Days 1-3: Real Inventory
Day 1 — Portfolio Snapshot
Action: List all active AI initiatives in the company. Include what IT calls a “project,” what departments call a “pilot,” and what marketing calls an “experiment.” Everything.
Output: A spreadsheet with columns: initiative name, owning department, status (active/paused/undefined), time active in months, estimated monthly cost.
Check: If you can’t fill the estimated cost column for more than 50% of initiatives, you have a financial visibility problem before you have a governance problem.
Day 2 — Assign Provisional Ownership
Action: For each initiative in the inventory, identify who’s really responsible. Not the original promoter: the current responsible person. Who gets blamed if it fails, who gets credit if it works.
Output: Additional column in the inventory: owner (name, not title), and whether it’s formalized or de facto.
Check: If more than 40% of initiatives lack a clear owner, the problem isn’t technical. It’s organizational. No AI model compensates for diffuse ownership.
Day 3 — Connect to Business Metrics
Action: For each active initiative, ask the owner to complete this sentence: “This initiative exists to [business verb] [specific metric] in [timeframe].” If they can’t complete it, the initiative lacks a success criterion.
Output: List of initiatives with attached business metric, and a separate list of initiatives without a metric.
Check: Any initiative without a clear business metric automatically enters the list of candidates for closure. Not as a penalty: as a diagnosis.
Days 4-7: Portfolio Hygiene and Kill Switches
Day 4 — Impact-Reversibility Matrix
Action: For initiatives with metrics, apply a simple two-axis matrix: potential impact (high/medium/low) and reversal cost (high/medium/low). Don’t use 1-10 scales: they create more false precision than they resolve.
Output: Four quadrants. High-impact, low-reversal-cost initiatives go first. Low-impact, high-reversal-cost initiatives are the first candidates for closure.
Check: If everything appears as “high impact,” you’re seeing confirmation bias, not reality. Force at least 30% to fall into low or medium impact.
Day 5 — Define Kill Criteria per Initiative
Action: For each initiative still active after the matrix, define an explicit kill criterion: the condition that, if met, triggers the decision to pause or close without needing a meeting.
Output: One line per initiative: “If [metric] doesn’t reach [threshold] by [date], the initiative is automatically paused and the owner presents options within 48 hours.”
Check: A kill criterion requiring consensus from five people isn’t a kill criterion. It’s a disguised meeting. The criterion must be triggered by a single responsible person.
Day 6 — Execute Closures for the Week
Action: Effectively close initiatives that didn’t pass the Day 4-5 filters. Formal closure: communicate to the owner, deactivate resources, record the reason for closure.
Output: List of closed initiatives with date, reason, and what resources become available.
Check: If the first closure produces significant political friction, you’re discovering that the governance problem was also a power management problem. The playbook doesn’t solve that: it requires explicit executive mandate. Without it, the operating system can’t function.
Day 7 — Review the Reduced Portfolio
Action: With the remaining initiatives, review the updated inventory. Compare against current quarter business objectives.
Output: Clean portfolio: maximum 5-8 active initiatives with owner, metric, and kill criterion. If there are more, the filter criterion was too lax.
Check: Can the current portfolio be explained in 3 minutes to any member of the executive team? If not, it’s too complex to be well-governed.
Days 8-11: Operating Cadence and Review Stack
Day 8 — Design Review Cadence
Action: Define the portfolio review rhythm: weekly for alert signals, monthly for adjustment decisions, quarterly for closures or new additions.
Output: Fixed calendar with three types of review: weekly signals (15 minutes, just owners), monthly portfolio review (45 minutes, executive team), quarterly criteria review (90 minutes, includes potential new initiatives).
Check: If the weekly review requires more than 30 minutes of preparation, the format is poorly designed. It should be automatic: owners report against their metrics without preparing a presentation.
Day 9 — Configure Executive Review Stack
Action: Select four weekly signals any executive can review in 10 minutes. Not a 40-metric dashboard: four signals with thresholds.
Reference signals:
- Decision quality: % of AI decisions reversed in the last 30 days.
- Escalation load: number of decisions AI systems have escalated to humans vs. previous week.
- Margin leakage: cases where AI intervention increased documented rework.
- Active portfolio: ratio of initiatives with green vs. yellow vs. red metrics.
Output: A four-row dashboard. Each with current value, alert threshold, and who decides if it’s crossed.
Check: If a signal crosses the threshold and nobody knows what to do, the decision component is missing. Each signal needs a predefined executive action: continue, correct, or stop.
Day 10 — Formalize Decision Rights
Action: Define, per initiative, who can make each type of decision without needing consensus: (a) modify scope, (b) reassign owner, (c) pause initiative, (d) permanently close it.
Output: Decision rights table: role, decision type, required authorization.
Check: If the answer to “who can close this initiative” is “the AI committee,” governance is theater. Decisions must have a singular responsible person, not a committee.
Day 11 — First Weekly Review Cycle
Action: Execute the first weekly review with the format designed on Day 8. Time the meeting. Check if the Day 9 signals are available without additional manual work.
Output: Minutes of the first review: state of each signal, decisions made (or not made), next steps.
Check: If the first review takes more than 30 minutes, the format needs adjustment. If no decisions are made, the system isn’t working: it’s reporting.
Days 12-14: Rollback and Handoff
Day 12 — Design Rollback Protocols
Action: For each active initiative, define the reversal procedure if something goes wrong: how it’s deactivated, who does it, in what timeframe, what systems are affected, and what manual process replaces it while.
Output: A rollback brief per initiative. Maximum one page per initiative.
Check: If a rollback requires more than 48 hours of coordination, the initiative has a reversibility design problem that needs to be solved before scaling.
Day 13 — Document the Operating System
Action: Consolidate into a single document (no more than 5 pages): the initiative inventory, kill criteria, review cadence, decision rights, and rollback protocols.
Output: The foundational document of the AI operating system. Not a manual: it’s the reference any new executive team member can read in 20 minutes and understand what’s happening.
Check: If the document can’t be read in 20 minutes, it’s overdesigned. The document’s complexity isn’t a sign of rigor: it’s a sign the system isn’t clear yet.
Day 14 — Handoff and First Cycle Closed
Action: Review the complete cycle. Compare the portfolio state on Day 1 vs. Day 14. Communicate results to the executive team: what was closed, what remains, what metrics each active initiative has.
Output: Governance sprint closure report: before/after, decisions made, active operating system.
Check: If the executive team can’t answer in 60 seconds how many initiatives are active and what their metrics are, the operating system hasn’t been installed.
Anti-Patterns: What Not to Do During the 14 Days
Don’t start with politics. If on Day 1 you spend time agreeing on who should be on the governance committee, you’ve wasted the first day on bureaucracy. Start with data: the inventory doesn’t require political approval.
Don’t close without criteria. Closing an initiative because “it doesn’t seem useful” is arbitrary and generates resistance. Close only against predefined kill criteria. Even if the criterion is simple, it must be written.
Don’t design the system for the ideal case. The operating system must work when the owner is missing, when there’s political friction, or when data is incomplete. If it only works under perfect conditions, it won’t work.
Don’t multiply review signals. The temptation to add more metrics to the review stack is constant. Resist it. Four signals with thresholds are more useful than twenty signals without owners.
Don’t delegate closure to the committee. The committee can recommend. The closure decision needs a singular responsible person. Without that, closure never happens or always comes late.
The Indicator that the System Works
At the end of Day 14, the AI operating system is working if you can answer these four questions in 60 seconds:
- How many AI initiatives are active right now?
- Who’s the owner of each, and what’s their business metric?
- When is the next executive review of the portfolio?
- What condition would trigger the closure of the most important initiative?
If you can’t answer all four, the system isn’t installed yet. The work is simple: go back to where clarity was lost.
Download the Full Playbook in PDF and Implement AI Governance in Your Company
This post is the complete content of the playbook. The PDF version includes work templates for each day: the inventory sheet, impact-reversibility matrix, kill criteria template, and four-signal dashboard.
Download the PDF and implement this same sprint in your company.
Related Articles
- AI Portfolio Hygiene: fewer use cases, better results
- Decision Kill-Switch: the protocol that prevents an AI initiative from staying alive by inertia
- Operating Cadence: the forgotten variable in AI teams
Translated from the Spanish original with AI assistance and reviewed for accuracy. Read the original in Spanish.