The 90-day pilot-to-production playbook is a three-phase programme that takes one proven AI pilot into live use within a single quarter. Days 1 to 30 map the real workflow and everything blocking it. Days 31 to 60 assign an accountable owner and the minimum governance needed. Days 61 to 90 deploy to real users, in real systems, with live monitoring.
Variants of the approach are known as a 30-60-90 day AI plan, a pilot-to-production roadmap, a POC escape plan or an AI scale-up sprint. The version set out here is based on the method Luise Freese describes in Part 2 of a recent conversation, with an executive decision gate added by AIPEX at the end.
The structure fits the management system standards many organisations already use. ISO/IEC 42001:2023 follows the Plan-Do-Check-Act cycle, and the NIST AI Risk Management Framework (AI RMF 1.0) organises the work into Govern, Map, Measure and Manage. The 90-day playbook is a practical way to run one full cycle on one use case, quickly.
Speed matters because the alternative is attrition. Gartner predicts that over 40% of agentic AI projects will be cancelled by the end of 2027 because of escalating costs, unclear business value or inadequate risk controls (Source: Gartner, 2025). Only 11.7% of organisations move directly from AI validation into full production rollout (Source: AvePoint, State of AI Report, 2025).
Why 90 days, and why only one pilot?
Ninety days is long enough to reach real users and short enough to hold executive attention for a single quarter. Choosing one pilot is deliberate. The aim of the first cycle is not to scale a portfolio; it is to build the route to production once, so that every later pilot can follow it. Freese reports running the programme with more than twenty organisations, each producing one or two live use cases.
Choose the pilot with care. It should be a narrow, bounded use case operating in a single system, with an outcome you can measure. It should matter to the business. If you pick the easiest pilot rather than a valuable one, you will learn how to ship things nobody needed.
Days 1 to 30: map the real workflow and its blockers
The first 30 days produce an honest map of how the work is actually done today and everything standing between the pilot and production. That means walking the real workflow, not the documented one, and listing every blocker: technical, organisational, political and regulatory. The gap between the two is often the most useful finding of the whole exercise, as AI as an Organisational Truth Serum explains. Foundational gaps in identity, device management and data access belong on this list too; see AI Readiness Starts Before AI.
Exit criterion: a complete blocker inventory, with each blocker assigned to someone who can remove it.
Days 31 to 60: assign ownership and the minimum governance
The middle 30 days make one person accountable and define what the AI is allowed to do. That means a single named business owner for the production outcome, a written delegated authority for the agent (decisions it takes alone, decisions that need approval, the data it may touch) and the minimum governance needed to go live safely. We cover the delegated authority in detail in Who Owns the AI Agent When It Breaks?
Exit criterion: a named owner, an approved delegated authority, and governance rules built into delivery rather than held in a meeting.
Days 61 to 90: deploy to real users with live monitoring
The final 30 days put the use case into real systems, with real users, under live monitoring. The pilot environment is retired. Monitoring, logging, escalation and rollback are working before the first user relies on the output, and the operational budget that keeps them running has been agreed. At scale that budget is driven by compute, which we cover in Scaling AI When Compute and Power Are Scarce.
Exit criterion: the use case is running in production, monitored, with a tested fallback.
Day 90: the executive decision gate
AIPEX adds one step to the method: a formal executive decision at day 90. The leadership team reviews the evidence and makes one of three decisions: scale the use case, rework it for another cycle, or stop it. Stopping is a legitimate outcome. What is not acceptable is the default that creates pilot purgatory: no decision at all.
The gate also captures the route itself. Whatever it took to reach production (the owner, the authority, the governance, the funding) becomes the template for the next pilot.
| Phase | Executive accountable | Exit criterion |
|---|---|---|
| Days 1–30: Map | Sponsoring executive | Complete blocker inventory, each blocker owned |
| Days 31–60: Own and bound | Named business owner | Approved delegated authority; governance in the pipeline |
| Days 61–90: Deploy | Business owner with technology lead | Live in production, monitored, fallback tested |
| Day 90: Decide | Executive team or board | Scale, rework or stop, recorded as a decision |
How does the 90-day playbook work, and what does it require from the CEO and CFO?
The playbook works by forcing the decisions that stalled pilots avoid, in a fixed order and to a fixed deadline. From the CEO it requires one visible sponsor and a named business owner with real authority over the workflow. From the CFO it requires the pilot to be funded as an operation from day 61, including monitoring, maintenance and support, rather than as an experiment that ends when the demo succeeds.
Neither requirement is technical. Both are the reason most pilots never reach production. For the wider diagnosis, read AI Pilot Purgatory.
If you want to run the first cycle with experienced leadership, a Fractional Chief AI Officer can own it from day one. The Executive Strategy Day selects the right first pilot and sets up the 90 days, and our briefing From Pilot to Scale: The Execution Gap prepares the wider leadership team.
Frequently asked questions
What is a 90-day AI pilot-to-production plan?
A 90-day AI pilot-to-production plan is a three-phase programme that takes one proven pilot into live use within a quarter: map the workflow and blockers, assign ownership and minimum governance, then deploy with live monitoring. For a full explanation, see The 90-Day Route from AI Pilot to Production: An Executive Playbook →
Which AI pilot should go first?
The first pilot should be a narrow, bounded use case in a single system, with a measurable outcome that matters to the business. Choosing the easiest pilot rather than a valuable one teaches the organisation to ship things nobody needed. For a full explanation, see AI Pilot Purgatory: Why Successful Pilots Never Reach Production →
What must be agreed before an AI agent goes live?
Before an AI agent goes live, the organisation must agree named owners, a written delegated authority, an operational budget, an evidence trail and tested rollback and escalation routes. For a full explanation, see Who Owns the AI Agent When It Breaks? Delegated Authority for Boards →