Every stalled AI transformation we have seen had a plan. Usually a good one. It had a sequence, a vendor path, a data workstream, a pilot timeline, and a launch milestone. It looked organized because it was organized.
The problem is that a plan of that kind answers one question well and leaves the important ones untouched. It explains how to launch an AI initiative. It does not explain who will make trade-offs after launch, who will redesign the work, who will manage adoption, who will govern risk, who will own the operating result, or how leadership will know whether the value endured.
The best diagnostic question we know for a draft AI plan is not about the model, the vendor, or the timeline. It is this: who owns the organization after the pilot works?
Most plans have no answer. That silence is the finding.
Why a Longer Plan does Not Fix It
The instinct, when a plan is exposed as incomplete, is to add to it. More workstreams, more milestones, more governance steps. That instinct produces a longer document and the same outcome. What is missing is not detail. It is a system: the integrated set of decisions, capabilities, routines, ownership, boundaries, guardrails, and feedback loops that makes transformation repeatable rather than heroic.
AI transformation fails in a specific way. One decision goes unmade, and another is quietly forced to compensate for it. Technology gets blamed for unclear strategy. Governance gets asked to resolve missing ownership. Training gets asked to repair poor work design. Pilots multiply because delivery has no path to scale. Activity gets reported because value was never defined.
The compensation is what makes the gap so hard to see. A project manager chases decisions the operating model should have assigned. Engineers manually clean information that a data function should have maintained. A champion persuades colleagues to use a tool that should have been embedded in the work. Those efforts can make a pilot look healthy. They do not create anything the next team can rely on.
The Minimum Complete System
In our book we call the answer the Jones-Roxon Nine, or simply the Nine. It is not an inventory of everything related to AI. It is the minimum complete set of connected decisions required to turn AI capability into measurable business outcomes, and it sits in four layers.
Direction answers the first leadership problem: where are we going, and who is accountable for getting us there? Vision and Strategy defines where AI will create differentiated value and what the organization will deliberately not pursue. Operating Model and Organization defines who owns the transformation, how decisions are made, and how accountability changes as people, copilots, automations, and agents share the work.
The Engine answers the next problem: can the organization actually build the capability it is imagining? Work Enablement and Process Evolution defines which work should be eliminated, simplified, augmented, automated, or redesigned. Data and Knowledge defines what information AI must access, whether it can be trusted, and who owns it. Technology and AI Foundation defines the models, platforms, integrations, observability, resilience, and cost controls that let AI operate reliably inside the business.
The Enabler and Guardrail layer answers the human and trust problem: will people adopt the change, and can the organization move quickly without losing control? People, Leadership and Adoption defines how leaders and employees build the confidence, capability, incentives, and habits to operate differently. Governance, Ethics and Trust defines how the organization creates responsible speed through clear accountability and proportionate controls.
The Lifecycle answers the last one: can an idea reach scale, prove its value, and improve from there? Delivery and Execution moves opportunities through qualification, prioritization, testing, rollout, and operational transition. Value and Continuous Improvement determines whether measurable value is actually being created, whether the unit economics hold, and whether the capability should be improved, scaled, or retired.
Nine is Only too Many If You Run Them as Nine Programs
The most common objection is that nine decisions are more than an executive team can carry under pressure. It is a fair challenge, and the answer is practical.
Nine is too many only if leaders treat them as nine separate programs. They are not nine programs. They are nine views of the same transformation. A customer-service agent is a strategic choice, a redesigned workflow, a knowledge problem, an operating-model decision, a change in roles, a governed capability, a delivery effort, and an economic proposition all at once. An organization can ignore one of those views. The work cannot.
Completeness also does not mean equal investment. A low-risk internal knowledge assistant may need a lighter operating model and a shorter governance path. An agent that can make customer commitments, move money, or influence employment decisions requires deeper evidence, controls, and ownership. The discipline is to ask every question and scale the answer to the consequence. Proportionate does not mean partial.
What changes in the room
The value of a complete system shows up in how reviews are run. When an initiative stalls, the conversation stops being “why is the AI not working” and becomes diagnostic. Was the outcome unclear? Was ownership unresolved? Was the workflow never redesigned? Was the knowledge unreliable? Were employees expected to adopt without changed incentives or management routines?
The answer rarely sits where the symptom appears. A hallucination can be a model problem, a retrieval problem, a source-quality problem, or an evaluation problem. Low usage can reflect poor training, or it can reveal that the tool added a step, threatened an incentive, or solved a problem employees did not have. Rising cost can require model routing, process redesign, renegotiation, or retirement.
The Nine does not produce the answer automatically. It makes the right question harder to avoid. That is usually the difference between a second pilot and a second capability.
Adapted from Feasible, Not Achieved: Why AI Transformation Stalls and the NINE Decisions That Finish It, by Weston A. Jones, MBA and Jacob C. Roxon, PhD (JRNine Press).
The Jones-Roxon Nine™ is the Operator’s System for AI Transformation.