A pilot goes well. The demo lands, the steering committee signs off, the roadmap slide gets a green tick next to “Phase 1 complete.” Then the same team tries to move that pilot into production, across a real business unit, with real customers and real compliance obligations, and the roadmap stops predicting anything. New stakeholders appear who were never consulted. The governance question that seemed hypothetical in the pilot becomes a blocking one. The people expected to run the new process were never trained for it. The roadmap did not survive the transition from proof of concept to production, because it was never built to.
Technology is rarely the cause. McKinsey’s 2026 report on the state of organisations, drawing on wider McKinsey research, notes that 88% of organisations are experimenting with or deploying AI, yet 81% report no meaningful bottom-line gains. The same report puts the share of companies that have tried to adopt AI and seen significant bottom-line impact at under 20%. A separate 2026 McKinsey survey of more than 500 marketers found the same pattern at the functional level: about 60% use AI multiple times a week, but fewer than 10% of organisations have captured value across marketing workflows, and only 28% say their company is pursuing the kind of fundamental rewiring that would let them. The technology clears the pilot stage reliably. The organisation around it does not clear production. A roadmap that only sequences tools and timelines will hit the same wall. One that sequences governance, operating-model change, and people investment ahead of scale has a chance of surviving it.
Where roadmaps actually break
AI transformation roadmaps are typically built as technology rollout plans: select a use case, run a pilot, scale it, repeat. That sequencing works for software features. It does not work well for AI, because the constraint that stops a pilot from scaling is rarely the model. The same McKinsey report calls the alternative “double transformation”: technical change paired with organisational change, where the unit being redesigned is the whole workflow rather than the point where AI touches it. The marketing report reaches the same conclusion from a different survey base, describing current AI use as often added onto an unchanged process, rather than decomposing the workflow by task, deciding what a person does and what an agent does, and redesigning the sequence end to end.
A roadmap that treats “deploy the model” as the finish line skips the step that determines whether the deployment produces value. It also explains why so many programmes look identical at the pilot stage and diverge sharply at the production stage: the pilot never tests the organisational rewiring that scale requires.
Governance has to be in place before scale
The reflex in many transformation plans is to treat governance as a compliance gate near the end: build the thing, then get it approved. McKinsey’s 2026 AI Trust Maturity Survey of roughly 500 organisations suggests the opposite works better. The survey scores each organisation’s responsible-AI maturity on a four-level scale, from foundational practices at level one to a comprehensive, proactive programme at level four, across five dimensions: strategy, risk management, data and technology, governance, and agentic AI controls. The average score has risen to 2.3 out of 4, up from 2.0 a year earlier, but strategy, governance, and agentic controls consistently lag the other dimensions, with only around 30% of organisations reaching maturity level three or higher on those three. McKinsey’s report on the state of organisations shows the same concern from the adoption side: regulatory, ethical or legal concern is the second-biggest barrier to adopting AI at scale, cited by 44% of leaders. Organisations know the risk exists and are often slow to build the controls for it, usually until a pilot needs to move into a regulated or customer-facing environment.
The AI Trust Maturity Survey also found that organisations with clear responsible-AI ownership score 2.6 out of 4 on maturity against 1.8 for those without a clearly accountable function, and that organisations investing $25 million or more in responsible AI are more likely to report EBIT impact above 5%. The survey links ownership and maturity to results, which makes governance built into the roadmap from the start a condition for scaling and a cost worth planning for. A roadmap that names an owner for AI risk and builds agentic controls into the design phase, before the production decision arrives, removes a common reason a scaling request stalls in review.
The people-to-technology ratio the roadmap usually gets backwards
McKinsey’s state of organisations report quotes a rule of thumb from one executive that is worth putting directly on the roadmap: for every $1 an organisation spends on AI technology, roughly $5 should go to the people who will use it. The report also cites separate McKinsey research on the agentic organisation finding that around 75% of current roles will need reshaping as AI enters the workflow, and its own survey shows 86% of leaders saying their organisations are not very prepared to adopt AI in day-to-day operations. The AI Trust Maturity Survey separately found that knowledge and training gaps are the leading barrier to responsible AI implementation.
Transformation budgets are typically built the other way round: licences and infrastructure are line items from day one, and enablement is a training session scheduled after go-live if there is time. A roadmap built to reach production treats enablement as continuous coaching tied to the specific tasks changing, and sizes the people investment against the technology spend before the business case is signed off, rather than waiting for the adoption to fail.
Leadership clarity decays the further it travels from the C-suite
The same McKinsey report measured how clearly leaders at different levels understand the organisation’s “must-win battles”: 56% of executives report clarity, and the figure drops sharply further down the organisation. The marketing survey adds a related finding about the people further down the path: 57% of marketers report anxiety about AI’s implications, a concern a roadmap owned only by a central team is unlikely to hear.
A roadmap owned solely by a central AI team, with milestones set and communicated only at the top, keeps losing clarity at every layer it passes through. The roadmaps that hold together cascade the vision explicitly to the managers and teams doing the redesign, and build a channel for frontline concerns to travel back up before they surface as resistance during rollout.
Sequencing a roadmap that reaches production
Put together, the evidence points to a specific ordering rather than a longer list of workstreams.
- Start from the workflow to be redesigned and decompose it by task before choosing tools or writing a timeline.
- Assign explicit ownership for AI governance and build the agentic controls the use case will need at the scale it is heading towards, rather than the scale of the pilot.
- Size the enablement budget against the technology budget using the people-to-technology ratio above as a starting point, and schedule it as ongoing coaching rather than a single session before go-live.
- Cascade the vision and the must-win battles to the managers who will run the redesigned workflow, and build a channel for their concerns to travel back up before the production decision is made.
None of this changes what the pilot proves. It changes what happens once the pilot succeeds and the same team is asked to do it again at scale, with governance, headcount, and customers attached. That is the point roadmaps are not built for, and it is the point that decides whether the AI investment shows up in the bottom line.
Zartis works with organisations on the AI Operating Model and AI Governance work that sits underneath a roadmap like this: naming the workflow to redesign, building the ownership and controls a production deployment needs, and sizing the enablement investment before the scaling decision is made rather than after it stalls. If a transformation roadmap is stuck between a successful pilot and a production rollout that keeps slipping, the gap is usually the organisational layer described above.
