Your AI strategy doesn't need a better roadmap. It needs better questions first. Companies are set to double AI spending this year, from 0.8% to 1.7% of revenue (BCG, 2026). Yet only about 6% of organizations are attributing real EBIT impact to AI (McKinsey, 2025). The gap isn't the tech. It's what leadership teams skip before the strategy gets written. Here's what I'd ask your leadership team in the first hour, before a single vendor call or roadmap: 1/ Which P&L line does this move, and who owns that number? → If the answer is "productivity" you don't have a strategy. You have a hope. → Tracking well-defined KPIs is the single practice most correlated with bottom-line impact (McKinsey, 2025). Reality: AI initiatives without a named executive owner become zombie pilots with a budget line. 2/ What decision are we willing to let AI make without a human? → This one question exposes your real risk appetite. → Only about 30% of organizations reach maturity level three or higher on AI strategy, governance, and agentic controls (McKinsey, 2026). Reality: In the agentic era, the risk isn't AI saying the wrong thing. It's AI doing the wrong thing. 3/ Which workflow would we redesign end-to-end if AI didn't exist as an excuse to avoid it? → High performers are 2.8x more likely to have fundamentally redesigned workflows, 55% vs 20% (McKinsey, 2025). → Layering AI on a broken process gets you a faster broken process. Reality: Roughly 70% of AI's value potential sits in core business workflows (BCG, 2026). 4/ Who in this room has personally spent 8 hours with these tools this month? → CEOs who invest at least 8 hours a week building their own AI capability generate more value from it (BCG, 2026). → You can't govern what you've never touched. Reality: The model is the cheapest part of your rollout. Leadership attention is the scarcest. 5/ What happens to the hours we save? → Redeployed to higher-value work? Reduced cost? Reinvested in growth? → If nobody can answer, the savings evaporate into the org chart. Reality: Time saved with no plan for the time is ineffective. 6/ Which of our current pilots would we kill today? → Gartner projects over 40% of agentic AI projects will be canceled by end of 2027. → Better to be the one holding the scalpel than the one explaining the write-off. Reality: Pilot purgatory is a choice. Nearly two-thirds of organizations still haven't begun scaling AI across the enterprise (McKinsey, 2025). 7/ If this works, whose job changes first, and have we told them? → Workforce transformation announced after the fact reads as workforce reduction. → Half of CEOs now believe their own job stability depends on getting AI right (BCG, 2026). Reality: If AI accountability reaches the CEO's chair, it reaches everyone's. Notice what's not on this list: model selection, vendor comparisons, or a single acronym. Because an AI strategy written before these answers isn't a strategy. It's procurement with a narrative.
This is the right way to frame AI strategy. A roadmap without these questions can look very mature and still fail to create value. The hardest questions are not about which model to use or which vendor to select. They are about ownership, economics, workflow change, risk appetite, and accountability. Which P&L line should move? Who owns the outcome? Which workflow needs redesign? What decisions can AI support or execute? What happens to the time saved? Which pilots should stop? Those questions force leadership to move from AI activity to AI value. They also expose whether the organization is ready to scale or just funding another layer of experimentation. AI strategy is not a technology plan with executive language around it. It is an operating model decision about how the business will work differently.
I particularly like: "What decision are we willing to let AI make without a human?" I wonder whether there is an even earlier question: Which decisions should never be delegated to AI at all? Because once we define: what outcomes matter, which workflows should change, and where autonomy is acceptable, we still need to answer: Who retains the authority to determine whether an AI-generated action should be allowed to become reality? I increasingly think many organisations are treating AI adoption as a technology strategy when it may first be a governance and decision rights exercise. The challenge isn't only: "What can AI do?" It's: "Which consequences are we willing to permit AI to create, under what conditions, and who remains accountable?" Because an AI strategy ultimately becomes real at the point where authority is delegated. #AIGovernance #AIRiskManagement #TauDIL #IFA #TauGuard
Question one is the whole thing. A named owner only bites if they actually control the number, not just report on it. Most zombie pilots have a sponsor for the launch and nobody accountable for the P&L line six months later.
Essential set of questions Carolyn. I would probably start with an eighth question: What are the critical decisions we’re trying to improve? Every AI investment ultimately exists to improve the quality, speed, consistency, or confidence of decisions. Yet most organizations still frame AI around tools, productivity, or automation instead of the decisions that create enterprise value. Once you identify the decisions that matter most, the rest of the questions become much easier to answer: ➺ Which workflows need redesign? ➺ Where can AI act autonomously? ➺ Which executive owns the outcome? ➺ What KPIs actually matter? ➺ Which pilots deserve to scale or stop? In my experience, organizations rarely suffer from a lack of AI ambition. They suffer from a lack of decision clarity. They have dashboards, copilots, and agents, but no shared understanding of which decisions drive growth or how to improve them. The companies that pull ahead in this next era won’t simply deploy more AI. They’ll become far more intentional about the decisions they choose to augment, automate, and govern.
Too many AI strategies begin with the technology rather than the problem, the workflow and the people expected to use it. In healthcare, the first questions should be: What clinical problem are we solving? What evidence will prove that it is safer or better? Who remains accountable when it fails? And what happens to the time we claim to save? AI layered onto a broken workflow simply creates a faster broken workflow. Real value comes from redesigning the system, involving frontline teams and patients, and building governance in from the start. The roadmap should follow the answers, not replace them.
A practical starting place is one core workflow with a named executive owner and one board-level measure. The test exposes gaps in process design, decision rights and staff readiness.
The strongest AI strategies I’ve seen don’t start with AI. They start with a business decision that must become faster, better or cheaper. Everything else, models, vendors and architecture, should follow that choice.
Exactly. A real AI strategy should leave the room with named workflows, accountable owners, baselines, decision rights, value-capture plans, and kill criteria. Without those commitments, the roadmap is usually procurement with milestones.
I like the focus on questions before tools. I've noticed that the quality of an AI strategy is usually limited by the quality of the conversations leadership has before implementation even starts.
AI strategy often fails because leaders jump to tools before agreeing on ownership, risk, and economic impact. A roadmap cannot fix weak decisions made before the roadmap begins. Clear answers on P&L, workflows, and accountability turn AI from an experiment into a business system.