Most companies that believe they have an AI strategy actually have a list of AI use cases someone compiled after reading a few LinkedIn posts. The list might be sensible. It might even be ambitious. But a list of things you'd like AI to do for you is not a strategy — it's a wishlist with better fonts.
A real AI strategy starts before the use cases. It starts with a clear-eyed understanding of the environment your organisation is operating in and an honest assessment of what you actually have to work with. Only then — once you know the landscape and your own position in it — does it make sense to decide what to do.
The two tools I return to most consistently when working with businesses on this are PESTEL analysis and the AI Strategy Canvas. Used together, they turn what is usually a political conversation ("whose AI idea gets funding?") into a structured one.
PESTEL: reading the landscape before you build
PESTEL is a classical strategic analysis framework — Political, Economic, Social, Technological, Environmental, Legal — that most business school graduates learned and then promptly forgot to use. In the context of AI strategy, it earns its keep, because AI adoption is shaped by external forces that most companies dramatically underestimate.
Political. The regulatory environment for AI is moving quickly and unevenly. The EU AI Act is now in force, with risk-based classifications that affect how AI systems used in hiring, credit scoring, and healthcare can be deployed. National AI strategies across Switzerland, Germany, and the US are creating different incentive structures for investment and compliance. If your organisation operates across borders, the political dimension of AI isn't abstract — it's a real constraint on what you can build and where you can deploy it.
Economic. AI investment costs are falling, but unevenly. Inference costs have dropped dramatically for standard models, but fine-tuning, infrastructure, and talent remain expensive. The economic question for most organisations isn't "can we afford AI?" — it's "where is the actual ROI, and how long until we see it?" Honest economic analysis forces you to prioritise use cases by expected return rather than strategic optics.
Social. Workforce acceptance of AI varies enormously by industry, age cohort, and organisational culture. A company that rolls out AI tools without understanding the human dynamics — fear of job displacement, mistrust of automated decisions, discomfort with AI-generated communications — will face adoption problems that no amount of technical sophistication can solve. The social dimension also includes your customers: how comfortable are they with AI making decisions that affect them?
Technological. What is the current capability landscape, and where is it heading? Which tasks can today's models do reliably enough to build a product around, and which are still too unreliable to put in front of users? What does your integration stack look like, and what would it cost to connect AI capabilities to your existing systems? The technological dimension requires genuine technical assessment, not vendor promises.
Environmental. Compute is energy-intensive. Large language model inference at scale has a material carbon footprint. For companies with public sustainability commitments — and increasingly, for those subject to ESG reporting — the environmental cost of AI deployment is a real consideration, not a footnote. This is particularly relevant when evaluating whether to run models locally versus using cloud inference.
Legal. Beyond regulation, the legal landscape for AI includes intellectual property questions (who owns AI-generated output?), data privacy requirements (what can you feed into a model?), liability frameworks (who is responsible when an AI system causes harm?), and contractual obligations with customers and employees. Legal analysis should happen before you ship, not after something goes wrong.
PESTEL forces you to have the uncomfortable conversations before the investment is made, not after the project is stalled waiting for legal sign-off.
The AI Strategy Canvas: turning analysis into decisions
Once you have a clear picture of the landscape, the AI Strategy Canvas provides the structure for making decisions within it. Think of it as a Business Model Canvas — but instead of mapping how your business creates and captures value, it maps how AI fits into that logic.
The canvas has seven fields, and the sequence matters:
- AI Ambition. What are you actually trying to achieve? Not "use AI" — that is not an ambition. The question is: where do you need to be better than you currently are, and do you believe AI is the most effective way to get there? A clear ambition filters everything that follows.
- Data Assets. What data do you have, and what is its quality? AI systems are only as good as the data they're trained on or grounded in. An honest audit of your data assets often reveals that the foundational work — cleaning, structuring, making data accessible — needs to happen before any model can be deployed effectively.
- AI Capabilities. What do you currently have — in tools, skills, and infrastructure — versus what you need? This is where the build/buy/partner decision gets made. Most organisations significantly overestimate their internal AI capability and significantly underestimate the time required to develop it.
- Use Case Prioritisation. List your candidate use cases and plot them on a simple matrix: impact on one axis, feasibility on the other. The cases in the high-impact, high-feasibility quadrant are where you start. This seems obvious; in practice, organisations routinely start with the most visible use case rather than the most tractable one.
- Human-AI Collaboration Model. For each use case, decide explicitly: what does the human do, what does the AI do, and where is the handoff? "Human in the loop" is not a collaboration model — it's a phrase. The collaboration model specifies exactly where human judgment is required, what the AI's output looks like when it arrives at that point, and how a human overrides or corrects the AI when needed.
- Governance. Who owns AI decisions in your organisation? Who can approve the deployment of a new model? Who reviews AI outputs for quality and safety on an ongoing basis? Governance questions feel bureaucratic until something goes wrong, at which point the absence of an answer is a crisis.
- Success Metrics. How will you know if it's working? Define measurable outcomes before you build — not because measurement is the point, but because clear metrics force clarity about what you're actually trying to achieve. If you can't define what success looks like, you have not finished thinking about the ambition.
Running it as a workshop
In practice, I run PESTEL and the AI Strategy Canvas as a two-session workshop — typically with a cross-functional leadership group that includes product, technology, legal, and business stakeholders.
The first session is PESTEL, run as a structured scan. Each dimension gets thirty minutes: a short briefing on what's relevant in that area right now, then group discussion of what it means for this specific organisation. The output is a set of constraints and opportunities that the strategy needs to account for.
The second session is the canvas, working sequentially through the seven fields. The PESTEL output feeds directly into it — the political and legal analysis shapes the governance field, the economic analysis informs use case prioritisation, the social analysis influences the human-AI collaboration model.
The result isn't a finished strategy. It's a structured foundation for one — a shared understanding of the landscape and a set of decisions that a real strategy can be built from. Getting leadership alignment on this foundation, before any use cases are committed to or budgets allocated, is the highest-leverage thing most organisations can do right now.
The alternative — which I see constantly — is announcing an AI strategy that is actually a procurement decision, wondering why adoption is low, and then commissioning a culture change programme to fix a problem that was always a strategy problem. That is an expensive detour. PESTEL and the canvas are not a guarantee against it, but they are a much cheaper way to find out where the real obstacles are.