What should our AI strategy actually say?

Why one page is enough

Many leaders feel they should have a thick AI strategy, and worry that they are behind. A long document is rarely the gap. The gap is clarity: people need to know what the company is trying to achieve with AI, what is allowed and who decides. A 2025 Gallup survey found that only 22% of US employees said their organisation had communicated a clear AI plan or strategy, and those who strongly agreed it had were 2.6 times as likely to feel comfortable using AI in their role.

A single page also helps leaders themselves. It turns a vague worry about falling behind into a short list of decisions you can make, explain to the board and revisit as you learn. It should be reviewed and updated, not framed and forgotten.

Could your leadership team agree its top three AI problems in one meeting? Book a free AI Value Briefing and we will look at it with you.

The seven things your page should say

1. The business problems. Two or three problems that matter: slow quotes, heavy reporting, long response times. Tools come later.
2. Where the value is and how it is measured. For each problem, a measure of time or quality taken before you start (a baseline), so results can be compared honestly.
3. Guardrails and data rules. Approved tools, data that must never be pasted in, and where a person checks the output.
4. Who owns it. One named executive who answers for results, with support from IT, legal and finance.
5. People and skills. Protected learning time, training on real work, and a clear statement of what freed hours will be used for.
6. A small portfolio of first uses. Three to five frequent, low-risk tasks, spread across teams.
7. A review rhythm. Dates where a named person looks at the measures and decides.

Plan for the dip

Economists Erik Brynjolfsson, Daniel Rock and Chad Syverson describe a "productivity J-curve": technologies like AI need investment in skills, new processes and data before they pay back, so early results can look weaker and later results stronger. In practice, a team often slows down while it learns. A strategy that names this in advance stops a normal dip being mistaken for failure, and review dates stop a failing project running on too long (Field Guide 30).

Learning matters more than tools

A 2020 MIT Sloan Management Review and BCG study of more than 3,000 managers found that only 10% of organisations reported significant financial benefits from AI. The authors linked those benefits to changing processes so that people and AI learn together, not to technology alone. That is why people and skills sit in the middle of the page.

What to leave out

A one-page strategy is as useful for what it does not say. Leave out long lists of tools, which date quickly and invite debate about products instead of problems. Leave out promises of fast savings or headcount targets based on forecasts: report measured results instead, and keep estimates clearly labelled as estimates (Field Guide 28). Leave out vague ambitions such as "becoming AI-first" unless you can say what would change on a Monday morning. And avoid making AI a side project for one enthusiast: the owner should be someone who already answers for results in the business.

Frameworks you can borrow

You do not need to invent governance. In the US, the NIST AI Risk Management Framework (2023) is voluntary and organises the work into four functions: govern, map, measure and manage. Internationally, the OECD AI Principles (2019, updated 2024) cover transparency, robustness, safety and accountability. In the EU, Article 4 of the AI Act, as amended by Regulation (EU) 2026/1744, asks providers and deployers of AI systems to take measures to support the AI literacy of their staff, so a line on training also helps you show you have done so. This is general information, not legal advice.

Goes deeper in our free AI Field Guide: 32. Leading AI adoption: what leaders do differently, 30. The AI J-curve: why results dip before they rise, 28. Measuring the value: hours given back, 10. Data safety: what never to paste, and other rules, 26. Guardrails and permissions: what AI may and may not do, 16. Choosing the right level: use what you have before you build.