Outgrace
OutgraceThe Leadership Magazine, home
StrategyQ&A primer

Using AI in long-range strategy work, answered

Where AI tools sharpen strategic analysis, where they mislead, and which published risk frameworks to set around them.

Most strategy teams have now tried an AI assistant on a market model, a scenario draft or a competitor memo. Far fewer have written down what the tool may be trusted with. The questions below are the ones a strategy team is most likely to raise. The short answers rest on three published sources: the US National Institute of Standards and Technology, the UK government and the European Commission. None of them addresses long-range strategy directly, so wherever we move from what the sources say to what we would do, we say so.

Is there an established framework we can point to for AI risk?

The most detailed public one we found is the AI Risk Management Framework from the US National Institute of Standards and Technology, released on 26 January 2023. It is intended for voluntary use, and it aims to help manage risks to individuals, organisations and society. NIST built it through an open process: a request for information, several drafts, public comment and workshops. It also publishes companion material, including a Playbook, a Roadmap, a Crosswalk and Perspectives. Treat the framework as a list of questions to ask, not a certificate to obtain.

Does a voluntary framework still matter to a company that is not obliged to follow it?

Yes, for two reasons. The framework is designed to help organisations build trustworthiness into the design, development, use and evaluation of AI. It is also being revised: NIST says version 1.0 is being updated under the White House AI Action Plan. Name the version you use in every board paper, and check whether a revised text has appeared before the pack goes out. A paper that cites an unversioned framework invites the first hard question and leaves no answer to it.

What is different about generative AI?

NIST published a generative AI profile, NIST-AI-600-1, on 26 July 2024. It is meant to help organisations identify risks unique to generative AI, and it proposes risk actions matched to an organisation's own goals and priorities. For strategy teams this is the more relevant document. In our experience, many of the tools used to draft scenarios, summarise research or size markets are generative. A profile does not tell you whether an answer is true. It tells you which failures to look for.

Does the EU AI Act apply to a tool we use for internal analysis?

It depends on what the tool does and where the company operates. The European Commission describes the AI Act, Regulation (EU) 2024/1689, as the first comprehensive legal framework on AI worldwide. It sets risk-based rules for developers and deployers, and it defines four levels of risk. Minimal-risk AI, such as spam filters, is not subject to its rules. Transparency rules take effect in August 2026, including the requirement to tell people when they are interacting with a machine. Whether a particular internal use is in scope is a question for counsel, not for the strategy team.

Which uses need the most scrutiny under EU rules?

The Commission's page we hold does not settle this. It was cut off before its list of high-risk use cases and its remaining guidance. What it does say is that high-risk AI systems face strict obligations from 2 December 2027. Read the full list on the Commission's site before you decide. Our judgement is that any tool which informs a decision about people, such as hiring, needs the strictest treatment. The Commission uses hiring decisions as its own example of why an opaque system can disadvantage someone without anyone noticing.

Do the UK rules add anything for a UK board?

The UK government's white paper sets out a pro-innovation approach to regulating AI. It states aims of growth, innovation, investment and public trust in AI, and its consultation closed at 11:45pm on 21 June 2023. The material we hold does not set out binding duties for a strategy team. The working governance therefore has to be written by the company itself. A UK board should not read the absence of a statute as the absence of a standard to meet.

Can an AI tool forecast for us?

None of the sources we hold addresses AI in long-range strategy or forecasting accuracy. So nothing in our set supports or rejects the idea. Our judgement is simple: an AI forecast is a hypothesis with a source, not a prediction. Ask the team to log every forecast the tool produces, date it, assign an owner and check it against outcomes at a fixed interval. A tool that has never been checked has no record that anyone can rely on.

Why can't we simply ask the tool to explain its reasoning?

The European Commission notes that it is often not possible to find out why an AI system has made a decision or prediction and taken a particular action. That opacity, it says, makes it difficult to assess whether someone was unfairly disadvantaged. A fluent explanation from a tool is more text the tool has generated. It is not an audit of how the answer was reached. Ask instead for the inputs, the sources and the assumptions, and check each of them yourself.

Who should own the governance?

A named individual, not a committee. In our experience the strategy function is the heaviest user of these tools and the least likely to have a designated risk owner. Give the role to someone who can say no to a tool, usually the chief risk officer or the finance lead, with a written line of reporting to the board. The owner should be able to pause a use case without first persuading the team that produced it.

What should a written policy contain?

Keep it to one page at first. List the tools the company permits, the uses permitted without sign-off, and the uses that need prior approval, such as anything touching decisions about people or money. Add a rule on how AI-assisted material is labelled in board papers, the forecast log, and a simple route for reporting a failure. The European Commission says deployers must ensure human oversight and monitoring, and that providers and deployers report serious incidents. Those two duties give a useful shape to the policy, even where the law does not apply to you.

Does the critical infrastructure work matter to us?

Only if you operate critical infrastructure. NIST released a concept note on 7 April 2026 for an AI RMF profile on trustworthy AI in critical infrastructure. The profile guides operators on risk practices when they engage AI-enabled capabilities. If you are not an operator, read it for ideas about controls, not as a set of obligations.

Are these frameworks enough for long-range strategy itself?

No. They concern risk to individuals, organisations and society. They do not tell you whether a scenario is plausible, whether a market model is right, or which option to choose. Use the frameworks to set guardrails around the tools, and keep your own method for the analysis. The tool may draft the scenarios. A person with the authority to reject them should decide which ones the board sees.

The mistake behind most of these

Most of these questions come down to one mistake: treating a tool's fluency as evidence. The output reads like analysis, arrives quickly and carries no visible doubt, so teams stop checking. The published frameworks point the other way. NIST describes its framework as voluntary and aimed at trustworthiness across design, use and evaluation. The European Commission treats opacity as a reason to look harder, not a reason to trust more. Strategy work is especially exposed because its errors are slow to appear. A weak market assumption can sit in a plan for a long time before anyone is forced to test it.

Start this week with three steps. Name the owner. Write a one-page list of the tools and uses the team already runs, and mark every use that touches decisions about people or money. Then open a forecast log with dates and an owner for each entry. Before the next board pack goes out, ask counsel to confirm which EU or UK duties, if any, apply to your specific uses. A note on sourcing: the OECD's AI principles are not covered here, because they were not among the sources we could check for this piece. Add them before relying on this article for any international policy.

Newsletter

The Briefing

One email a week: the leaders we profiled, the ideas worth stealing, and nothing else.

Weekly, free, one click to leave. Read our privacy policy.