Skip to content

Glossary

AI strategy

An AI strategy states where an organisation intends to use AI and where it does not, and the second half is what makes it useful.

In plain terms

A document saying what the organisation will and will not do with AI. Most of them contain only the first half, expressed as ambitions nobody could disagree with, and a document nobody could disagree with has not decided anything. The useful ones are shorter and contain refusals, because a refusal is the only part that changes what happens on a Tuesday.

01

Why it matters

Because in the absence of decisions, every individual purchase becomes its own argument, and the arguments are settled by whoever is most confident rather than by anything agreed. A strategy that rules things out saves that conversation repeatedly, which is a concrete saving of time and goodwill. One that only expresses enthusiasm adds a document without removing a single question.

02

How it works

The negative half is what carries the weight. Where the organisation will not use AI, what material must not go into it, which decisions stay with a person: these settle questions in advance and are what somebody can actually act on. Ambitions about transformation settle nothing, which is why they are easy to agree and easy to write.

It has to name who decides what is not covered, because the interesting cases are always the ones nobody anticipated. A strategy is a set of decisions made in advance and the world produces situations outside them continuously, so the route for handling the unanticipated case is worth more than another paragraph of coverage.

It works best written after some real use rather than before any. A strategy composed by people who have not yet used these tools describes what they imagine, and the imagining is usually wrong in both directions: fearful about things that turn out fine and relaxed about things that turn out awkward. A few months of actual experience makes the document considerably more specific.

The test is whether reading it would change anybody's behaviour this week. If a colleague could read the whole thing and continue exactly as before, it is a statement of position rather than a strategy, and it will be treated as one. That test takes a minute and is the most reliable thing available in a subject that attracts a great deal of writing.

Two documents, one page each

Two documents, one page eachThe left column is not the product of carelessness. Each line is defensible, it survives review by every department, and it commits the organisation to nothing it might later regret, which is exactly why it gets written and approved. The right column will be argued about, because each line takes something away from somebody, and that argument is the whole value of the exercise. What is worth noticing is that the right-hand version is not longer or more sophisticated; it is simply specific enough to be wrong, which is the property that makes a document capable of guiding anything. An organisation with the left-hand page still settles every real question by whoever is most confident in the room, and has spent the effort of a strategy process to arrive there.Changes nothingWe will be an AI-enabledorganisation.We will build capability acrossevery team.We will use AI responsibly.Changes somethingClient deliverables arereviewed by a person, always.Customer records do not go intounadministered tools.Anything unanticipated goes tothis named person.Nobody would disagree with theleft column, which is preciselythe problem: a document thatcannot be disagreed with has notdecided anything, and it will befiled rather than used.
The left column is not the product of carelessness. Each line is defensible, it survives review by every department, and it commits the organisation to nothing it might later regret, which is exactly why it gets written and approved. The right column will be argued about, because each line takes something away from somebody, and that argument is the whole value of the exercise. What is worth noticing is that the right-hand version is not longer or more sophisticated; it is simply specific enough to be wrong, which is the property that makes a document capable of guiding anything. An organisation with the left-hand page still settles every real question by whoever is most confident in the room, and has spent the effort of a strategy process to arrive there.
03

Seen in the wild

  • Deciding that a general assistant is sanctioned for internal drafting but not for anything going to a client unreviewed.

    ChatGPT

  • Deciding that unattended automation may summarise and route records but may not send anything externally without a person.

    n8n

  • Deciding that anything reaching across internal systems is administered centrally rather than adopted by a department.

    Glean
04

Common misconceptions

People assume

We need a strategy before we start.

In fact

Usually the reverse. A document written before anybody has used these tools describes what was imagined, and the imagining tends to be wrong in both directions. A few months of real use produces a far more specific and more useful document, and the interim can be covered by a short policy about material rather than a strategy.

People assume

A strategy is about which tools we will buy.

In fact

Tool choices date within a year and the decisions worth writing down do not. What material may be used, what stays with a person, who decides the unanticipated case: those hold while the products underneath change completely, which is what makes them worth the effort of agreeing.

05

Telling them apart

AI strategy vs AI policy

AI strategy

Where the organisation intends to use this, and where it will not.

AI policy

The rules for using it: what may go in, what must be checked.

The strategy chooses the ground; the policy governs conduct on it. Most organisations need the policy sooner.

06

Questions

What makes one useful?
Refusals. Naming what the organisation will not do, what material must not be used, and which decisions stay with a person settles questions that would otherwise be argued case by case. A document containing only ambitions has added a paragraph without removing a single conversation.
How do we tell whether ours is any good?
Ask whether reading it would change what a colleague does this week. If somebody could read it entirely and carry on unchanged, it is a statement of position, and it will be treated as one however carefully it was written. The test takes a minute and is the most reliable one available.
When should we write it?
After some real use rather than before. Written first, it describes what was imagined and is usually wrong in both directions. Written after a few months, it is specific about things that actually happened. A short policy about permitted material covers the interim and is the more urgent document anyway.
07

Key takeaways

  • The refusals are the part that does any work.
  • Name who decides the case the document did not anticipate.
  • Written before any real use, it describes what was imagined.
  • If reading it changes nobody's week, it is a position rather than a strategy.
08

Related terms

09

Tools that use this

  • ChatGPT

    Sanctioned for internal drafting, not for unreviewed client work.

  • n8n

    May summarise and route, may not send externally without a person.

  • Glean

    Anything reaching across systems is administered centrally.

Last checked July 2026

All glossary terms