AI can do the groundwork for a strategy: gather the basis, work through it systematically, and lay out the options. It cannot make the choice, because the choice is about what risk you will accept. The gain sits in the groundwork, which is exactly what usually gets skipped.
The question arrives in two versions. Either whether AI can write the strategy, or whether the whole thing is overblown.
Both versions miss where the gain actually is.
Strategy work consists of groundwork and a choice. The groundwork is heavy, tedious, and therefore rarely done properly. The choice is short, hard, and requires somebody to carry the responsibility.
AI moves the groundwork. The choice stays where it is.
What AI is genuinely good at here
Gathering the basis. Market data, competitors' public material, your own numbers and documents. That work takes a person several days and usually gets done halfway.
Working through it systematically. A human looks at a situation through the lens they are best at. A structured pass looks at the same situation from fifteen angles without getting bored. That is not intelligence, it is stamina, and it is what most analyses lack.
Laying out the options. Given the basis: which three to five routes exist, and what does each require?
Keeping track of whether it happens. After the decision, which is where most strategies quietly die.
What it cannot do
Choose. An analysis can say that if you do A, B probably follows. It cannot say you ought to do A. That question is about values and risk appetite.
Know what is written nowhere. That a customer segment exists only because your largest customer asked for it. That a key employee leaves if you close a department. That kind of thing is not in the spreadsheet, and the analysis does not account for it.
Handle the unexpected. If regulation shifts or a crisis lands, the system still follows the rules it was given, and those rules were written for the world that was.
Have the conversation. It can prepare you thoroughly for a difficult conversation. It cannot conduct it.
Those four limits are not temporary. They are why the human stays in the middle. Developed in what AI agents can actually do.
Decide how much it does alone
This is the decision that determines whether it works, and it often never gets made explicitly.
There are three levels: insight only, where you read and decide. Proposal with approval, where a human says yes before anything happens. And full automation, where the system acts within fixed rules.
The choice is made per task and turns on two things: what a mistake costs, and how stable the rules are. Strategic choices always belong in the first. Data collection can sit in the third. Most things sit in the middle.
Note that this is not about how capable the model is. A better model does not move an acquisition down to the third level, because precision is not the problem.
What it requires of you
Three things, and none of them are technical.
Classify your data first. What may be sent, and what never? Without that answer the AI question stays stuck in a matter-of-principle debate while employees use a free account for the same thing. The method is in the four classification levels.
Decide what gets documented. Which model, which data in, what came out, what you used. Four lines per analysis. That is substantially the whole obligation under the EU AI Act for a company using AI for analysis rather than for decisions about people.
Agree a ceiling. A limit on what may be spent between two human looks. Not as a prohibition, but so that a setup starting to behave differently gets noticed within a month.
A week that produces an answer
Day 1. Pick one recurring analysis you already do and nobody looks forward to. Competitor overview, customer feedback, margin development.
Day 2. Run it on purely public data. Competitors' annual reports, market figures, your own website. No personal data at all.
Day 3. Read the result with whoever usually does the analysis. The question is not whether it is impressive. It is whether it is right, and whether it contains anything you did not know.
Day 4. Write down what it missed. This is the most important day, because it is where you learn what the groundwork has to be supplemented with.
Day 5. Decide whether the task runs this way from now on, and what that saves.
Five days produces an answer for your company. That is faster than reading about it, and the answer actually applies to you.
How it fits with everything else
AI does not change what a strategy is. The work still falls into three phases: overview, development, execution, and the first is still the one that gets skipped. The phases are in the strategy chain.
What AI changes is the price of the overview. When the groundwork costs hours instead of weeks, it gets done, and then the choice rests on something.
How the work divides concretely, and why the order matters, is in from chatbot to strategic department.
What the board will ask
Sooner or later the item reaches the agenda. Three questions establish whether the vendor has thought it through: where does our data physically sit, can we see what the system built its answer on, and may we choose something other than the recommendation.
If those cannot be answered clearly, the system is not ready for your decisions. The three demands are in three things your board should demand.
The honest answer to whether AI can write your strategy is no. The honest answer to whether it is overblown is also no. It can do the groundwork properly, and the groundwork is what usually goes missing.