It is genuinely useful at both ends: gathering the inputs and drafting the wording. It is useless in the middle, because strategy is a set of refusals and no model will refuse something on your behalf.
The honest version of this question is not which tool. It is which part of planning you are asking AI to do, because planning is three jobs and they have completely different answers.
The three jobs
| Job | Can AI help? | What actually happens |
|---|---|---|
| Gather the inputs | Yes, substantially | Summarises research, customer notes, last year's results |
| Decide | No | It will argue for whichever option you argue for |
| Keep it alive | Yes, at the edges | Drafts updates, flags what has not moved |
The middle row is the one every vendor demo skips. A model asked whether you should enter a new market will produce a thoughtful case for entering. Ask it the opposite and it will produce an equally thoughtful case against. That is not a flaw in the model, it is what agreeable text generation does, and it is the exact opposite of what deciding requires.
Strategy is mostly a list of things you have refused. Nothing in a chat window will refuse on your behalf, and if it did you should not accept it.
The four uses that genuinely work
Summarising inputs. Six customer calls, last quarter's numbers, a competitor's pricing page. Reading and condensing is the strongest thing these models do, and the downside of a miss is small.
Drafting the wording. Turning "we need to be better at onboarding" into three candidate objectives with measurable key results. Speed, not insight, and speed is worth plenty when a leadership team is staring at a blank page.
Arguing the other side, when you ask explicitly. Prompt it to make the strongest case against the plan you have just written. It will, and it will usually find two things nobody in the room wanted to raise. This is the single most undervalued use.
Writing the update nobody wants to write. The fortnightly progress note, drafted from what changed. Low risk, genuinely tedious work, and the thing that decays first when people are busy.
What to be careful with
A plan that only says yes. Generated plans are almost always additive. They list what to do and never what to stop. If the output has no refusals in it, the hard half of the work has not been done.
Confident numbers. Ask for a market size or a benchmark and you will get one, stated plainly, often wrong. Anything numeric needs a source you can open.
Mistaking fluency for agreement. A well-written plan reads like consensus. It is not consensus. The test is still whether five people can name the top three priorities without looking, and that has nothing to do with how the document reads.
The part that decides whether any of it mattered
Sull, Homkes and Sull showed in Harvard Business Review in 2015 that strategies fail in execution far more than in conception, and that the commitments which fail most are the ones spanning several teams, because nobody owns them.
No amount of AI in the planning phase touches that. A better-worded plan with the same ownership gaps fails in the same way, slightly more elegantly. The mechanics are in why strategy execution fails.
Which is where the tooling question actually lands. Empiraa GPS is a strategy and accountability platform: one named owner per commitment, measures that can move down, progress collected rather than chased, and ANI built into the platform rather than sold as a separate seat. The AI sits where it is safe, drafting and summarising on top of a current record, rather than where it is not.
Related reading: AI prompts for strategic planning that are worth running, and what agentic AI changes about strategy execution.
Strategy execution has a definition on this site, and strategy execution software covers the category.
