Ask any decent AI to write a strategic plan for your business and you’ll have one in about two minutes.
It will have an objective, four or five goals with plausible-looking numbers, strategies in the right format, measures with owners, initiatives with quarters against them. It will look better than most of the strategy documents I’ve seen inside real companies.
And it will be worthless.
Not wrong, exactly. Worthless in the specific sense that nobody will change what they do on Monday because of it.
That isn’t a limitation of the technology. It’s a fact about what strategy is. Once you accept it, the question stops being can AI write my strategy and becomes the useful one: what is AI actually good for here?
What is AI genuinely good at in a strategy session?
More than the sceptics allow. Four things, and they’re not small.
It has no stake in the outcome. The big one. In every strategy session I’ve sat in, some part of the room is defending something — a product line someone built, a market someone owns, a headcount someone fought for. An AI has no career riding on Strategy 3, and has never had lunch with the person whose favourite project should be cut. Ask it what should stop and you get an answer, not a negotiation.
It has infinite patience for the uncomfortable question. A human facilitator asks “and what would you have to stop doing?” twice, reads the room, and moves on to keep things pleasant. An AI asks a fourth time in the same even tone. That’s not stubbornness, it’s the absence of social cost — and the fourth asking is usually where the honest answer lives.
It recognises patterns. A goal of 28% growth sitting above three cost-reduction initiatives doesn’t add up. A strategy every competitor would also write hasn’t chosen anything. An empty “we will not” column means the hard conversation was skipped. These are the common OGSM mistakes, and they are almost always thinking mistakes rather than template mistakes — which is exactly the kind of thing a machine spots quickly.
It holds a consistent standard. Underrated, this one. Humans get tired. By hour four of a workshop the fifth strategy gets waved through with less scrutiny than the first. An AI applies the same bar to the last item as the first, at 9pm on a Thursday, on the fourth evening.
That’s a genuinely useful colleague. A fast, well-read sparring partner who will ask the same hard question on a Sunday night as on a Tuesday morning, and who has no stake in which way you answer it.
But it hasn’t lived through your last five years of business.
What can AI not do?
Four things, and each one is load-bearing.
It doesn’t know your market. It knows what has been written about your market, which is a thinner thing. It doesn’t know the competitor everyone’s worried about has quietly lost their best engineer.
It doesn’t know your customers. It knows archetypes. It doesn’t know your three biggest accounts all came from one person’s relationships, and that person is sixty-one.
It doesn’t know your team’s real capacity. It can count initiatives against headcount. It cannot know Dan has been carrying two roles since March, or that ops said yes to the last three plans and delivered one.
It doesn’t know what happened last time you tried this. Every business has a graveyard — the segment you entered in 2021 and quietly exited, the pricing change that lasted six weeks. That history is the most useful input to any strategy discussion, and it exists nowhere except in the heads of the people in the room.
A strategy that ignores those four things is a well-formatted guess.
Why is a two-minute plausible OGSM the worst possible outcome?
Because plausibility is the trap.
An obviously bad plan gets rejected. A clearly generic one gets sent back. But a plan that looks right — right format, right vocabulary, sensible-looking numbers — gets adopted. It goes on a slide. It gets presented at an all-hands. And then it sits there doing nothing, because underneath the format there was never a decision.
Strategy is a choice. A specific direction, paid for by everything you decide not to do.
An AI can generate the appearance of a choice instantly. It cannot make one, because making a choice requires having something to lose.
This is why most strategies fail: not because leaders aren’t smart enough, but because they mistake aspiration for direction. AI-generated plans are aspiration at industrial scale — beautifully formatted, internally consistent, and nobody had to give anything up to produce them.
Why does a strategy nobody chose become a strategy nobody defends?
Here’s the test that matters. It’s June. A good customer asks for exactly the kind of work your strategy says you’ll stop doing. Real revenue, this quarter, sitting on the table.
Who says no?
Only someone who made the choice. Not someone who received it, agreed with it in the meeting, and can’t quite reconstruct the reasoning six months later. Agreement is not alignment.
A plan you nodded at is a plan you’ll make an exception to. And strategies don’t fail dramatically — they drift, quietly, one reasonable exception at a time, until the gap between where you said you’d be and where you actually are is too wide to close without a serious conversation.
The defence of a strategy happens in small moments, by people who remember why. If the reasoning lives in a document and not in a person, there’s nothing to remember.
That’s the argument for facilitation over generation. Not that a facilitated plan is better written — often it’s worse written. It’s that a facilitated plan is owned, and ownership is the only thing that survives contact with a tempting exception.
Why is the discomfort of the strategies phase the point?
In the OGSM Strategy Builder, the strategies phase is the longest and designed to be the least pleasant. It generates more options than you’ll keep and then forces the cut. It insists on the “we will NOT” column. It asks what you’d have to stop, and keeps asking.
People sometimes read that as friction badly designed. It’s friction deliberately designed.
Good strategy always feels uncomfortable. Because it forces you to say no.
If the whole session was comfortable, something was avoided — and it’s always the same thing: the trade-off that costs a specific person something specific. That’s the moment the plan becomes real, and exactly the moment a generative tool skips, because nothing in it requires generating.
A strategy that cannot tell you what to say no to is not a strategy. It is a wish list dressed in a slide deck. An AI will happily produce the slide deck.
Where must the human decide?
Four places, and no tool should take them from you.
The objective. An AI can draft three genuinely different bets and show what each rules out — the craft of it is in how to write a great objective. Which one is your business is not a question it can answer.
The trade-offs. Which customer gets less attention. Which product line stops. Which market you’re not entering. These land on people, and the person who has to live with it has to be the one who says it.
The numbers you’d be embarrassed to miss. An AI can tell you a goal is unfalsifiable. It cannot tell you whether 92% is ambitious or safe in your business.
Who owns what. A measure without an owner is a wish, not a commitment — and only you know who will actually answer for it at the monthly review.
Everything else — the structure, the questions, the quality checks, the format, catching activity metrics dressed as outcomes — hand it over. That work is real, it’s tedious, and machines are good at it.
What about Doubt and Confusion — the ones no document fixes?
Two of the Seven Cardinal Sins make the point better than I can.
Confusion is not understanding — a strategy no one can repeat. If the people who have to deliver it can’t tell you what it is, you don’t have a strategy. You have a secret. An AI can write a clearer sentence. It cannot stand in front of your team and say it, twice a month, until it sticks. Communication isn’t a document problem; it’s a leader-showing-up problem.
Doubt is not believing — the plan is sound, the people aren’t sold. The most expensive of the seven, and no amount of writing touches it. You cannot communicate your way out of doubt. Someone who disagrees has heard you and disagrees. Their team watches what they do, not what’s on the wall.
A good AI process can name both, which is worth something — for most of the seven, naming is most of the cure. Not for these two. Here naming is where the job ends, and it should say so. That’s why the stress test always closes with what it cannot verify, including this: is there anyone who has to carry part of this who doesn’t believe in it?
An AI strategy audit produces observations, not decisions.
So what should you actually use AI for?
Use it as a facilitator. Let it run the process, hold the standard, ask the question you’d rather not answer, catch the goal without a baseline and the strategy without an owner. Let it draft options — three real ones, not three wordings — and choose yourself.
Don’t use it as an author. A plan you didn’t build is a plan you won’t defend, and an undefended strategy is a document with good posture and nothing behind it.
The best strategy is still made by people who understand the business. AI helps them think more rigorously about what they already know.
If you want that process rather than the two-minute version, the OGSM Strategy Builder is in the shop — eight phases, and it will make you choose at every one of them.
Rock on.
Leave a Comment