AI Skills for Strategists: From Research to Recommendations

Strategy work has a shape: gather, structure, interpret, recommend. AI compresses the first two dramatically and is actively risky in the second two — not because it produces obvious nonsense, but because it produces well-argued, internally consistent cases for whatever framing you handed it.

That's the specific hazard for strategists. A weak analysis announces itself. A confidently-structured argument for the wrong question doesn't.

Skill one: frame the decision, not the topic

The most consequential thing you do in a strategy engagement happens before any analysis: deciding what question is being answered.

"Should we enter the German market?" and "What would have to be true for German entry to beat expanding our current accounts?" are different questions producing different work. The first invites a case. The second invites a comparison.

AI will faithfully answer whichever you give it, at length, with structure. It won't tell you the question was wrong. Framing is the highest-leverage input and the one most often skipped.

Skill two: make it interview you first

The best move in AI-assisted strategy work is refusing to let it start.

Before asking for analysis, ask it to ask you questions — what it would need to know to give a useful answer. You'll get fifteen questions. Three will be ones you can't answer, and those three are usually the actual work.

This inverts the normal failure mode. Instead of the model filling gaps with plausible assumptions, it shows you where the gaps are.

Skill three: demand the case against

Ask for a recommendation and you get a recommendation, with supporting arguments, expressed confidently. Ask for the strongest case against your preferred option and you get something far more useful.

Better still: ask what would have to be true for the option you've rejected to be correct. That question tends to surface the assumption you've been treating as fact.

Skill four: separate what it knows from what it's inferring

A strategy output blends three things: things that are verifiably true, things that are reasonable inference, and things that are pattern-matched from similar situations that may not apply.

They arrive in the same voice, formatted identically. Asking the model to mark which is which — explicitly, claim by claim — is a thirty-second step that changes what you'd stake a recommendation on.

Idris — the well-argued wrong question

Idris is a strategy lead at a healthcare group. He was asked to assess whether to build an in-house data platform, and he ran a thorough AI-assisted analysis: build versus buy, cost model, capability comparison, risk assessment. It was good work.

In the steering group, a clinical director asked what problem the platform was solving. The honest answer was that three departments each had a reporting complaint, and nobody had checked whether they were the same complaint.

The analysis was sound. The question was wrong, and no amount of rigour downstream would have caught it — the AI had answered exactly what he asked, thoroughly.

He now spends the first session of any engagement getting the question challenged before any analysis starts, and asks the model to argue that the stated question is the wrong one.

Wren — the fifteen questions

Wren is an independent strategy consultant. Her habit is to give AI the brief and ask what it would need to know before it could answer well.

On a recent pricing engagement it came back with fifteen questions. Twelve she could answer from the brief. Three she couldn't: what proportion of revenue came from customers on legacy terms, whether the sales team had discretion to discount, and what had happened the last time prices moved.

She took those three to the client. Nobody in the room knew the third. Finding out took a week and changed the recommendation — the previous increase had triggered churn concentrated in exactly the segment they were now planning to target.

The model didn't find that. It found the shape of the hole.

The one thing

AI collapses the cost of gathering and structuring, which means the bottleneck in strategy work moves entirely to framing and judgement.

That's good news if you're strong at those and dangerous if you've been relying on the effort of analysis to slow you down enough to notice a bad question. The rigour that used to come for free now has to be deliberate.