Claude Prompts for Business Strategy That Work

The short version: Claude is good at business strategy work if you feed it real numbers, real constraints and a clear role, but most people write prompts so vague they get back the same consultancy mush you'd find in a 2015 slide deck. Give it your actual figures, ask for the uncomfortable option too, and treat the output as a first draft from a sharp junior, not a verdict from an oracle.

Why most strategy prompts produce rubbish

"Help me create a growth strategy for my business" is not a prompt, it's a wish. Claude will answer it, because it's built to be helpful, and the answer will sound plausible and say almost nothing. Three pillars. Customer-centric approach. use synergies. You've read it a hundred times because that's what you get when you ask a vague question of a model that has to guess your industry, your budget, your team size and your actual problem.

The fix isn't a magic prompt template, it's information. Strategy work is only as good as the constraints you feed in. Claude can hold a lot of context in one conversation, which means you can paste in your P&L summary, last quarter's numbers, your competitor list and your actual headcount, and it will reason against that rather than against the internet's average idea of a business.

The uncomfortable bit nobody tells you

Here's what doesn't get said in most "AI for strategy" posts: Claude will agree with you. If you ask "is launching a second product line a good idea?" and you've already decided it is, the model will find supporting arguments, because that's the shape of the question you asked. It is not a neutral board member who'll tell you your idea is weak. You have to deliberately ask it to argue against you, and even then it will soften the blow unless you tell it not to. Treat every strategy output as one voice in the room, not the final word, and ask it explicitly to be the skeptic at least once in the process.

Prompts that get you something usable

These are built around a fictional example so you can see the shape of a good prompt. Say you run a 12-person marketing agency doing £900,000 a year in revenue, with three big clients making up 60% of that income, and you're trying to decide whether to specialise in one sector or stay generalist.

• Diagnosis prompt: "I run a marketing agency, £900k annual revenue, 12 staff, three clients make up 60% of income. Before you suggest anything, list the five biggest risks in this business as it stands today, ranked by how much damage they'd do if they happened in the next six months."
• Options prompt: "Give me three strategic options for this agency over the next 18 months: specialise in one sector, stay generalist but raise prices, or build a productised service line. For each, tell me the realistic downside, not just the upside, and what would have to be true for it to work."
• Stress-test prompt: "Take the specialisation option and argue against it as hard as you can, as if you were a competitor trying to talk me out of it in front of my team."
• Numbers prompt: "If I raised prices by 15% on new clients only, keeping existing contracts the same, model what that does to revenue if I lose two of my smallest clients but keep everyone else. Show the maths."

That last one matters more than people think. Claude is surprisingly solid at working through arithmetic step by step when you ask it to show its working, which also lets you catch it if it's made an assumption you didn't intend, like assuming a 12-month contract when you meant month-to-month.

A worked example, step by step

Take that same fictional agency. Here's a sequence that goes from vague worry to a usable decision, roughly how you'd run it in one sitting:

• Step 1: Paste in the real numbers. Revenue, client concentration, staff costs, your biggest fixed costs. Ask Claude to summarise the financial picture back to you in plain English, so you can check it understood the inputs correctly before it reasons from them.
• Step 2: Ask for the five riskiest things about the business as it stands, ranked by impact. This surfaces the client concentration problem explicitly instead of letting it hide in a paragraph about "diversification."
• Step 3: Ask for three distinct strategic responses to that top risk, each with a one-sentence "what would have to go right" and a one-sentence "what would have to go wrong."
• Step 4: Pick the option that worries you least and ask Claude to argue against it as the skeptic, specifically targeting your weakest point (cash flow, team capacity, or client trust, whichever applies).
• Step 5: Ask for a 90-day action plan with named owners if you have a team, broken into weeks, not vague quarters, so you can hand it to someone on Monday.

Run that whole sequence and you'll have spent maybe 40 minutes and produced something closer to a working strategy memo than a slide deck full of pillars. The output still needs a human to sense-check it against things Claude can't know, like whether your biggest client is secretly unhappy or whether your best account manager is about to quit.

What Claude is bad at in strategy work

It doesn't know your market unless you tell it, and it will fill gaps with generic industry knowledge that might be two years out of date or simply wrong for your specific niche. It also has no sense of your team's actual capability, so a 90-day plan that looks reasonable on paper might assume skills nobody in your business has. And it has no skin in the game, so it will happily suggest bold moves because it isn't the one who has to make payroll if it goes wrong.

Use it for structuring your thinking, stress-testing assumptions and doing the maths fast. Don't use it as the final decision-maker, and don't let a confident-sounding paragraph replace a conversation with someone who knows your numbers, like your accountant or a co-founder.

Which Claude model to use for this

For strategy work specifically, the model choice matters more than people assume. If you're just drafting a quick internal memo, a faster model is fine, but for genuine reasoning through options and numbers you want the heavier model. I've written a full breakdown of which Claude model to pick for different writing and reasoning tasks, and separately on when the lighter Haiku model is the right call rather than overkill. For strategy specifically, don't use the cheap fast model and expect depth, it's built for speed, not for arguing with you.

If you're building out recurring strategy reviews rather than one-off questions, it's also worth looking at how Claude's dashboard features work, since you can set up a running view of your key numbers rather than re-pasting your P&L every time you want a check-in.

Where this fits with the rest of your AI stack

Strategy prompts don't live in isolation. If you're already using Claude for writing and are curious where other tools fit, Canva's integration is worth knowing about if your strategy work turns into actual presentations or client-facing material, which I covered in a piece on what you can do with Canva from inside Claude or ChatGPT. And if part of your strategy involves how you talk about the business publicly, the same "give it real constraints, don't accept the first draft" rule applies to social copy too, which I go into in a separate post on prompts for tweets that don't sound like AI wrote them.

If you'd rather not build all this yourself, this is exactly the kind of thing an AI consultant earns their fee doing, turning a messy set of numbers and worries into a working process rather than a one-off chat. I've written about what that costs if you're weighing up doing it yourself versus bringing someone in.

A simple rule for writing your own prompts

Every good strategy prompt I've described has three things in it: a specific number, a named constraint, and an explicit ask for the downside, not just the upside. Miss any one of those three and you're back to generic pillars. Put all three in and you get something you can act on by Monday morning.

Frequently asked questions

Can Claude replace a business strategy consultant?

No, and treating it that way is where people go wrong. Claude is strong at structuring options, doing maths quickly and stress-testing an idea, but it doesn't know your market nuance, your team's real capability, or the office politics that kill good plans. Use it to sharpen your thinking before or after talking to someone who knows the business.

What's the single biggest mistake people make with Claude strategy prompts?

Being vague. Asking "how do I grow my business" gets you generic consultancy language. Pasting in real numbers, naming a specific constraint, and asking explicitly for the downside of each option gets you something usable.

Which Claude model should I use for strategy work?

Use the heaviest model you have access to for genuine reasoning and multi-step arguments, and save the faster, lighter model for quick internal memos where depth doesn't matter as much.

How do I stop Claude from just agreeing with my plan?

Ask it directly to argue against your preferred option as if it were a competitor or a skeptical board member. Without that explicit instruction it will tend to support whatever framing you've already given it.