How to Use Claude to Run Your Business (Without Wasting Money)

The smartest AI on the planet is now sitting one click away, and most business owners are using it like a fancy search bar. That's a mistake. Used right, Claude does real work for your business - drafts, plans, research, tools your team actually uses. Used wrong, it either gives you weak answers or quietly runs up a bill.

I'm Jonathan Floyd. I run LeadValets, where we get service businesses found on Google and in the AI answers buyers now trust. I test these tools every day, so this guide skips the hype and gives you what to do.

Here's how to use Claude to run your business better.

What Claude can actually do for you

Claude is an AI assistant from Anthropic. You talk to it in plain English and it hands back finished work - not a rough paragraph you have to rewrite.

A few things it does well:

• Build an interactive chart or calculator to understand a number in your business
• Build a spreadsheet - any spreadsheet - with the formulas already working
• Co-write anything after you feed it your notes and files
• Build a slide deck or PowerPoint
• Run deep market research across dozens of sources and hand back a decision doc

The ceiling is much higher than "rewrite this email." That's the part most owners never see.

There isn't one Claude - there are four

Claude comes in different levels of intelligence, and you pick the one you talk to. This is where owners either get weak answers or a surprise bill, so it's worth 60 seconds.

From cheapest to smartest:

• Haiku - the fastest and cheapest. Good for simple, high-volume stuff.
• Sonnet - the middle option. Solid quality, low cost.
• Opus 4.8 - very smart, and by most benchmarks cheaper and stronger than Sonnet for real work. This is my daily driver.
• Fable 5 - the smartest model on Earth right now. Also the most expensive, because you pay per use.

The easiest way to picture it: Fable 5 is a $1,000-an-hour senior partner. Opus is a sharp $150-an-hour associate. You don't put the senior partner on routine paperwork. You put him on strategy, then hand the plan to the associate to execute. Do that and Claude costs you a little. Ignore it and it costs you a lot.

Don't run everything through Fable 5

Fable 5 is billed per use. The cost builds up two ways: by "turns" (your message plus Claude's reply) and by length. And every time you send a message, Claude re-reads the whole conversation - so a long back-and-forth gets expensive fast.

To put numbers on it: a quick question and answer is a few cents. A long thread of twenty exchanges can run into real money. Push it to forty turns and you'll feel it.

The fix is simple:

• Start a new chat for each new job. Don't let one thread run all afternoon.
• Ask Fable your hard question in one or two turns - the strategy, the plan, the tough call.
• Switch the model to Opus 4.8 for the drafting, the edits, and the follow-ups. Opus is included in your plan, so it doesn't run the meter.

If you have a team, this matters even more. A dozen people leaving Fable running all day with no guidance is how you wake up to a bill that stings. Go into your organization settings and set usage limits before you roll it out to everyone, not after.

Short version: aim Fable at expensive problems only. Paying per use to "rewrite this email" is shooting a bird with a bazooka.

The one habit that changes everything: give it goals, not tasks

For years we all learned to spoon-feed AI. "Rewrite this." "Make it shorter." "Summarize that." Small steps, because the old models lost the plot after a few.

Claude doesn't work that way anymore. It plans. It asks you questions before it starts. It pushes back. It checks its own work before handing it over - like an employee who re-reads before hitting send. So stop feeding it tasks and start handing it goals.

A task (weak): "Write a follow-up email to this client."

A goal (strong): "This client owes me two invoices and went quiet three weeks ago. Here's the full thread. Goal: get paid without burning the relationship. Plan your approach, draft whatever's needed, and ask me what you don't know."

Same AI. Completely different result. The second one hands you a strategy and the email - and it flags the thing you forgot.

Four lines I add to almost every serious prompt, because they fix Claude's most annoying habits:

• "Lead with the answer." Put the bottom line in the first sentence, details after. No wading through the setup to find the point.
• "Only tell me what you can back up. If something isn't verified, say so." Kills the fake "all done!" confidence on longer jobs.
• "I'm thinking out loud - don't write anything yet" or "Handle this end to end, only check with me if something's irreversible." Tell it whether you want advice or action, or it'll write the whole email when you only wanted feedback.
• "When you have enough to act, act. Give me a recommendation, not a menu." For when you want an answer, not ten options.

You're not prompting anymore. You're setting rules and handing over a goal. Claude figures out the how.

Four jobs worth handing Claude this week

These are the prompts I'd actually use to run a business. Start each one in Fable 5, then switch to Opus 4.8 once the thinking is done.

1. Stress-test a big decision

I'm about to make a significant decision and I want it pressure-tested before I commit, because reversing it later will be expensive. Ask me for the decision, the options on the table, my priorities in order, my hard constraints, and my deadline. Interview me a few questions at a time until you understand it better than I've explained it. Then give me your one-sentence recommendation, how each option scores against my priorities, the best/base/worst case for your pick, the strongest argument for the option you rejected, and the single most likely way you're wrong. Don't hedge with "it depends" - I'll have told you what I value.

2. Research you can act on

I need a research report I can make a decision from, not a survey of everything. Ask me the question I'm answering, the decision it feeds, and any sources I already trust. Then research it end to end. Open with the answer in three sentences, then support it. Cite a source for every real claim. Where sources disagree, tell me which you'd trust and why. Before you deliver, attack your three biggest claims like a skeptic paid to find the flaw, and fix anything that doesn't survive. Flag anything you can't verify. Keep it under two pages.

3. Turn your messy week into a plan

I'm going to paste the raw exhaust of my work week - notes, message threads, to-do lists, half-finished docs. Turn it into an operating plan for next week. Give me four sections: Commitments (who owes what to whom and by when, flag anything overdue), Decisions (what was actually decided vs. just discussed), Conflicts (where two people or docs point opposite ways - quote both), and Next Week (the five actions that most move my top priority). Ground everything in what I gave you. Don't invent tasks that aren't in the material.

4. Write in your voice, not "AI voice"

Here are three things I've written [paste them]. Study my rhythm and phrasing. Then help me write [the piece], for [audience], to be published on [where]. Match my voice, not a polished corporate style. Flag any place you're asserting something my material doesn't support so I can add proof or cut it. Don't invent quotes, stats, or stories to fill gaps - leave a visible gap instead.

That last one matters. AI-sounding content is a liability now - customers can smell it, and it flattens the exact personality that makes people choose you over the next guy. Feed Claude your real writing and it sounds like you, not like everyone.

The real unlock: give Claude your business, not just prompts

Here's what nobody tells owners. Claude read most of the internet. It has never read your business - your clients, your pricing logic, your deadlines, the way you talk. Out of the box, it answers like an average of everyone. Generic in, generic out.

You fix that with context, and there are two easy ways to do it:

• Projects - a folder inside Claude that remembers your files and instructions across every chat. Drop in your service list, your best case studies, your pricing, your brand voice. Now every answer starts from your business instead of a blank page.
• Connectors - plug Claude into the tools you already use: your inbox, your CRM, your calendar, your notes, your drive. Now it works from what's actually happening this week instead of guessing.

The owners getting real results aren't better at prompting. They've just handed Claude more context. That's the whole game.

Two more tools worth knowing by name. Cowork lets Claude work directly with the files and folders on your computer and hand back finished spreadsheets, decks, and documents. Claude Code is the version developers use to build software - overkill for most owners, but it's why people now build custom tools without hiring an engineer.

The bottom line

Most business owners are still typing "rewrite this email" into an AI like it's 2022. The ones pulling ahead are handing it goals, feeding it their real business, and letting it do a day's work in an hour.

Here's your move: block twenty minutes this week. Open Claude, pick one prompt above, and run it on something real - a decision you're stuck on, the mess from last week, a piece you've been putting off. Use Fable to think, switch to Opus to finish. You'll see it inside one session.

That's how you go from playing with AI to running your business with it.

- Jonathan Floyd, LeadValets

References

• Anthropic - Claude Models Overview: https://platform.claude.com/docs/en/about-claude/models/overview
• Anthropic - Claude Platform Pricing: https://platform.claude.com/docs/en/about-claude/pricing
• Anthropic - Claude Cowork: https://www.anthropic.com/product/claude-cowork
• Anthropic - Get Started with Claude Cowork: https://support.claude.com/en/articles/13345190-get-started-with-claude-cowork
• Anthropic - Claude Code: https://claude.com/product/claude-code
• Anthropic - Prompt Engineering Overview: https://docs.claude.com/en/docs/build-with-claude/prompt-engineering/overview
• CloudZero - Anthropic Claude API Pricing 2026: https://www.cloudzero.com/blog/claude-api-pricing/