The Business Playbook

The difference between a mediocre answer and a great one is usually the prompt. A strong business prompt has five parts. Say them in plain English and Claude does the rest:

ROLE + CONTEXT + TASK + FORMAT + CONSTRAINTS

Watch it in action

A popular, fast-paced tour of prompting techniques you can apply to every business task.

"Google's 9-Hour Prompt Engineering Course in 20 Minutes" — by Tina Huang · watch on YouTube

You don't have to write these five parts in any special order, and you don't need perfect grammar or fancy wording. Think of it less like programming and more like briefing a brand-new employee on their first morning: the more clearly you explain who they are, what your business is like, what you need, and what "good" looks like, the less they have to guess. The formula is simply a checklist that stops you from leaving out the parts people forget most — usually the format and the constraints.

A Closer Look at Each Part

Each part of the formula does a specific job, and a small tip for each one will noticeably lift the quality of what you get back.

• Role sets the point of view. Naming a role — "you are our bookkeeper," "you are a seasoned copywriter" — pulls the answer toward the right expertise and tone. Tip: be specific about seniority and specialty. "A detail-obsessed contracts paralegal" gets a more careful answer than "a lawyer."
• Context is the background only you know: who your customers are, what you sell, what stage you're at, what happened last week. Tip: include the "why" behind the task. "This email goes to lapsed customers we haven't seen in six months" changes the whole approach.
• Task is the single, concrete thing you want done. Tip: use one clear action verb — write, summarize, compare, clean, draft, list. If you catch yourself asking for three things at once, split them into three prompts and you'll get sharper results on each.
• Format describes the shape of the answer: a table, five bullets, a 120-word email, a subject line plus body, a checklist. Tip: if you know roughly how long you want it, say so — length is the setting people forget most, and it's the easiest one to fix up front.
• Constraints are the guardrails: the tone, the things to avoid, the policies to respect, the reading level. Tip: phrase these as clear "do" and "don't" rules — "keep it warm and plain-spoken; no jargon, no hype, no discounts" — so there's nothing to misread.

Weak Prompt vs. Strong Prompt

Here's the same job asked two ways. The weak version leaves Claude guessing; the strong version uses all five parts and gets a usable result on the first try.

Weak prompt (vague)

Write a marketing email.

Strong prompt (the formula in action)

ROLE: You are our email marketer.
CONTEXT: We sell handmade candles to eco-conscious buyers who care about natural ingredients and sustainable packaging.
TASK: Write a launch email for our new autumn scent, "Amber & Ash."
FORMAT: A subject line, then about 120 words of body copy, then one clear call-to-action button label.
CONSTRAINTS: Warm, friendly tone. No discounts or urgency gimmicks. Mention the natural soy wax.

Notice what the strong version did: it told Claude who to be, gave it the customer and product it's selling to, named exactly one deliverable, spelled out the structure down to the button label, and set the tone and the no-go list. There's almost nothing left to guess, which is why the first draft comes back usable instead of generic.

One more before-and-after

The same pattern rescues everyday internal tasks, not just marketing. Watch a vague ask become a finished document.

Weak:  Summarize this meeting.

Strong:
ROLE: You are my executive assistant.
CONTEXT: This is the transcript of our Monday leadership meeting. The team is small and busy, so they skim.
TASK: Turn the transcript into meeting notes.
FORMAT: A 2-sentence summary at the top, then "Decisions" as bullets, then "Action items" as a table with owner and due date.
CONSTRAINTS: Only include things that were actually decided or assigned. If an owner or date is unclear, mark it "TBD" rather than guessing.

Five Reusable Templates

Copy these, fill in the blanks, and keep them handy. They already follow the formula.

Template A — Weekly report

ROLE: You are my operations analyst.
CONTEXT: Our business is [what you do] and this data is [what the file contains].
TASK: Summarize the attached [sales / ops / KPI] data into a weekly report.
FORMAT: A 3-sentence headline summary, a small table of the top numbers, and 3 bullet-point recommendations.
CONSTRAINTS: Plain English, no jargon. Flag anything that changed more than [X]% vs last week.

Template B — Customer reply

ROLE: You are a customer-service rep for our [type of business].
CONTEXT: A customer wrote: "[paste their message]". Our policy on this is [your policy].
TASK: Draft a reply that resolves their issue.
FORMAT: A short, friendly email — greeting, answer, next step, sign-off.
CONSTRAINTS: Apologetic but not over-promising. Do not offer refunds beyond our policy. Under 150 words.

Template C — Data cleanup

ROLE: You are my data assistant.
CONTEXT: The attached spreadsheet is our [customer list / inventory / expense export].
TASK: Clean it up — remove duplicates, fix inconsistent formatting, and standardize the columns.
FORMAT: Return a tidy CSV plus a short note listing what you changed.
CONSTRAINTS: Do not delete any rows without flagging them first. Ask me before merging records that look similar.

Template D — Social post from one idea

ROLE: You are our social media manager.
CONTEXT: We're a [type of business] and our audience is [who follows us]. This week's news is [what happened].
TASK: Write posts announcing it for [the platforms you use].
FORMAT: One version per platform, each with a hook, 2-3 short lines, and 3 relevant hashtags.
CONSTRAINTS: Match our voice: [friendly / expert / playful]. No emojis if we don't use them. Keep each under the platform's limit.

Template E — Turn a decision into options

ROLE: You are my strategy sounding board.
CONTEXT: We're deciding whether to [the decision]. Here's the situation: [budget, constraints, goal].
TASK: Lay out my realistic options.
FORMAT: A short table with each option, its main pro, its main con, and a rough cost or effort level.
CONSTRAINTS: Be honest about trade-offs. Flag anything I'd need to verify before committing. Don't pick for me — help me think.

A generic prompt gets a generic answer. When Claude knows your brand voice, your prices, and how you do things, the output goes from "fine" to "ready to send." The good news: you can load that context once and it applies to every chat afterward.

Three Ways to Give Context

Claude Projects are a persistent workspace. Upload your brand guide, price list, standard procedures, and past reports once, and every chat inside that Project already knows your business — no re-explaining.

Think of a Project as a labeled folder with a memory. You might keep one Project called "Marketing" with your brand guide and voice examples, another called "Finance" with your reporting templates and chart of accounts, and a third called "Customer Support" with your FAQ and refund policy. Each one stays focused, so the answers you get inside it are focused too. When your prices change or you write a new procedure, you update the file once and every future chat in that Project is instantly current — a huge step up from pasting the same background into a fresh chat over and over.

Custom / style instructions let you set standing rules: your tone of voice, formatting preferences, spelling conventions, and things to always avoid. Claude follows them automatically. This is the place for the small, universal preferences that apply to everything — "always write in British English," "use our product's exact name, never a nickname," "default to short paragraphs" — so you never have to repeat them.

Reference files are templates you want Claude to copy — your report layout, your email signature, your proposal structure. Give Claude the mold and it fills it in consistently. The trick is to hand over a real, finished example rather than describing it: an actual past proposal teaches Claude your structure, headings, and phrasing far better than a paragraph of instructions ever could.

Two more, briefly: Skills package repeatable know-how (like "how we write a monthly close") so Claude does the task your way every time, and MCP is a standard way to safely connect Claude to your existing tools and data. You don't need the technical details — just know these exist when you're ready to go further. In plain terms: a Skill is a saved recipe for a task you do repeatedly, and MCP is the safe "adapter" that lets Claude reach into a tool you already use (say, your calendar or your files) instead of you copying and pasting between them.

Things Worth Uploading

A few concrete examples make this real. A landscaping company might upload its rate card, a list of the services it does and doesn't offer, and two past quotes it was proud of — now every new estimate comes out priced correctly and worded like the owner would word it. A boutique might upload its size chart, return policy, and a page of "how we talk to customers" — now support replies and product descriptions sound on-brand without anyone editing them. The rule of thumb: if you find yourself explaining the same thing to Claude twice, that thing belongs in a Project.

You don't have to build the perfect knowledge base on day one. Start with the two or three documents you reach for most, and add more as you notice gaps. Keep the files clean and current — an outdated price list is worse than none, because Claude will trust it. Treat your Project like a tidy filing cabinet, not an attic.

Don't automate the hardest, most important thing first. Automate the boring, repetitive thing first. Score each candidate task on three questions:

• Time saved: how long does it take you by hand each time?
• Frequency: how often do you do it — daily, weekly, monthly?
• Ease: how rule-based and repetitive is it? (Clear rules = easy; nuanced judgment = hard.)

Tasks that are high-frequency + repetitive + rule-based are your first wins. Anything that needs human judgment, relationships, or negotiation should keep a human in charge.

A Worked Example

Let's score a real task the way you would on a napkin. Say you run a small dental clinic and every Friday you spend about 45 minutes writing appointment-reminder texts for the following week.

• Time saved: 45 minutes a week, every week. That's high — call it a 3 out of 3.
• Frequency: weekly, without fail. Also high — a 3 out of 3.
• Ease: the task is almost entirely rule-based (name, date, time, a friendly line), so it's very repeatable — another 3 out of 3.

A near-perfect score means this is a "start now" task. Contrast that with "deciding which new service to add to the clinic": maybe an hour of work, but you do it once a year, and it needs real judgment about your market and money. Low frequency, low ease — that one stays firmly with a human. You don't need a spreadsheet or precise numbers here; a quick high/medium/low gut-check on the three questions is enough to sort your whole to-do list into "automate," "try later," and "keep human."

Every business is different, but the patterns repeat. Here are concrete starting points for seven common industries. Pick the closest one and try a single use-case this week. You'll quickly notice that most of these come down to the same three buckets — recurring paperwork, regular reports, and customer messages — dressed up in the language of each trade.

The fastest way to fail is to announce "we're all using AI now" on day one. The fastest way to succeed is to start with a couple of eager people, get a visible win, and let it spread. Here's a simple three-phase plan. The whole approach is built on a simple truth about human nature: people trust what they've seen work for a colleague far more than what they've been told to do in a memo.

How to Pick Your Champion

Your champion isn't necessarily your most technical person — it's your most curious and most trusted one. Look for someone who already tinkers with new tools, is respected by their peers, and enjoys helping others. A well-liked champion who can say "here's the exact prompt I use every Monday" will do more for adoption than any all-hands presentation. Give them a little protected time to experiment, and make it clear that early stumbles are expected and fine.

How to Run the Pilot

Keep the pilot deliberately small so a win is quick and a failure is cheap. Pick one or two tasks that score high on the automation method, agree on what "success" looks like before you start (for example, "reminders sent in 10 minutes instead of 45"), and check in briefly at the end of each week. Resist the urge to automate everything at once; one clear, repeatable win is worth more than five half-finished experiments, because it becomes the story you tell the rest of the team.

How to Measure Hours Saved

You don't need a fancy dashboard. Before you start, jot down roughly how long a task takes by hand. After a few weeks, ask the people doing it how long it takes now, and multiply the difference by how often it happens. "This one change gives us back three hours a week" is the single most persuasive sentence you can bring to a skeptical owner or team. Track two numbers only: hours saved per week, and how many people are actually using it. Those two tell you almost everything about whether it's working.

How to Handle Skeptics

Some resistance is healthy — take it seriously rather than steamrolling it. Most worries fall into three buckets: "it'll replace my job," "it'll make mistakes," and "I don't have time to learn it." Answer the first honestly: the goal is to remove drudgery so people do more of the work only humans can do. Answer the second with guardrails: a human reviews anything that touches money or customers, so mistakes get caught. Answer the third by doing the first task with them, live, so they see it take five minutes instead of an afternoon. Never force it — let the skeptics watch a colleague's win and come around on their own timeline.

What Managers Should Focus On

• Name a champion: one person who owns adoption and answers questions.
• Share prompt templates: a shared doc of "prompts that work here" removes the blank-page problem for everyone.
• Measure ROI: track hours saved per week — the clearest, most convincing number.
• Set simple usage guidelines: a one-page "do this / not that" is enough to start.
• Set clear guardrails: spell out what's OK to share with AI and what always needs a human review before it goes out.
• Celebrate wins out loud: when someone saves an afternoon, tell the team — social proof is your best adoption engine.

You don't need the biggest plan to start. Most owners and managers begin small and move up only when more people are relying on it. Pricing below is approximate and changes over time — always check anthropic.com for current pricing before deciding.

The honest advice: don't overthink the choice up front. The cost of starting one tier too low is tiny — you upgrade in a couple of clicks the moment you hit a limit. The signals that it's time to move up are simple: you're bumping into usage limits during a normal workday, or more than a couple of people want their own access and you'd rather manage it centrally. Until then, a smaller plan lets you prove the value before you spend more.

About Cowork: Cowork is included on all paid plans, and Claude for Small Business runs inside Cowork at no extra charge beyond your Claude license and whatever partner tools you already pay for.

You don't need a security team to use AI responsibly — you need a handful of habits. Run through this checklist and set these as your team's ground rules. Most AI mishaps at small businesses aren't dramatic breaches — they're someone pasting a spreadsheet full of customer details into a personal free account, or acting on a confident-sounding answer without checking it. The habits below prevent almost all of that with very little effort.

Do I need to know how to code?

No. You work in plain English. The only time code is involved is if you decide to build a hosted app — and even then, Claude writes the code, not you. Your job is to describe what you want.

Will this replace my team's jobs?

It removes drudgery — the repetitive paperwork and busywork — so your people spend more time on higher-value work: judgment, relationships, and decisions. Keep humans on the things that need a human. Think of it as giving each person a tireless assistant. In practice, most owners find their team gets more done and enjoys the work more, because the tedious parts shrink. The framing that lands best with staff is honest and simple: this handles the busywork you never liked anyway, so you can focus on the parts of the job that actually need you.

Is my business data safe?

Yes, with the right plan. Team and Enterprise plans don't train on your data by default. The most important habit is simple: don't paste secrets (passwords, card numbers, customer PII) into a general chat. See the Privacy section for the full checklist.

How much does it cost?

You can start for free. For real day-to-day use, Pro is roughly $20/month. Bigger plans exist for power users and teams. Pricing is approximate — check anthropic.com for current pricing.

How accurate is it?

Very useful, but not infallible. It can be confidently wrong. Treat it like a sharp assistant whose work you review — always verify anything that affects money, legal matters, or customers, and keep a human in the loop. The good news is that mistakes are usually easy to catch when you ask it to show its reasoning or cite where a number came from. A quick "how did you get that?" often surfaces a wrong assumption before it costs you anything.

How do I start today?

Pick one boring task you do every week. Write a prompt using the formula (Role + Context + Task + Format + Constraints). Try it in Cowork. That single win is the whole beginning. Don't wait until you "understand AI" — you learn far more from one real task than from any amount of reading, and the whole thing forgives mistakes because nothing is final until you say so.

What if I don't like the answer I get?

Just say so, in plain English. "Too formal," "make it half as long," "you missed the refund policy," "try three different angles" — each follow-up sharpens the result. Getting a so-so first draft isn't failure; it's the normal middle of the process. The people who get the best results are simply the ones who keep the conversation going for two or three more turns.

Do I need a different tool for each job?

No. The same Claude handles writing, spreadsheets, summarizing, planning, and drafting replies — you change what it does by changing what you ask, not by buying separate apps. Starting with one tool for most tasks keeps things simple and keeps your context in one place.

How long until we see results?

Usually the same day. A first useful task — a cleaned-up spreadsheet, a drafted email, a summarized report — often takes minutes. Meaningful, measurable time savings across a team typically show up within the first few weeks once a couple of workflows become habit. This isn't a big IT project with a long runway; it's a tool people can benefit from on day one.

What's the difference between Cowork and Claude Code?

Cowork is the no-terminal, plain-English teammate that works across your files and apps to deliver finished work — the right starting point for almost every owner and manager. Claude Code is for actually building and shipping software. If you're not planning to build an app, you'll live happily in Cowork.

Official Anthropic Resources