Anthropic shipped the menu: 15 prebuilt workflows for finance, ops, sales, marketing, service, and HR. This is the cookbook for it. Seven role playbooks, each with a named weekly ritual, a project template you can fill in an hour, twelve copy-paste prompts, one integration walked through end-to-end, and the L4 sketches you should aim at — including the ones you shouldn't build yet. Written for the owner running the place, not the consultant selling the transformation.
Every role below sits on the same ladder. The biggest mistake we see: a business reads about autonomous agents (Level 4) and tries to build one before anyone in the company has spent a week on Level 1. Walk up the rungs.
Open Claude. Paste a contract, a CSV, a customer email. Ask questions, draft outputs. No setup, no integrations.
Use Claude Projects. Load your brand voice, SOPs, chart of accounts, customer list once — reuse everywhere. Consistency across the team.
Connect QuickBooks, HubSpot, Canva, Google Workspace, Microsoft 365. Claude operates inside the tools you already use.
Scheduled workflows. Autonomous tier-1 handling. Custom MCP servers for your specific data. Always human-in-the-loop on what matters.
In a small business, "roles" are sometimes one person wearing four hats. Read the one that's most you, then skim the others — the patterns repeat. Each playbook has the four levels, one anti-pattern to avoid, and what success looks like in 90 days.
You're the bottleneck on judgment. AI doesn't replace your judgment — it gives you a faster way to surface what your judgment should be applied to. The full Owner playbook walks through all of this with copy-paste prompts, a project template, and a wired integration.
Using Claude as a yes-man for decisions you've already made. The value is forcing yourself to articulate the question well enough for someone else to engage with it.
You have a weekly "Owner's Hour" ritual, a single "My Business" project everyone on the leadership team can use, and a Monday cash briefing you didn't write.
Operations is the highest-leverage role for AI in a small business. Every recurring process is a candidate. Start where the same thing happens more than twice a month. The full Operations playbook walks through the SOP Hour ritual, the project template, twelve prompts, and a wired client-onboarding workflow.
Trying to systematize a process that's still half-figured-out. If you can't write the SOP yourself, Claude can't either — it'll just write a confident-looking wrong one.
Your three most-repeated processes (onboarding, monthly reporting, vendor renewals) each have a documented playbook, a project, and at least one integration doing the boring middle.
The role with the most AI hype and the biggest gap between hype and reality. The win is consistent voice and faster iteration — not "AI writes my marketing." The full Marketing playbook walks through the Source Hour ritual, the Brand Voice project, twelve prompts (including one that flags whether your draft sounds like AI), and a quarterly performance-to-calendar workflow.
Posting AI-drafted content without editing. Your audience can tell. The lift is in volume × consistency, not "look ma, no hands."
You're shipping 3× the content with the same head count, your voice is more consistent than it was a year ago, and every post starts from a project, not a blank page.
The pre-meeting brief is the gateway drug. Once a salesperson walks into a call with a 90-second AI-generated brief and closes it, they're hooked. The full Sales playbook walks through the Pipeline Hour, the won/lost-deal log project, twelve prompts, the inbound triage workflow, and a full intent-signal outbound pipeline.
Letting Claude write outreach with zero personalization signal. "Hi {{first_name}}, I noticed your company {{company}}…" is what gives AI a bad name. Always pass real context in.
Every booked call has a pre-meeting brief. Every lost deal gets a 5-line post-mortem. Your CRM hygiene is the best it's been in 3 years — because Claude is doing the typing.
The role with the clearest ROI and the highest stakes. A bad AI answer to a real customer is a real problem. Move slow on autonomous; move fast on draft-assist. The full Customer Service playbook walks through the Service Hour, the project template, twelve prompts, the help-desk draft-assist workflow, and the careful path to a tier-1 agent.
Turning on a tier-1 agent without a weekly transcript review. Drift is inevitable. The review is the entire control loop — skip it and you'll find out the hard way.
Average first-response time is cut in half. Your team is spending the saved time on the 20% of tickets that actually need a human, not the 80% that don't.
QuickBooks integration is the headline. The real value is something quieter: a chart of accounts that's coded consistently across the whole team, every month, forever. The full Finance playbook walks through the Variance Hour, the project template, twelve prompts, a month-end close assistant, and the careful path to a cash-optimization agent.
Trusting AI-categorized transactions without spot-checks for the first 60 days. Your COA rules have edge cases you don't know about until Claude finds them — for better and worse.
Month-end close takes 40% less time. Your cash forecast is updated weekly, not "when we get around to it." You catch one anomaly per month that you'd have missed.
A 12-person company doesn't have an HR department, but it has 100 hours a year of HR work. AI doesn't replace that work — it makes the owner stop avoiding it. The full HR & People playbook walks through the People Hour, the project template, twelve prompts, an offer-to-onboarding workflow, and the explicit lines AI must never cross with people.
Using AI in employee-facing communication without disclosure. Trust is the only HR asset that matters. "Claude helped me draft this — here are my actual thoughts" is fine. Stealth AI is not.
Hiring loops are tighter, onboarding is the same every time, and your handbook is actually used — because you can ask it questions in plain English.
Playbooks are scaffolding. Once you've climbed past them, these are the principles that separate the businesses getting compounding leverage from the ones getting a 1.2× typing bump.
Block 60 minutes every Friday. Use Claude on the single most important question of the week — not email drafts. Pricing review, churn analysis, a hard conversation you've been postponing, vendor renegotiation prep.
This is the practice that converts AI-curious owners into AI-leveraged owners. The output isn't a deliverable. It's that you actually thought hard about the thing you've been avoiding.
Legal advice you'll act on without a lawyer. Medical decisions. Tax positions with real exposure. Anything where a wrong answer costs more than verifying it would.
Claude is great at drafting the question, terrible at being the final answer for high-stakes regulated work. Use it to prepare for the conversation with the professional, not to skip the professional.
Most small businesses don't need a multi-agent system or a custom MCP server. They need somebody to set up four integrations, write seven good prompts, and create three Claude Projects.
If a consultant's first move is "let's build you an agent," they're optimizing for their invoice, not your business. The right first move is almost always Level 2.
Anything that leaves your business — to a customer, the IRS, a contract counterparty, a regulator — gets verified. Anything internal, anything in draft, anything you'd edit anyway — trust and move.
Without this rule, owners either trust nothing (and get zero leverage) or trust everything (and have an incident). The rule is liberating once you internalize it.
Anthropic built Claude for Small Business around "Claude proposes, owner approves." That's not training-wheels for beginners — it's the right architecture for almost every workflow in a 50-person company.
The point of automation isn't to remove yourself from the loop. It's to remove yourself from the boring parts of the loop, so your judgment shows up where it actually matters.
Don't roll out Claude to sales, marketing, finance, and customer service in the same month. Pick one role. Get it to Level 2. Measure something. Then add the next.
Patience wins this. A half-working workflow rolled out across four roles produces four sets of frustrated users; a fully-working one in a single role produces a believer who pulls the rest of the team along.
If you do these four weeks, you'll be ahead of 90% of small businesses experimenting with AI right now. The trick isn't the prompts. It's not skipping a week.
A consultant publishing the list of things you should do yourself is counterintuitive — which is exactly why we do it. If you're on the left column, you don't need us. If you're on the right, the 30-minute call is free.
Anthropic gave you Claude. We give you the cookbook for your specific kitchen. Start with the 30-day plan. If you stall, the call is free and direct — no pitch, no deck.