The Agency Playbook

Most AI agencies cap under $10K/mo. It is rarely a sales problem. It is almost always a structure problem.

This is the structure that broke me past it, and what every agency I coach inside AI Agency OS is running. Steal what you can use.

1. One offer, one outcome

Pick a single outcome and become the obvious choice for it. The agencies that compound are the ones who refuse to sell five things.

Three filters for your offer

• Specific outcome the buyer already wants. Booked appointments. Recovered leads. Hours back. Fewer no-shows. Faster quoting. Lower CAC. If you have to explain the outcome, it is too abstract.
• One vertical or one role. "Roofers." "HVAC." "Wedding photographers." "Ops managers at SaaS." Generic equals invisible. Niching down doubles your close rate and halves your delivery cost because every client looks the same.
• Provable in under 14 days. If your result takes 6 months to feel, your offer is too vague. Pick something the client can measure within 2 weeks of going live.

The one-line offer formula

Examples that work:

• "We help roofing companies turn every missed call into a booked appointment within 14 days, using a 24/7 AI receptionist."
• "We help med-spa owners recover the leads that went cold inside their CRM, in under 30 days."
• "We help solar installers quote 3x faster, with an AI that drafts the proposal in under 90 seconds."

B2B or B2C, pick one

Selling AI to businesses is a different sport from selling to consumers. The hook, the price, the timing, the proof all change. Trying to do both halves your close rate. Pick whichever side your first paying customer is on, then double down for the next 12 months.

2. The AI builds we sell

You do not need ten products. You need three, packaged tightly. Here are the three we sell most often, who buys them, the rough delivery, and a sample of the prompt or workflow.

Build 1: The Missed-Call Recovery Agent

Who buys it: any business that gets phone leads. Roofing, HVAC, plumbing, locksmiths, med-spas, dentists, solar.

What it does: answers every missed call within seconds, qualifies the lead in a short script, books an appointment into the calendar, and notifies the team in Slack or Telegram with a summary.

Typical numbers: recovers 20-40% of previously-missed calls, books 5-15 extra appointments per month for a single-location business, pays for itself in week one.

Install footprint: a voice agent (Retell or VAPI), a calendar connection, a CRM webhook, one fallback notification, and an analytics dashboard. Usable on day one of go-live.

Sample voice-agent prompt opener:

Build 2: The Lead Follow-Up Engine

Who buys it: any business with a steady flow of new leads from forms, ads or platforms. Real estate, coaches, agencies, solar, fitness.

What it does: every new lead receives a personalised reply in under 60 seconds. AI references something specific they wrote. Retries on day 2, day 4 and day 7 with different angles. Flags a "hot lead" the moment intent jumps.

Typical numbers: 2x to 3x reply rate vs the client's previous follow-up. Roughly 20-30% more booked calls inside the first month, without any new ad spend.

Install footprint: a webhook from the lead source, an LLM-drafted reply step, a CRM stage update, a "stop sequence" rule when the lead replies, and a Slack notification for hot leads.

Sample reply prompt:

Build 3: The Founder Brief

Who buys it: founders and small business owners drowning in tabs.

What it does: a daily AI summary pulling calls, calendar, leads and revenue into one short brief sent every morning. Includes today's calls, top open leads, yesterday's revenue, and one suggested action.

Typical numbers: replaces about $4K/mo of operations payroll. Saves the founder 1 to 2 hours a day, every day.

Install footprint: 3 to 5 data sources into a single Claude run, output formatted to a Telegram message or Notion page, scheduled via cron.

3. Sales and outreach

Most agency owners overthink offers and underthink sales conversations. Here is the shortest path to your first 5 paying clients.

The cold DM that works

Short. Specific. No pitch.

Send 20 of these a day for a week. You will get 2 to 5 real conversations. Those become discovery calls.

The 15-minute fit call

Cut the discovery call in half. It is killing your close rate.

1. 1 minute - "Walk me through what you do." Stop. Listen.
2. 4 minutes - "What's the current bottleneck stopping you doing more of that?" Listen for time, leads, or quality.
3. 2 minutes - You name the exact outcome you can deliver and how long it takes.
4. 3 minutes - You quote price, scope and timeline, on the call.
5. 5 minutes - Q&A and next step. The next step is always either "send a deposit link" or "no fit, here's someone better suited."

If your discovery call is longer than 15 minutes you are educating. Education is content. Sales calls are for fit and a yes or no.

The 3 objections you'll hear and what to say

• "Send me a proposal." -> "Happy to. I can also just tell you the price on this call so you don't wait. Want that?"
• "Can you guarantee results?" -> "I guarantee the install in 14 days. The result is yours to claim once it's live."
• "It's expensive." -> "Compared to what? A salaried hire that costs you 5x and works one shift?"

The 4 numbers to track from day one

Before $10K/mo, vanity metrics will lie to you. Track only what feeds revenue. Four numbers, on a sticky note above your laptop, updated every Sunday.

• DMs sent per day. Should be 20+ until you cross $10K.
• Conversations started per week. Real back-and-forth, not just sends.
• Fit calls booked per week. From conversation to actual calendar.
• Close rate on those fit calls. If it's under 30%, fix the offer before sending more DMs.

The whole business lives in those 4 numbers until your first $10K. Build the dashboard later. Use a sticky note now.

Revive the leads that ghosted you

The lead you didn't close is your cheapest pipeline. Every 30 days, message every prospect from the last 90 days who went cold, with a fresh angle:

Roughly 1 in 5 of those replies. Half of those book a call. Free pipeline you already paid for once.

4. Pricing and the 7-day scope

The right shape for a young AI agency is a one-time setup fee plus a monthly retainer for the system. Stop hourly billing the moment you can.

The starting frame

• Setup: $2K to $5K paid up front, fully refundable if not live in 14 days.
• Retainer: $1.5K to $3K/mo for monitoring, tweaks, and one small improvement per month.
• Scope cap: 7-day install. Anything beyond goes into a separate small project quote, not a "free addition."

Quote in outcomes, not in deliverables. "We turn every missed call into a booked appointment, live in 7 days, $3K setup, $2K/mo." Same product. Completely different close rate.

When to raise prices

• After 3 paying clients on the same offer, raise by 30%.
• After 6 clients, raise again to a price you genuinely feel slightly uncomfortable saying out loud. That is your real number.
• If your close rate is above 50%, you are underpriced. Raise.

The refund policy that closes deals

One line. Put it in your proposal and on your DM closer:

This kills the "what if it doesn't work" objection before it forms. You will rarely be asked to refund because the offer is built to ship in 7 days.

5. The 7-day delivery system

If every project is unique, you are running a custom shop, not an agency. Templatise the install.

Day-by-day checklist

1. Day 1 - Kickoff. 30-min call. Collect logins. Brand voice + service list captured into a doc. Welcome email auto-sent. Client gets a Loom recap that night.
2. Day 2 - Build. Clone your template repo. Wire up the client's tools (calendar, CRM, lead source). Drop in their voice and brand.
3. Day 3 - Build continued + first internal test. Run a full sandbox flow. Catch obvious gaps.
4. Day 4 - End-to-end test with real-looking data. Send the client a 2-minute Loom showing the system handle 3 realistic scenarios.
5. Day 5 - Soft launch. Route 20% of real traffic through the system. Monitor every interaction in real time.
6. Day 6 - Full launch. 100% of traffic. Monitoring on. First "win moment" notification ready for the client.
7. Day 7 - Wrap-up call. Walk them through the dashboard. Set the cadence for the retainer (weekly check-in, monthly improvement).

What goes in your template repo

• Pre-built n8n / Make workflows for the build.
• Voice agent prompt templates with variables for client name, services and tone.
• A Notion handoff doc template for the client.
• A loom-recording script so every kickoff Loom sounds the same.
• The monitoring dashboard, ready to point at any client's data.

Build each step once. Every new client follows the same 7 days. After 5 clients you stop thinking and start shipping.

6. The stack that one operator can run

• Build / orchestration: Claude Code, n8n, sometimes Make. Claude Code for the custom logic, n8n for the connectors.
• Voice: Retell or VAPI. Retell is faster to get live, VAPI is more flexible.
• CRM glue: GoHighLevel or HubSpot, whichever the client already has. Never make them switch.
• Calendar: Google Calendar via MCP. Cal.com if the client is more technical.
• Comms: Slack and Telegram for ops alerts. Gmail / business email for client-facing.
• Dashboards: a tiny Vercel app per client, fed by their data. 1 page, 4 numbers, no clutter.
• Payments: Stripe links and subscriptions. No manual invoicing.
• Project / SOPs: Notion. One template repo, duplicate per client.

Keep this stack short. Every extra tool is one more place the system can break and one more login your future hire has to learn. Add tools only when a real client problem forces it.

7. Hire and system in the right order

You do not hire to do more. You hire to take work off yourself in a specific order, while you build the system that makes the next hire unnecessary.

The hire order

1. An operator / project manager ($1.5K to $3K/mo full-time overseas, or 20 hrs/wk local). Owns the 7-day install. Frees you from delivery first.
2. A senior automation builder ($3K to $5K/mo). Implements the templated builds at quality. Frees you from technical execution.
3. A part-time growth hire (commission + base, or $1.5K/mo). Owns outbound DMs, content prep and lead capture while you stay client-facing.

How to actually find them

• Operators / PMs: Onlinejobs.ph (Filipino market). The good ones cost more, take them.
• Automation builders: Twitter and Skool communities for n8n / Make / Claude Code. The best ones already run their own micro-agencies.
• Growth hires: your audience. Whoever has been DMing you about "how do I get into this" for 6 months is your first hire.

The 30 / 60 / 90 plan for every hire

• First 30 days: shadow you on every call. Own the SOP for one task by day 30.
• 60 days: own the entire 7-day install for one client end-to-end. You only review on Day 4.
• 90 days: own 3 clients in parallel. You stop touching delivery completely.

Every hire slots into a documented system. The goal is not headcount. It is "the agency runs without me on it every day." When that lands, the agency runs in 2 to 3 hours of your time a day. The rest of your week goes to content, coaching, and the next thing you want to build.

8. Common mistakes that kill young AI agencies

• Selling 5 things. Pick 1. Refuse the rest for 3 months.
• Custom-building every project from scratch. Reuse 80% of the last one.
• Quoting hourly. Hourly caps your business at your hours.
• Charging too little because "I'm just starting." Clients trust higher prices. Charge $3K from day one.
• Promising on every call. Quote on every call. Two different things.
• Hiring developers before hiring an operator. Operators 10x your time. Devs 1x it.
• Spending months on a website. Spend hours on a one-page offer doc.
• Posting AI buzzwords. Post outcomes. Sound like a business owner, not a coder.
• Treating retainers as charity. Build something real into the retainer (monitoring, monthly improvement, dashboard updates).
• Building "the agency platform" in year one. Use existing tools. Custom platforms come later, not now.

9. The 30-day launch plan (do this, in order)

If you did nothing else but follow these 30 days exactly, you would have your first paying client. No theory. Just the actions, in sequence.

Week 1 - Offer and list

• Day 1: Write your one-line offer (section 1). Pick ONE niche and ONE build. Do not move on until both are written down.
• Day 2: Build a list of 100 businesses in that niche. Name, owner, a link to their page, one specific detail you can reference. A spreadsheet is enough.
• Day 3: Set up your outreach. Clean personal profile, banner with your one-line offer, a 90-second Loom showing the build working (use a demo account, not a client).
• Day 4-7: Send 20 DMs a day from the script in section 11. Reply fast. Book any "yes I'm interested" straight onto a 15-min fit call.

Week 2 - Conversations to calls

• Keep sending 20 DMs a day. Never stop the top of the funnel.
• Run every fit call with the script in section 12. Quote on the call.
• Send a Stripe deposit link within 10 minutes of any "yes, let's do it."

Week 3 - First install

• Run your first client through the 7-day delivery (section 5). Build slowly, document every step as you go. This first build becomes your template.
• Keep sending 10 DMs a day even while delivering. Pipeline never sleeps.

Week 4 - Win and repeat

• Capture the first result the moment it happens. Screenshot it. That is your first case study.
• Post the win. Use it in your next 100 DMs. Raise your price for client number two.

10. The numbers behind a 6-figure solo agency

Six figures sounds huge until you break it down. It is smaller than you think.

• $100K/year is $8,333/mo.
• At $3K setup + $2K/mo, four active retainers plus one new setup a month clears it.
• Four clients is not a team. It is a laptop, a template, and 2-3 hours a day.

The mistake is chasing 40 clients at $200/mo. That is a job with 40 bosses. Chase 5-8 clients at $2K+/mo with a system underneath, and you have a business you can actually run from anywhere.

11. Swipe file: the full outreach sequence

Copy these word for word, swap the brackets, send them. This is the exact sequence that turns a cold name into a booked call.

Message 1 - the opener (day 0)

Message 2 - the soft proof (if they reply with interest)

Message 3 - the call ask (after they watch)

The no-reply follow-ups

• Day 2: "Hey [name], floating this back up in case it got buried. Worth a quick look?"
• Day 4: "Last nudge from me. Just landed a similar build for [comparable business], thought of you. Want the Loom?"
• Day 7: "All good if the timing's off. I'll leave it here. If [pain] ever becomes a priority, you know where I am."

The 30-day re-engagement (the cheapest pipeline you have)

12. The discovery call, word for word

15 minutes. You are diagnosing fit, not educating. Follow the beats.

1. Open (30s): "Thanks for hopping on. To keep this useful and quick, I'll ask a few questions, then if it's a fit I'll tell you exactly what I'd build, the price and the timeline. Sound good?"
2. Diagnose (5 min): "Walk me through how you currently handle [the thing your build fixes]." Then: "What does that cost you in [time / leads / money] right now?" Let them quantify the pain. Write the number down. You will quote against it.
3. Mirror the cost (1 min): "So if I'm hearing it right, [pain] is costing you roughly [their number] a month. Is that fair?" Get the yes.
4. Prescribe (2 min): "Here's what I'd build. [One build, one outcome, in plain English.] Live in 7 days, and you'd start seeing [measurable result] within two weeks."
5. Quote on the call (1 min): "It's [X] to set up and [Y] a month to run and maintain it. If it's not live in 14 days, you get every dollar back."
6. Silence. Stop talking. The first person to speak after the price loses. Let them respond.
7. Close (remaining): "Want me to send the deposit link and we kick off Monday?" If yes, send it on the call. If "let me think," ask: "Totally fair, what's the one thing you'd need to be sure of to say yes?" Handle that, then re-ask.

13. 50 niches buying AI right now

Stuck on a niche? Pick from businesses that already lose money to slow response and manual admin. These all convert because the pain is obvious and measurable.

Home services: roofing, HVAC, plumbing, electricians, landscaping, pest control, garage doors, pool service, solar, locksmiths, junk removal, cleaning companies.

Health and beauty: med-spas, dental practices, chiropractors, physio clinics, aesthetics, hair salons, barbershops, gyms, personal trainers, vets.

Property: real estate agents, property managers, Airbnb / short-term rental operators, mortgage brokers, letting agents.

Professional services: accountants, law firms, recruiters, insurance brokers, financial advisers, marketing agencies, consultants, coaches.

Local and retail: auto repair, car dealerships, restaurants (booking + reactivation), event venues, photographers, wedding vendors, tutoring centres, driving schools.

14. Build it with Claude Code (the actual setup)

This is how one person delivers agency-quality builds without a dev team. Claude Code is the brain. Everything else is a connector.

Your client template repo

One folder per client, cloned from a master template. The master has a CLAUDE.md kernel that teaches Claude Code how you work:

The 5 skills every agency build reuses

• lead-engine - drafts and sends the follow-up, updates CRM stage, flags hot leads.
• call-handler - qualifies the inbound, books to calendar, summarises to Telegram.
• daily-brief - pulls calls, leads and revenue into one morning message.
• reactivation - runs the 30-day cold-lead message on a cron.
• reporting - writes the weekly client update from the dashboard data.

Build each skill once. Drop it into every client folder. The client's CLAUDE.md changes; the skills barely do. That is the whole leverage.

The one command that delivers

15. Your proposal and onboarding pack

You need exactly three documents. Build them once, reuse forever.

The one-page proposal

That is the whole proposal. If it needs to be longer, your offer is not clear enough.

The kickoff intake form (send the moment they pay)

• Business name, owner name, best contact.
• Logins / access: calendar, CRM, lead source, phone number.
• Service list and pricing.
• Brand voice in 3 words + 2 example messages they would send a customer.
• The one scenario that must never go wrong.

The kickoff Loom script

16. Your first $10K/mo

The job for this guide ends at $10K/mo. Past that, the bottlenecks change and the answers stop being one-size-fits-all.

To get to $10K from zero, the order is simple:

1. Pick one offer and one niche. Refuse the rest for 90 days.
2. Send 20 DMs a day for 7 days. Book the conversations.
3. Quote on the call. $3K setup, $2K/mo, 7-day install.
4. Land 3 to 5 clients on the same offer. Templatise as you go.
5. Document every objection, every win, every mistake.

By the time you cross $10K, the offer should sell itself. That is when you and I jump on a call, and we figure out the level after this one together.