AI Automation Agency

The shortcut: Most new automation agencies pitch "I'll automate your business" and get tire-kickers. The ones charging $3K-$5K per build pick one industry — dental offices, real-estate brokerages, law firms — and sell the same packaged workflow over and over.

Industry: Software & Tech Investment level: Small — $5,000-$15,000 Time to launch: 6-10 weeks (n8n or Make Partner setup + a packaged offer + first 2 paying clients gate the launch) Best for: Someone who can already wire up a Zapier or Make scenario without watching a tutorial, write a clear scoping doc, and sit with a non-technical client through "what does your week actually look like." You're a fit if you can read a webhook payload, debug an OpenAI prompt that's returning the wrong JSON, and tell a client no when they ask you to automate something that shouldn't be automated. What you'll likely make: $1,500-$3,000 month 3, $4,000-$7,000 month 6, $8,000-$15,000 month 12. Math is in Section 4.

Market Opportunity

Every small business owner has the same conversation in their head once a quarter: "There has to be a way to stop doing this manually." The "this" is invoice extraction, lead routing from form submissions, onboarding email sequences, monthly report generation, transcript-to-CRM-note translation. They know a computer could do it. They don't have a tech person on staff. They've tried Zapier on their own and gave up around step four when the CSV broke.

That's the door. The 2022 version was "Zapier consultant" — rule-based connections between SaaS apps. The 2025-2026 version is different because LLMs now handle unstructured inputs rule-based automation never could: a PDF invoice with a weird layout, an inbound email in someone's casual voice, a form field where the user typed three things into one box. n8n added native AI nodes in version 2.0 (n8n pricing + AI features). Make added GPT-4 modules. The stack got dramatically more useful in 18 months.

The wedge that closes deals: pick one vertical and own it. "Lead nurture automation for real-estate brokerages" beats "I automate business processes" because the buyer recognizes their own problem in the first sentence. A dental-office automation package — new-patient intake form to PMS to confirmation SMS to insurance verification email — sells for $4,500 fixed price. Sell that same package to ten dental offices in your metro and you've built a real business without ever writing a custom proposal. Generalist agencies still exist; the ones charging real money have a niche.

Launch With AI

Pro section. AI doesn't replace the work — it cuts the parts that drained you (writing scoping docs, scripting demo videos, debugging LLM JSON, drafting personalized cold emails). Spend the saved time on what AI can't do: picking the right vertical, sitting through three discovery calls a week, and reading webhook payloads when a client's automation breaks at 11pm.

The meta-trap: you're building AI workflows for a living, but most new automation founders never run their own business on the same workflows. They scope clients on a Google Doc, write SOWs from scratch, and personalize cold emails by hand. The agencies clearing $10K/mo treat their own business as the first case study — every internal task that takes more than 30 minutes a week becomes a Make scenario or a saved ChatGPT prompt.

Important up-front: AI doesn't pick your vertical. That comes from your network and three discovery calls — "who keeps DMing me automation questions?" The tools below are pure leverage after you've chosen.

AI Tools You'll Use

Tool	Price	What it does
ChatGPT Plus	$20/mo	Vertical pain-point research, discovery-call summaries, cold-email personalization at volume
Claude Pro	$20/mo	Long-context contract + SOW drafting, debugging LLM steps that return wrong JSON
Apollo Basic	$59/mo	50 vertical-matched contacts/week (dental owners, real-estate brokers) with verified email
Loom Business + AI	$15/mo	Auto-trim silences, AI chapter titles, transcripts for your packaged-build demos
Cursor Pro	$20/mo	n8n custom-code nodes + webhook handlers without writing TypeScript from scratch

The Workflow

Pick the vertical (ChatGPT, ~30 min). Paste:

"I'm starting a vertical-specific AI automation agency. I have prior network exposure to [list 3 industries you've actually worked in or near]. For each, give me: (1) the 3 most-automated workflows surfacing in their trade forums and subreddits, (2) the SaaS apps they already use daily, (3) typical IT spend per practice/firm, (4) the decision-maker title, (5) one packaged offer ($2,500-$4,500 fixed price) that solves their #1 pain. Skip industries where decisions go through an IT committee."

Pick the vertical where you can name 3 specific decision-makers in your phone contacts this week. If you can't name 3, pick a different vertical.
Build the demo on your own data first (Make + Claude, ~6 hours). Don't pitch what you can't demo. Run the packaged workflow against your own inbox or invoices. When the LLM step returns malformed JSON, paste the payload and the error into Claude:

"This n8n OpenAI node should output strict JSON {vendor, invoice_number, total, due_date} from a PDF invoice. It's returning an extra notes field and sometimes nulling the total. Rewrite the prompt with strict output formatting, a confidence score, and explicit handling for missing fields. Show me the new prompt and the JSON schema check that should follow it."

Loom-record the working demo (4-6 min). That recording becomes your sales asset for the next six months.
Generate the SOW + AI disclosure clause (Claude, ~45 min). Upload your one-page packaged offer. Paste:

"Generate a 4-page Statement of Work for the attached automation package. Include: scope (numbered deliverables), exclusions (list 8 things explicitly out of scope), milestones (kickoff, build, UAT, handoff), payment terms (50% upfront, 50% on UAT signoff), an AI-generated content disclosure clause aligned with EU AI Act Article 50, a GDPR Article 28 DPA reference, and change-order pricing at $200/hour. No legalese padding."

Cuts a 4-hour scoping document into 30 minutes of editing. Have an attorney review the resulting template once before your first $5K+ contract — $300-$500 one-time, then reuse forever.
Build the outbound list and personalize at volume (Apollo + ChatGPT, ~2 hrs/week). Pull 50 contacts in your vertical + metro from Apollo. Paste 10 of their LinkedIn bios into ChatGPT:

"For each LinkedIn bio below, write a 90-word cold email. Open with a specific detail from the bio (not 'I noticed you're a CEO'). One sentence on a problem specific to [vertical]. One sentence demo offer with a Loom link placeholder. Soft CTA: 15-min discovery call. Vary openers — no two emails should sound similar. Output as plain text, one email per bio."

Personalized cold beats spray-and-pray by ~5x reply rate. Cap at 30 sends/week so you can actually follow up on the replies — the follow-up is where the meeting gets booked, not the first send.
Record one demo, recycle for 6 months (Loom AI, ~30 min once). Walk through the working automation in 4-6 minutes. Loom AI auto-trims silences, generates chapter titles, and produces a searchable transcript. Embed the Loom in cold emails and on your packaged-offer landing page. One reusable demo replaces 20 live discovery calls — buyers self-qualify before booking your time.

Time Saved Per Week

Roughly 8-10 hours/week once your packaged offer is locked and your demo is recorded:

SOW + scoping doc per new client: 4 hours → 30 min (Claude template, you edit)
Cold outbound personalization: 3 hours → 45 min (Apollo + ChatGPT bulk-generation)
Live demo calls with unqualified prospects: 90 min × 4/week → 0 (Loom replaces all of them)
LLM prompt debugging when client automations misfire: 2-3 hours → 45 min (paste payload, get fix)

Trade that time for: three real discovery calls per week, one new packaged offer for a second vertical by month 6, and two written case studies for your landing page. Those compound — outbound volume doesn't.

Total AI Stack Cost

Budget tier ($40/mo): ChatGPT Plus + Claude Pro only. Skip Apollo (use LinkedIn Sales Navigator's free trial), Loom Free instead of Business. Right for the first 30-60 days while you're still locking the offer.
Full tier ($134/mo): ChatGPT + Claude + Apollo Basic + Loom Business AI + Cursor Pro. Worth it once you've signed your second client and outbound volume is the bottleneck. The Apollo seat alone replaces ~6 hours/week of manual LinkedIn prospecting.
Compare: A part-time SDR doing the same outbound runs $2,500-$4,000/month. A freelance technical writer billing your scoping docs is $75-$150/hour, ~$3,000/month at one build per week.

Cancel anything you don't open in a 7-day window — per-seat tools at $20-$30 each stack into $300/mo if you keep saying yes to "just one more."

Your First Win

30 minutes from now you'll have a packaged offer + a vertical. Open ChatGPT (free tier works for this one task). Paste:

"I'm starting an AI automation agency. I have past work or network exposure in [list 2-3 industries]. For each, give me: (1) one specific workflow that's painful and repeatable enough to package as a $2,500-$4,500 fixed-price build, (2) the 3 SaaS apps that workflow touches, (3) the typical decision-maker title and where I can find them (LinkedIn, trade conference, local chamber), (4) two competitor agencies already serving that vertical so I know it's a real market with budget."

Pick the vertical where (a) you've actually done the workflow before in your career and (b) you recognize one of the competitor agencies named. You've just cut 6 hours of "what should I sell?" into 30 minutes. Next: build the demo against your own data this weekend — that's your sales asset, not a polished website.

Product / Service Offering

You're selling three things in some combination:

Packaged automation builds — fixed-price, fixed-scope, vertical-specific. Examples: "Real-estate lead-to-CRM pipeline" ($2,500-$4,500), "Dental new-patient onboarding" ($3,500-$5,500), "Law firm intake-to-conflict-check" ($4,500-$7,500). Two to four weeks of work. Defined deliverables. No hourly billing.
Monthly maintenance + expansion retainer — $1,000-$3,000/month per client. You keep their existing automations alive (platform updates break things), monitor the exception queue, and ship one or two new automations per month from their backlog. This is where the real money lives — at four retainer clients you've cleared $5K-$10K MRR on top of new builds.
One-off custom builds — custom work for clients who don't fit your packages. Charge more ($5K-$15K) and only take these when the build will become a future packaged offer.

Pick one vertical for your first 90 days. Build the same workflow three times for three different clients in it. By client number three you'll know which steps break, what exception handling looks like, and how to scope it in a 30-minute call. That's when you raise the price.

Revenue Model

Unit economics for a solo agency on Make + n8n + OpenAI/Anthropic, no employees, working from a laptop:

Service	Price	Variable cost (platform + API)	Your time	Take-home per engagement
Packaged automation build (small, 3-step)	$2,500	$50 (Make ops) + $20 (LLM API for testing) + $73 Stripe (2.9% + $0.30)	20-30 hours	~$2,350
Packaged automation build (mid, AI doc parsing + dashboard)	$4,500	$80 + $40 + $131 Stripe	40-60 hours	~$4,250
Monthly maintenance + expansion retainer	$1,500/mo	$30 platform + $50 LLM API + $44 Stripe	6-10 hrs/mo	~$1,375/mo
Custom one-off build	$8,000	$100 + $80 + $232 Stripe	80-100 hours	~$7,600

Your first $1K month = one small packaged build at $2,500 (delivered in weeks 3-4, paid 50% upfront, 50% on delivery). The deposit alone clears you past $1K. Your first $3K month = one mid-size build ($4,500) plus the second half of an earlier build, OR two retainer clients at $1,500/month.

The retainer math is what makes this a real business. Four retainers at $1,500/month = $6,000 MRR with ~30 hours of monthly work. Add one build per month at $4,500 and you're at $10,500/month gross, ~$9,500 take-home. Switching costs are high — a client with 12 live automations you built and understand isn't going to re-onboard a new contractor to maintain them.

Startup Costs

Automation platforms. Make at $9/month for 10,000 ops on the Core plan, scaling to ~$29/month for the Pro tier with multi-user — Make pricing. n8n Cloud starts at $24/month, or self-host on a $5-$10/month VPS for unlimited workflows — n8n pricing. Zapier Pro at $19.99/month for 750 tasks scales fast and gets expensive (100K tasks runs $800+/month) — Zapier pricing. Recommend Make or n8n to clients with high volume; keep Zapier for clients who already use it.
LLM APIs. OpenAI GPT-4o for email parsing and content gen (~~$2.50 per 1M input tokens). Anthropic Claude 3.7 Sonnet for long-context document work (~~$3 per 1M input tokens). Budget $50-$200/month across testing and small production loads — Anthropic docs | OpenAI docs.
Adjacent tools clients pay for or you re-bill. Airtable ($20/seat/mo) for structured data + client-facing portals — Airtable pricing. Retool ($10/standard user/mo) for internal dashboards — Retool pricing. Typeform or Tally for triggers.
Make Partner status. Free to apply once you've delivered a few projects. Gets you white-label access, reseller pricing on client subscriptions, and inbound leads from Make's partner directory — Make Partner program. The Zapier Experts directory is the same idea on Zapier's side — Zapier Experts.
LLC + EIN + insurance. $35-$500 LLC filing depending on state — LLC University 50-state table. EIN free at IRS EIN Online — never pay a third party. Professional liability (E&O) insurance: $800-$2,000/year for solo software services via Hiscox or Insureon. Don't skip this — automation that mishandles data is the textbook E&O claim.
Contracts. Bonsai or Fiverr Workspace MSA + SOW templates ($20-$30/month). Get an attorney to review your standard MSA once before your first $5K+ contract — $300-$500 one-time cost that pays for itself.
Sales + ops. Stripe (2.9% + $0.30/transaction — Stripe pricing). Calendly free or Acuity at $20/month. A simple Notion-based CRM is fine until you have 10+ clients.

Realistic all-in: $5,000 if you defer the LLC for 60 days, skip Retool, and ride Make's free trial through your first build; $15,000 if you bind a year of E&O upfront, pay for an attorney to review contracts, build a simple landing page on Webflow ($25/mo) with 3-5 case studies, and budget $3,000-$5,000 for the first six months of platform + LLM costs while clients ramp.

Legal & Formation

Business entity. Single-member LLC the moment you sign a paying client — separates your personal assets from a "your automation routed our customer data to the wrong person" lawsuit. Sole prop is fine for the first 30 days while you test offers. Get your EIN free directly at IRS EIN Online — never pay a third party. Once your net profit clears roughly $80K-$100K/year, run the math on an S-corp election via IRS Form 2553. With four retainer clients plus three builds a year you can hit that ceiling in year two — it's worth knowing the threshold up front.

Licenses & sales tax. No state license required for automation work. Sales tax is the wrinkle: most states don't tax custom development billed as professional services, but if you're re-selling the client's Make or Zapier subscription as a bundled monthly fee, that bundled fee may be taxable in roughly 25 states that tax SaaS. Cleanest setup: have the client pay the platform directly for their own subscription on their own credit card, and you bill only for your build + maintenance time. Once you cross $100K in sales or 200 transactions in any single state you've got economic nexus there post-Wayfair — use Stripe Tax or Avalara to track it once you've got more than five clients.

Industry-specific risk. The single most consequential trap is AI-generated content disclosure to your clients' end users. If your automation generates customer-facing emails, chat replies, support responses, or reports using GPT-4o or Claude, the people receiving that content may have a right to know it's AI-generated. The EU AI Act Article 50 transparency obligation applies to certain AI-generated content for EU recipients starting August 2, 2026. Some US states are moving in the same direction. On top of that, any automation that touches EU personal data (a real-estate lead form that captures an EU contact, an invoice with an EU vendor's name and email) makes you a data processor under GDPR Article 28 — you need a Data Processing Agreement with that client. Most solo automation founders skip this entirely and it's a real liability gap. Bake an AI disclosure clause into your standard SOW: "Client acknowledges that automations include LLM-generated content and is responsible for end-user disclosures as required by applicable law." Ship a one-page DPA template for any client whose data flows include EU contacts. It takes 20 minutes per client and shuts down the largest legal risk in this business.

Marketing & First Customers

Your first three clients come from people who already trust you, not from cold outbound. The order that actually works:

Pick a vertical and tell your network. One LinkedIn post: "Building automation packages for {dental offices / real-estate teams / law firms}. First three clients get 50% off and lifetime maintenance pricing in exchange for a case study." You'll get 1-3 inbound DMs from your existing network within 72 hours.
Local industry meetups + chambers. Real-estate broker breakfasts, dental practice management conferences, your county bar association tech committee. Show up with a one-page automation menu specific to that industry. Three meetings = one signed client at this stage.
Make Partner directory + Zapier Experts. Free listings worth doing once you have 2-3 case studies. Inbound from these is slower but qualified — the buyer has already decided they need this.
Loom-based cold outreach. Once you have a packaged offer, record a 90-second Loom showing your packaged automation running, send it to 20 SMBs in your vertical per week via personalized email. Conversion runs 2-5% to discovery call, ~30% of those to paid build.
Indie Hackers + dev community content. A monthly write-up on Indie Hackers about a real client automation you built (anonymized, with permission) builds long-term inbound for the kind of clients who self-identify as ready to buy.

Skip Google Ads and Facebook Ads for the first six months. The buyer journey here is consultative — paid acquisition pipes underqualified leads into a sales process that doesn't yet exist.

First 90 Days

Week 1. File LLC. Get EIN. Bind E&O insurance ($800-$2,000/year). Pick one vertical (dental, real estate, law, e-commerce, or one you have prior network exposure to).
Week 1-2. Build your packaged offer. Write a one-page scoping doc: 3-step packaged automation, fixed deliverables, fixed price ($2,500-$4,500), 2-3 week delivery. Build the workflow in Make or n8n on your own data first so you can demo it.
Week 2-3. Pull MSA + SOW templates from Bonsai. Add the AI disclosure clause and the GDPR DPA template. Get an attorney to review for $300-$500 before your first signed contract over $3K.
Week 3-4. Post on LinkedIn announcing the offer to your network. Reach out to 30 warm contacts in your chosen vertical. Aim for 1-2 signed builds by end of week 4.
Week 4-7. Deliver the first build. Document every scope question, exception case, and "thing the client asked for that wasn't in scope" — this becomes your scoping doc v2.
Week 7-8. Capture a written case study + testimonial from client #1. Apply for Make Partner status. Quote client #2 at the full $4,500 price (no discount).
Week 8-10. Pitch every completed build into a $1,000-$1,500/month maintenance + expansion retainer. Aim for 50% retainer attach rate.
Week 10-13. Target: 2-3 packaged builds delivered or in flight + 1-2 retainer clients = $4,000-$7,000 month 3. Raise package pricing 20% for the next round of inbound.

Common Pitfalls

No exception handling on LLM steps. A GPT-4o invoice parser will silently misread an edge-case PDF and route the wrong amount downstream. The client processes 200 wrong records before noticing. Every LLM step needs a confidence-score check, an exception queue, and a human review point for consequential decisions. Build this on the first automation, not the fifth.
Letting clients pay you for the platform subscription. Client churns, you forget to cancel their Zapier account on your card, you eat $200/month for nine months. Always have the client pay platform fees on their own credit card on their own account. You bill build + maintenance only.
Generalist pitch instead of vertical pitch. "I automate business processes" closes 5% of discovery calls. "I build new-patient onboarding for dental practices" closes 30%. Pick one vertical for the first 10 clients, even if you think you're leaving money on the table. You're not.
Skipping the AI disclosure clause and DPA. The biggest legal exposure in this business isn't the code breaking — it's the client's end-users receiving AI-generated content with no disclosure, or EU personal data flowing through your automations with no processor agreement. Both fixes are paperwork, not engineering. Don't sign a build contract without them.

Get your full launch plan — take the free 60-second quiz.