I run five small businesses from Brooklyn with zero full-time employees. The whole AI and automation stack costs about $600 a month. Forbes published the summary version on April 15 (Gene Marks, Quicker Better Tech column). This is the itemized receipt.
No marketing talk. No "you can replace your whole team" hype. Just the actual tools, the actual prices, and what each one does.
| Tool | What it does | Monthly |
|---|---|---|
| Claude API (Anthropic) | Core reasoning for voice agent scripts, outreach sequences, review drafting, P&L summarization, content writing | ~$180 |
| Voice agent platform (Retell / Vapi tier) | B2B outbound calls in 3 languages, 100-200/day for OD Granite | ~$120 |
| ElevenLabs | Voice synthesis for the agent in Ukrainian and Russian | ~$44 |
| OpenAI API | Secondary reasoning and speech-to-text backup | ~$40 |
| Telegram Bots (custom) | Morning P&L, inventory alerts, ad spend watchdog, news digest | $0 (self-hosted) |
| n8n self-hosted | Cross-business pipeline orchestration | $0 (Hetzner €6 VPS) |
| Hetzner VPS | Hosts n8n, cron jobs, Telegram bots, Playwright scrapers | ~$8 |
| Playwright (open source) | Scraping, posting, browser automation | $0 |
| Amazon SP-API | Listing sync, pricing, review responses | $0 (free tier) |
| Shopify API | Order sync into P&L pipeline | $0 (included with Shopify plan) |
| Stripe API | Revenue data into P&L | $0 (free) |
| Perplexity API (occasional) | Research queries for intel digest | ~$20 |
| Zapier premium (legacy, some automations still there) | Simple integrations before n8n | ~$50 |
| Notion API | Data storage for P&L pipeline | ~$10 (paid plan) |
| GitHub | All scripts and configs | $0 |
| Cloudflare (website, CDN) | negodiuk.ai hosting | $0 |
| Domain renewals (6 domains) | Prorated | ~$10 |
| Total | ~$482-600 |
The total varies month to month based on call volume (voice agent scales with calls) and Claude token usage. $482 is a light month. $600 is a normal-to-heavy month.
Every decision the stack makes that needs judgment routes through Claude. Voice agent scripts dynamically adapt based on prospect responses. Email sequences personalize based on firmographic data. Amazon review responses drafted individually for each new review, not templated. Cross-business daily P&L summary written in natural language. When I pitch a new idea, Claude drafts the first version of the content before a human sees it.
Claude costs scale with tokens. Most months I run 80-120 million tokens across all five businesses combined. That works out to roughly $150-220 depending on model mix (I use Sonnet for most routine, Opus for high-stakes work).
I use a combination of Retell for the outbound flow orchestration and Claude for the conversation reasoning. ElevenLabs handles the voice synthesis in Ukrainian and Russian (English uses the platform native voice). The voice agent is set up once per business, then lives on the platform. Cost scales with call minutes. For 100-200 cold calls a day, I run around $120 a month in platform fees plus $44 in voice synthesis.
n8n runs self-hosted on a $8 Hetzner VPS. Every cross-business automation lives there. The daily P&L pipeline. The social posts scheduler. The ad spend watchdog. The inventory reorder trigger. The morning intelligence digest from 85 Telegram channels.
Moving off Zapier to n8n cut my automation bill from $200 a month to $8. The only catch is that n8n takes a Saturday to set up properly. After that, the marginal cost of a new automation is zero.
All three have free API tiers that cover a small operator. The API calls go into the n8n pipeline, which calls Claude to summarize and Telegram to deliver.
Every one of those tools I used to pay for. Every one was replaceable with Claude plus a cron job.
I get asked weekly: "what human roles does this stack replace?" My answer is specific. Two to three full-time operators.
Total payroll avoided: roughly $145,000 a year in base compensation, plus about $30,000 a year in benefits, management overhead, and tools for those three people. Call it $175,000 a year in total cost.
Stack cost: $600 a month times 12 months = $7,200 a year.
Net savings: about $167,800 a year.
This is also why I pay zero employees. Not because I do not want to. Because the math does not yet say hire.
I get asked less often but this is the more honest part. The stack breaks roughly once a week.
Every single one is fixable from my phone in under 20 minutes. That is also the skill. You do not need an engineering team to operate this. You need someone who can read a log file.
That is what a Fractional AI Officer actually is. Someone who sets this up, then fixes it when it breaks.
Yes. Everything on the list is public. Claude API docs are good. n8n tutorials are on YouTube. Retell and Vapi both have starter templates. If you are technical, you can get to 80 percent of this stack in two or three weekends.
The remaining 20 percent is the part that takes a year of running it on your own business to learn. What breaks first at scale. Which prompts drift over time. When to trust the agent and when to pull it out. That is the expensive part. It is the only thing I actually sell as a service.
Want me to build this for your business?
I deploy a variant of this stack for $1M-$10M revenue businesses in 2 to 8 weeks. Audit at $2,500, sprint at $5,000-$8,000, full deployment at $10,000-$15,000, retainer at $3,000-$5,000 a month. Every playbook is field-tested on my own five businesses first.