Every number in this lesson is an assumption for the sake of the example. It isn't a forecast, a promise of income, or financial or tax advice. Service prices are given as of October 2026 only where they've been verified; for current prices and versions, see What's current. Work out everything else from your own data.
The gist
Most AI builders know their costs. "I pay Anthropic $200, Vapi $100, infrastructure $50, so $350/month total." And that's where their financial analysis ends.
But that isn't unit economics. That's total billing. Total billing tells you "how much I'm spending." Unit economics answers a different question: how much does each customer bring in vs how much do they cost, and where does the math break once I start growing?
Without this math you're flying blind. You don't know whether ads will pay for themselves. You don't know how many customers you need to break even. You don't know when to hire a second person. You don't know whether your $9.99/month SaaS price is sinking you or feeding you.
This lesson is the full financial math for 3 typical levels of an AI business in 2026. With worked calculations, a breakdown of pitfalls, and a checklist to test your own model.
The core formula (memorize it for life)
Unit Economics health = LTV - CAC > 0
Four numbers you should know about your business by heart:
| Metric | Formula | What it means |
|---|---|---|
| LTV (Lifetime Value) | avg revenue/month × avg months retained × margin | How much a customer brings in over the whole relationship |
| CAC (Customer Acquisition Cost) | total marketing spend ÷ new customers | How much it costs to win one customer |
| Margin (gross profit) | (revenue - variable costs) ÷ revenue | The % of revenue left after direct costs |
| Payback period | CAC ÷ (monthly revenue × margin) | How many months it takes to earn back CAC |
Rules of thumb from SaaS practice (heuristics, not laws):
- LTV/CAC ratio ≥ 3:1 (below that, it's fragile)
- Gross margin ≥ 70% (below that, revisit your pricing or your stack)
- Payback period ≤ 12 months (B2B) or ≤ 6 months (consumer)
- Concentration risk: no single customer is more than 30% of total revenue
Next comes a breakdown by 3 business levels, using made-up numbers. The revenue ranges for the levels are arbitrary: they classify business size for the calculation, they aren't a goal or a forecast.
Level 1: Solopreneur AI service (roughly: revenue up to a few thousand $/month)
Who this is: an AI consultant, a freelance content creator, an individual coach who uses AI, a narrow B2B "done-for-you" service. One person, 1-10 active clients.
Costs (fixed monthly)
| Tool | $/month | What for |
|---|---|---|
| Claude Pro | $20/month (or $17/month annual = $200/year), as of October 2026 | Main workflow |
| ElevenLabs Starter | $6, as of October 2026 | Voice (if your product has it) |
| Cloudflare Workers | $0, as of October 2026 | The free tier covers solo volume |
| Notion (paid plan) | assumption: $10 | Client workspace, knowledge base |
| Domain + email | assumption: $2 | $24/year, spread out by month |
| Total fixed | ≈ $38 (monthly Pro) / ≈ $35 (annual Pro) | The bare minimum |
Variable costs (per client)
- API tokens (Anthropic): assumption: $0.50-2 per active interaction
- Active client: assumption: ~$5-15/month (depends on how heavily they use it and which model you pick)
Pricing (assumptions for the example, not market data)
- AI consulting: $200-500/month per client
- Content-as-a-service: $300-800/month (5-15 posts / scripts)
- Coaching + AI workflows: $300-1000/month
Contribution per client: $300 revenue − $15 variable = $285 contribution margin (95%)
Unit economics math
| Metric | Value |
|---|---|
| Avg revenue per client | $300/month |
| Variable cost per client | $15/month |
| Gross margin | 95% |
| Avg retention | 4-8 months (assumption) |
| LTV | $300 × 6 × 0.95 = $1710 |
| CAC (cold outreach + time spent on content) | $30-100 (if you do it yourself, it's your time at $30/hour × 1-3 hours) |
| LTV/CAC ratio | 17:1 to 57:1 (very high in this teaching example; in real life, measure it on your own data) |
| Payback period | <1 month |
Break-even
Fixed costs are about $40/month. One client covers fixed costs with room to spare ($285 contribution).
In the teaching example: 3 clients × $285 − $40 ≈ $815, and 5 clients × $285 − $40 ≈ $1385 per month, before taxes and without paying yourself for your time. Real results depend on whether you find clients and keep them.
Scale ceiling
- Solo capacity limit: in the example, 10-15 active clients (even with AI, you run out of time for onboarding, support and calls)
- Revenue ceiling in the example: $3000-5000/month before you need a team
- After that, either hire or move to Tier 2 by turning the service into a SaaS product
Honest verdict
In the example's math, a solo AI service has high margins and a fast return on what you put in, at low risk. But everything hinges on whether you find clients, and on your time. Decide based on your own data whether to productize or hire.
Level 2: SMB / small agency (roughly: revenue $5K-30K/month)
Who this is: an AI agency serving 5-20 clients, a micro-SaaS with 50-300 paying users, a B2B tool for a narrow niche. 1 founder + 0-2 contractors.
Costs (fixed monthly)
| Tool | $/month | What for |
|---|---|---|
| Claude Max 5x | $100/month, as of October 2026 | Main dev + delivery workflow |
| Anthropic API budget | assumption: $200-500 | Production traffic. Prices per 1M tokens as of October 2026: Sonnet 5.5 $2/$10, Opus 5.5 $4/$20 |
| OpenAI API (fallback) | assumption: $50 | Backup. For example, gpt-6.1-sol at $2/$10 per 1M tokens as of October 2026 |
| Vapi (voice, if you have it) | assumption: $100 | Minutes + numbers; Vapi's platform fee is $0.05/min, everything else is billed separately (see the Call Support AI lesson) |
| Cloudflare Workers Paid | from $5, as of October 2026 | Beyond the free tier |
| PostHog (analytics) | assumption: $30 | Product metrics (there's a free allowance) |
| LLM observability (LangSmith or similar) | assumption: $20 | Token tracking, debugging |
| GitHub Pro | assumption: $4 | Private repos |
| Stripe | $0 fixed | 2.9% + 30¢ in the US, different in other countries |
| Total fixed | ≈ $510-810 with the assumptions above | Depends on Vapi |
Variable costs
- API tokens per customer: assumption: $20-100/month (depends on the product)
- At 10 customers: $200-1000/month variable in total
Pricing (assumptions for the example, not market data)
- B2B AI service: $300-1500/month per customer
- Micro-SaaS subscription: $29-99/month per user
- Hybrid (subscription + overage): $99 floor + usage
Unit economics math (B2B service example)
| Metric | Value |
|---|---|
| Avg revenue per customer | $500/month |
| Variable cost per customer | $60/month (API + voice + share of tools) |
| Gross margin | 88% |
| Avg retention | 12-24 months (assumption; B2B is usually more stable) |
| LTV | $500 × 18 × 0.88 = $7920 |
| CAC (paid ads + content + sales time) | $100-400 |
| LTV/CAC ratio | 20:1 to 80:1 |
| Payback period | <2 months |
Break-even
Fixed costs ~$700/month. Contribution per customer is $500 − $60 = $440, so ~2 customers cover fixed costs.
In the teaching example: 10 customers × $440 − $700 ≈ $3700 per month, before taxes and without paying yourself for your time.
Scale ceiling
- With 1 person: in the example, 20-30 customers max (capacity runs into onboarding + support + delivery)
- Revenue ceiling in the example: $10K-30K/month before you need contractors or employees
- Beyond that: move to Tier 3, or choose not to grow (a lifestyle business)
Honest verdict
In the example's math, Tier 2 gives good margins (85%+) and a reasonable CAC, and B2B clients tend to be sticky. The main thing is not to spread yourself thin: one niche, one product, repeatable delivery.
Level 3: Production / mid-size business (roughly: revenue $30K-300K/month)
Who this is: an established SaaS with 100-1000 paying users, an agency serving 20-100 B2B clients, a vertical AI product with a team of 3-10 people.
Costs (fixed monthly)
| Tool | $/month | What for |
|---|---|---|
| Anthropic API | assumption: $1000-5000 | Heavy production. Prices per 1M tokens as of October 2026: Sonnet 5.5 $2/$10, Opus 5.5 $4/$20 |
| OpenAI API (fallback + alternative model) | assumption: $200 | Redundancy for specific tasks |
| Vapi production | assumption: $500-2000 | Voice at scale, numbers, minutes |
| Infrastructure (Cloudflare/AWS) | assumption: $200-800 | Beyond Workers: databases, R2 |
| LangSmith (LLM observability) | assumption: $50 | Production traces, debugging |
| Sentry (error tracking) | assumption: $50 | App-level monitoring |
| HubSpot Starter (CRM) | assumption: $20 | Sales pipeline |
| PostHog | assumption: $30 | Product analytics |
| Claude Team (5 seats) | $100-125, as of October 2026 | Team Standard is $25 per seat billed monthly or $20 billed yearly (×5) |
| Custom domain + SSL + email | assumption: $20 | Production grade |
| Legal / accounting (spread out by month) | assumption: $500-1500 | Contracts, taxes, compliance |
| Stripe (variable %) | 2.9% of revenue (US) | Payment processing |
| Total fixed | ≈ $3000-10000 with these assumptions | Depends on team size; team salaries aren't included in this table |
Variable costs
- Per customer: assumption: $50-500 (depends heavily on the use case)
- Sales commissions: assumption: 10-20% on new ARR (if you have an SDR)
- At 50 customers: $2500-25000 variable
Pricing (assumptions for the example, not market data)
- B2B SaaS: $500-5000/month per customer
- Enterprise contracts: $999-4999/month or annual deals of $30K-100K
- Tiered SaaS: $99 / $499 / $999/month
Unit economics math (B2B SaaS example)
| Metric | Value |
|---|---|
| Avg revenue per customer | $1500/month (across tiers) |
| Variable cost per customer | $250/month |
| Gross margin | 83% |
| Avg retention | 18-36 months (assumption) |
| LTV | $1500 × 24 × 0.83 = $29,880 |
| CAC (paid + SDR + content + tools) | $1000-3000 |
| LTV/CAC ratio | 10:1 to 30:1 |
| Payback period | ≈ 1-2.5 months by the lesson's formula (CAC ÷ (revenue × margin)); with a long sales cycle it takes longer in practice |
Break-even
Fixed costs $7K. Contribution per customer is $1500 − $250 = $1250, so ~6 customers cover fixed costs. Team salaries aren't in this calculation: with salaries, the break-even point moves noticeably higher.
In the teaching example: 30 customers × $1250 − $7K ≈ $30,500 per month, before taxes, salaries and commissions.
Scale ceiling
- With a team of 3-10 people: in the example, 200-500 customers
- Revenue ceiling in the example: $300K/month ($3.6M ARR) before you need a major organizational overhaul (Series A, departments, a real sales org)
Honest verdict
Tier 3 is a business you manage rather than one you run with your own hands. In the example, margin drops from 95% to 80% because of the team and infrastructure, but the absolute numbers grow by an order of magnitude. In practice, the payback period is usually longer than the formula says (long sales cycles, discounts), so you need capital or patient cash flow.
3 full setups with the math (typical AI products of 2026, made-up numbers)
Setup A: AI Content Studio
Stack: the lessons AI copywriting, The full content pipeline and Social media automation.
| Metric | Value |
|---|---|
| Pricing (assumption) | $1500/month for 30 posts/month for a client social media agency |
| Variable per client | $40/month (assumption: Claude API + Buffer + image tools) |
| Gross margin | 97% |
| 1 person handles | 5-8 clients (assumption) |
| Revenue ceiling | $7500-12000/month solo (in the example) |
| LTV (avg 12 months) | ≈ $17.5K |
| CAC | $200-500 (assumption) |
| Payback | <1 month |
Verdict: in this example, the margin is the highest of the three setups: minimal variable costs, maximum leverage. The main risk is commoditization (low barrier to entry).
Setup B: AI Customer Support Bot
Stack: the lessons AI Customer Support, Call Support AI and AI chat on your website (RAG + chatbot + voice).
| Metric | Value |
|---|---|
| Pricing (assumption) | $500/month for a 24/7 chatbot that replaces support for a small business |
| Variable per client | $80/month (assumption: Claude API + voice minutes for the whole stack + Pinecone) |
| Gross margin | 84% |
| 1 person handles | 20+ clients (mostly autopilot after setup, assumption) |
| Revenue ceiling | $10K+/month solo (in the example) |
| LTV (avg 18 months, sticky B2B) | ≈ $7.6K |
| CAC | $300-700 (assumption) |
| Payback | 1-2 months |
Verdict: in the example, the best scaling potential. A set-it-up-once-then-mostly-passive model. Clients are sticky because the bot replaces a human support team. The main risk: the quality bar is high (one bad answer = churn).
Setup C: AI SaaS Tool
Stack: the lessons Final architecture and Build-Along: SaaS MVP (productized SaaS) + a custom feature stack.
| Metric | Value |
|---|---|
| Pricing (assumption) | $49-99/month SaaS with a freemium funnel |
| Variable per user | $5/month avg (assumption: heavy users $30, light $0.50) |
| Gross margin | 85-92% |
| Scale | 1000+ users, everything automated |
| LTV (avg 14 months retention) | ≈ $830-900 |
| CAC | $30-80 (organic + content) or $150-300 (paid), assumption |
| Payback | 1-6 months |
| Upfront investment | $20K+ (product development + 6-12 months to product-market fit), assumption |
Verdict: in the example, the highest ceiling and the longest payback. It needs upfront capital and doesn't guarantee the product will find its users.
Hidden costs (the ones people forget)
| Cost | How much (example) | Why it matters |
|---|---|---|
| Your time | $30-100/hour × 20 hours of setup = $600-2000 | The equivalent cost. At scale = $10-30K/year |
| Taxes | The rate and rules depend on your country and business structure | Calculate after taxes, not gross. Check with an accountant or tax advisor where you live |
| E&O insurance (B2B) | The cost depends on your country and the scope of work | Protection against client lawsuits |
| Refund / dispute buffer | 1-2 months of revenue in reserve | Stripe disputes, refunds, chargebacks |
| Cash flow gap | 2-7 days for Stripe payouts | Don't confuse revenue with cash you can actually use |
| Bad debt | 1-3% of revenue | Clients who don't pay, go bankrupt or dispute |
| Tool sprawl | $50-200/month of "invisible" subscriptions | Once a quarter, audit and cut |
Buffer and safety margin (survival rules)
| Rule | Minimum | Why |
|---|---|---|
| 20% buffer on top of the plan | Always | Experiments, an unexpected API spike, an urgent fix |
| Gross margin ≥ 70% | Mandatory | Below that, the business is fragile and any shock can break it |
| 3-month runway | Minimum | Fixed costs × 3 in reserve in case of churn waves |
| Customer concentration <30% | Health rule | One client >30% of revenue = fragile |
| Provider redundancy | At least Anthropic + OpenAI | An outage shouldn't take down your business |
3 typical pitfalls (what breaks unit economics)
Pitfall 1: Hidden complexity per customer
Symptom: every client needs a little something custom: custom prompt tuning, a custom integration, a custom report. Variable cost doesn't scale linearly; it grows quadratically.
Impact on the math: margin drops from 88% to 60% at 20 clients.
Fix: standardize your offerings strictly. Productize delivery: a playbook, templates, automated onboarding. Custom work = a premium add-on with a clear price.
Pitfall 2: CAC grows faster than LTV
Symptom: customers used to come in at $50 CAC, now it's $300. LTV isn't growing. LTV/CAC ratio goes 20:1 → 4:1 → 2:1.
Impact on the math: the payback period goes from 2 months to 8 months. Cash flow breaks.
Fix: retention first, growth second. If retention is 6 months, pause acquisition and fix the product/onboarding. Rising CAC with flat LTV = "pouring water into a leaky bucket."
Pitfall 3: A low SaaS price with heavy API users
Symptom: SaaS at a flat $9.99/month. 10% of users burn $30/month in API costs. The math: $9.99 revenue − $30 variable = −$20 contribution per heavy user.
Impact on the math: the more heavy users you have, the more you lose.
Fix: usage-based pricing + tier limits. A $9.99 floor + 1M tokens, overage beyond that. Or tiers at $9.99 / $49 / $199. Heavy users pay more.
Free hosting tiers: the terms change
| Platform | What's known as of October 2026 | What to check |
|---|---|---|
| Vercel Hobby | Free, but only for personal non-commercial projects; commercial projects need Pro | Limits and terms on Vercel's site |
| Cloudflare Workers | Free: 100,000 requests per day | CPU limits and Paid plan terms |
| Render, Fly.io, Railway | Their free tier terms have changed in recent years | Open their pricing pages before you choose: the free tier may have disappeared or become a trial |
Insight: free tier terms change fast, so check them again before every project. The free Vercel Hobby plan isn't an option for a commercial project.
Free vs Pay-as-you-go vs Subscription (decision matrix)
| Model | When it works | When it hurts you |
|---|---|---|
| Free tier | Product-led growth, >5% conversion to paid | Conversion <2%, free users burn API costs |
| Pay-as-you-go | Irregular usage (API, voice minutes, batch jobs) | Predictable usage: clients want a fixed bill |
| Subscription (flat) | Regular usage, predictable revenue for you | Heavy users eat your margin |
| Hybrid (subscription + overage) | A good choice in many cases | Harder to explain to the client |
Default recommendation for 2026: Hybrid. A subscription floor (predictable revenue) + overage charges (protects your margin from heavy users).
Prices of typical tools (as of October 2026, checked on the services' websites)
| Category | Tool / Tier | Price/month |
|---|---|---|
| AI coding IDE | Cursor Pro | see the service's pricing page |
| AI coding IDE | Devin Desktop (formerly Windsurf) Pro | see the service's pricing page |
| AI coding | GitHub Copilot Pro | see the service's pricing page (since June 1, 2026, chat and agents use token-based credits) |
| AI writing | Claude Pro | $20, or $17 billed yearly |
| AI writing | ChatGPT Plus | $20 |
| AI writing | Google AI Pro | $19.99 in the US (formerly Gemini Advanced) |
| AI voice | ElevenLabs Starter / Creator | $6 / $22 |
| AI image | Midjourney Basic | see the service's pricing page |
| AI music | Suno Pro / Premier | $8 / $24 |
| Payments | Stripe | 2.9% + 30¢ in the US (no monthly fee) |
Prices change: open the service's pricing page before you run your numbers. For the prices we track, see What's current.
Pricing psychology + price points (reference points, not market data)
| Tier | Price range | Audience | Sales cycle |
|---|---|---|---|
| Prosumer / individuals | $9-29/month | Solopreneurs, hobby users | Easy decision, instant signup |
| Small business | $49-99/month | Teams of 1-10 employees | Needs an ROI calculation, 1-3 day decision |
| Mid-market | $199-499/month | 10-100 employees | Procurement review, 1-4 weeks |
| Enterprise | $999-4999/month | 100+ employees | Contract + legal + security review, 1-6 months |
| Strategic | $5000+/month | Custom contracts | Multi-year deals, negotiated |
The main insight: there are psychological gaps between tiers. $29 → $99 is an easy sell (3x). $99 → $499 is a different decision-making process (you need a champion and an ROI case). $499 → $4999 is a different sales motion (you need a salesperson).
3-tier summary table
The numbers in the table come from the teaching examples above. They're assumptions, not a forecast.
| Level | Fixed costs/month | Rough revenue scale | Margin | Time to profit in the example | Ceiling in the example |
|---|---|---|---|---|---|
| Solo | ≈ $35-40 | up to a few $K/month | 90-95% | <1 month | $3-5K/month |
| SMB | ≈ $510-810 | $5K-30K/month | 80-88% | 2-3 months | $10-30K/month |
| Mid-business | ≈ $3K-10K | $30K-300K/month | 70-85% | 6-12 months | $300K/month |
Healthy unit economics checklist
If 3 or more boxes are unchecked, your financial model is fragile. That's not a disaster; it's a signal of what to fix next.
Practice: calculate your own model right now
Step 1: Gather your fixed costs
Open every subscription dashboard (Anthropic, Stripe, Cloudflare, GitHub, Notion, any other SaaS). Write down each subscription with its price. Add them up. That's your fixed monthly burn.
Step 2: Calculate variable cost per customer
Take last month. Total API spend ÷ active customers = avg variable cost per customer.
Step 3: Calculate avg revenue per customer
Total revenue ÷ active customers for the same month.
Step 4: Calculate retention
How many months, on average, a customer pays before they churn. If your business is less than 12 months old, estimate pessimistically (6 months B2C, 12 months B2B).
Step 5: Calculate LTV
avg revenue × retention months × gross margin
Step 6: Calculate CAC
Total acquisition spend (ads + content production cost + your sales time × hourly rate) ÷ new customers last month.
Step 7: Compare with the benchmarks
- LTV/CAC ≥ 3:1?
- Margin ≥ 70%?
- Payback ≤ 12 months?
If all three are yes, you're healthy. If even one is no, you have a concrete thing to work on next quarter.
Tracking tools
- Stripe Dashboard: MRR, churn, ARR out of the box
- Baremetrics: SaaS metrics on top of Stripe (LTV, churn, MRR cohorts)
- ChartMogul: an alternative to Baremetrics
- Claude Console: token spend tracking per project
- LangSmith: LLM cost attribution per customer (Helicone also works, but since March 2026 it's been in maintenance mode)
- PostHog: product analytics + revenue tracking
- Google Sheets + a monthly review: the simplest option if revenue is under $10K/month
Key takeaways
Unit economics isn't accounting. It's a compass. Without LTV / CAC / margin / payback you're flying blind and don't know whether to run ads, hire a second person, or cut features. Your numbers are your main decision-making tool.
Reference points for a healthy AI business: margin ≥ 70%, LTV/CAC ≥ 3:1, payback ≤ 12 months, concentration <30%. These are heuristics, not guarantees. If even one number is off, fix it before you grow. Growing on broken economics = speeding toward a wall.
Tier 1 (solo) has the highest margin and lowest risk in the examples. Tier 2 (SMB) gives a good ratio to hours worked. Tier 3 (production) has the largest absolute numbers and the longest payback. Each tier needs its own mental model: you can't run Tier 3 like Tier 1.
Sources
- Y Combinator Startup School — Unit Economics
- Claude Pricing (Anthropic moved from anthropic.com → claude.com)
- Claude API Docs Pricing
- Stripe Pricing
- PostHog Pricing (free tier)
- Vapi Pricing (voice)
- ElevenLabs Pricing
- Cloudflare Workers Pricing
Next lesson
The mark stays in this browser only and is never sent anywhere. My progress