Business basics for vibe coders: what to know before you launch

If you search for the business basics you need before launching an app you built with AI, most answers fall into two camps. One is a list of fifty startup terms with no numbers attached. The other is classic SaaS advice written before every user action triggered a paid model call. Neither tells you whether your app makes money on each customer.

That question matters more now. In April 2026 GitHub announced it was moving Copilot to usage-based billing because "a quick chat question and a multi-hour autonomous coding session can cost the user the same amount," and said the old model was "no longer sustainable" (GitHub blog). If a company that size had to reprice around inference cost, a solo builder should do the math before launch.

This post covers five concepts that decide whether an AI-built app becomes a business, works one made-up example end to end, and ends with a pre-launch checklist. At the end we point to a 45-lesson course that teaches the full set.

1. Size the market from the bottom up

"There are 30 million small businesses, we only need 1%" is a top-down estimate, and it hides every assumption that matters. A bottom-up estimate makes you name three numbers: how many customers you can actually reach, what share you might win, and what each one pays. OpenStax's entrepreneurship text lists market demand, market size and structure, and margins as the screening questions for any opportunity (OpenStax).

Start narrow. Y Combinator's advice is that "a small group of customers who love you is better than a large group who kind of like you" (YC). Define the segment by behavior ("plumbing firms that lose leads because quotes take more than a day") rather than a label ("small businesses").

2. Contribution margin, including AI cost

Contribution margin is "the amount by which a product's selling price exceeds its total variable cost per unit," and it is what each unit contributes toward fixed costs (OpenStax). For an app, variable costs are anything that grows with a customer's usage: payment processing, hosting, support, and model tokens.

Model cost is the line builders forget. Cursor's pricing terms, for example, calculate model fees from total tokens across input, output and cache (Cursor pricing policy). Your app works the same way: every prompt has a cost, and heavy users cost more. Calculate margin by segment, not just on average.

3. Break-even

Break-even is how many customers you need for total contribution to cover fixed costs: your salary, tools, base infrastructure. Cost-volume-profit analysis studies what happens when price, variable cost, fixed cost or volume changes (OpenStax). Treat it as a sensitivity test, not a forecast. The useful output is which input moves the answer most.

4. CAC, payback and LTV

Customer acquisition cost (CAC) is what you spend to win one customer. Payback is how many months of contribution it takes to earn that back. Lifetime value (LTV) is the contribution a customer brings in before they churn. David Skok's SaaS Metrics 2.0 says the best SaaS businesses have an LTV:CAC ratio "higher than 3," that many recover CAC "in 5-7 months," and that profitability is "anemic" when recovery runs past 12 months (SaaS Metrics 2.0).

Skok also notes that early companies often miss these guidelines and improve toward them. For a first pass, we use a rough formula: monthly contribution divided by monthly churn rate. It ignores discounting, so treat it as a sanity check rather than a valuation.

5. Pricing: pick the unit before the number

Stripe's Atlas guide treats packaging (what each plan includes and limits) as the main decision, with the price point downstream (Stripe Atlas). For AI apps, the unit matters because cost varies by user. GitHub's fix kept base prices (Copilot Pro at $10 a month) and added included AI credits with token-metered usage beyond them (GitHub blog). The same Stripe guide advises pricing "a bit ahead of where you think your maturity is," since raising prices on existing customers later is harder.

A worked example (made-up numbers)

Say you built QuoteBot, an AI tool that drafts job quotes for small plumbing and HVAC firms. Every number below is invented except the card fee, which is Stripe's standard US rate of 2.9% + 30¢ (Stripe pricing).

Market. You can reach 20,000 firms through one trade-association channel. If you win 3%, that's 600 customers at $29 a month, or about$209,000 a year.

Contribution per typical customer, per month:

Break-even. Fixed costs are $6,000 a month, so you need 6,000 / 18.36 = 327 customers.

Acquisition. Paid search costs $150 per customer. Payback is 150 / 18.36 = 8.2 months. With 418.36 / 0.04 =$459, so LTV:CAC is about 3.1.

That looks acceptable. Now add the heavy users. Suppose 20% of customers run quotes all day and burn $22 a month in model usage instead of$6. Their contribution drops to $2.36. The blended picture changes:

Break-even now needs two-thirds of your entire 600-customer estimate. Nothing about the product changed. One usage pattern ate the margin. The fix is a pricing decision: include a monthly quote allowance in the $29 plan and charge for usage above it, so heavy users pay for the cost they create.

A pre-launch checklist

This is our own checklist, built from the sources above:

1. Write your segment as a behavior, and estimate reachable customers through one named channel.
2. List every variable cost per customer, including model tokens at light, typical and heavy usage.
3. Calculate contribution margin for each usage level, not just the average.
4. Compute break-even customers and compare them to your market estimate. If break-even needs most of the market, change something before launch.
5. Estimate CAC for your first channel and check payback against Skok's 12-month warning line.
6. Choose a pricing unit that makes heavy users pay more, then set the number.

Learn it hands-on

MBA for Vibe Coders is a 45-lesson course that teaches MBA-level business judgment to people who ship apps with AI tools. After an orientation lesson, eight modules of five lessons each cover business models and market sizing, accounting and unit economics, finance and fundraising, pricing and growth, customer discovery, operations, leadership, and law and AI risk. Each module ends with a case lab built from public filings and company sources, including Figma, the pricing of Cursor, Lovable and Replit, Duolingo, and GitHub Copilot. Lessons include runnable calculators like the ones above.

The four-lesson capstone has you build a Venture Operating System for a real or plausible app: a venture thesis, a financial model with AI cost sensitivity, a go-to-market plan, an operating dashboard, a risk register, and a board-style update.