Skip to content
The Lean CPTO
Menu
Get the kit · US$47

Guides

What an AI-built MVP really costs: 5 places your money leaks

Short answer: the tool subscription is the smallest part of the bill. An AI-built MVP costs you in five places: build credits, hosting and database, AI inside your product, paying a human when the AI gets stuck, and your own time. Most people only budget for the first one. Write all five down, budget for your worst month, and pause anything nobody has paid for. The cheapest credit is the one you never spend.

Why the bill is never what you expected

Your AI MVP was meant to cost the price of a subscription. Then the credits ran out. Then the hosting bill arrived. Then the AI started charging you to fix its own bugs.

AI didn’t cure our urge to overbuild. It made it worse. Building has never been easier. Which is exactly the problem.

I’ve spent more than 10 years coaching founders, and this part hasn’t changed. They loved it in the meeting. Then they built the tech anyway. I get why. The manual version isn’t sexy.

There are three ways the money goes.

Problem 1: you budgeted for the subscription and nothing else

The subscription is the number on the pricing page, so it’s the number that goes in the plan. The rest shows up later, one bill at a time. If you only look at last month, you’ll think it’s cheap. Then one bad month arrives.

The fix: sell it first, so you know what the money is for. Then write down all five costs, with last month and your worst month side by side.

Problem 2: you’re paying for the same bug three times

You asked it to fix the login. It broke the dashboard. So you ask it to fix the dashboard, and the login breaks again. Each try uses credits, whether it worked or not. The AI charges you to create the bug, charges you to fix it, then puts it back.

The fix: prove each change before you make the next one. One change per prompt. Save a working version first. After three failed fixes, stop.

Problem 3: you’re paying to run something nobody has bought

This is the big one. Every tool on your list is a monthly cost for a product with no customers yet. If you’re a business owner, it’s a cost for a tool nobody has signed off or started using.

The fix: build only what someone has paid for. Only what someone will pay for this month. Park the rest.

The 5 places your money leaks

  1. Build credits. Lovable, Bolt, Replit, Cursor and the like. Debugging loops burn the most.
  2. Hosting and database. Traffic, bots and bugs can all spike these bills. Add the small stuff here too: domain, email, forms and payments.
  3. AI inside your product. If your product calls an AI model, that cost grows with every customer who uses it.
  4. Paying a human. The developer you call when the AI can’t fix it. The bill grows the longer you wait.
  5. Your time. The one nobody counts. Weeks in debugging loops that a developer could fix in hours.

What it looks like on one page

Here’s a made-up example, in US dollars a month. It isn’t a real client, and tool prices change often, so use your own numbers.

  • Build credits: $25 last month, $85 in the worst month.
  • Hosting: $20, and $20.
  • Database: $0 on a free plan, and $0.
  • AI inside the product: $30, and $90.
  • Developer fix: $0, and $300.
  • Domain, email and forms: $12, and $12.
  • Total: $87 last month, $507 in the worst month.

Now add your time. Five hours a week fixing and prompting, at $100 an hour, is about $2,167 a month. In this example your time costs more than four times the worst month of tools. Use your own hourly rate, but do count it.

Build your cost map in 20 minutes

Pen and paper is fine.

  1. List every tool you pay for, under the five headings.
  2. Write down two numbers for each: last month’s bill and your worst month.
  3. Mark what you need before someone pays. Be honest. Most of the list can wait.
  4. Add up your time. Hours a week, at your rate.
  5. Circle your three biggest numbers. That’s where to start.

Three rules for your budget

  1. Budget for your worst month. In the example that’s $507, not $87. Credits and AI usage spike when things break.
  2. Pause what nobody has paid for. Anything you marked as not needed waits until a buyer pays. Then the business funds it.
  3. Keep AI cost under about 30% of your price. That’s my rule of thumb. If it’s higher, change your price or your AI model before you add customers.

Switch on caps and alerts today

Many tools have no spending limit on by default, and some budgets only warn you. Check each of these now.

  1. Set a monthly ceiling before anyone pays.
  2. AI builder: usage limit on, automatic top-ups off.
  3. Hosting: set a spend limit, and check whether it pauses the service or only sends a warning.
  4. Database: set a budget alert, and stay on the free plan while you can.
  5. AI provider: set a monthly limit and an email alert.
  6. Payment: use one low-limit card for every MVP tool.

The biggest cost cut is not building yet

The cheapest version of your MVP is the one you deliver by hand. AI does the heavy lifting, and you deliver the result. All you need is email, a form, a spreadsheet, a payment link and one AI chat plan. That’s a small monthly bill compared with the example above, and there are no debugging loops.

Sell first. Deliver it by hand, with AI doing the heavy lifting. Build only what paying customers keep asking for.

When paying customers need the same manual step every week, build that step, nothing else. Every manual step you repeat is a feature someone has already paid for.

If you’re a business owner, it’s the same idea. Run the job by hand or on tools you already pay for, and let the repeated step tell you what’s worth building.

FAQ

So how much does an AI-built MVP cost? It depends on what you’re building, and tool prices move too often for a price table to stay true. Your own cost map, with your worst month and your time, will tell you more than anyone’s average.

When should I stop prompting and pay a developer? Set your line before you need it. When one bug has cost you a set amount or a set number of hours, a fixed-price developer is cheaper than more credits. Ask what exactly is broken, what fixed will look like, whether it’s a fixed price, and whether everything stays in your name.

Do I need to count my own time if I’m not paying myself? Yes. Those hours could go to talking to buyers. Five hours a week of debugging is five hours a week of not selling.

What should I charge so the AI costs don’t eat my margin? My rule of thumb is 3 to 5 times what a customer costs you to serve. Add their share of your monthly tools, their AI usage and any hours you spend delivering by hand. Times 3 is your minimum price. Times 5 is your target. Then say it out loud to a buyer.

Is it worth building at all if nobody has paid yet? A clickable mock-up to show buyers is fine. Paying monthly to run the real thing before anyone has paid is a guess with a price tag. Let your first customer fund the next part of the build.

Tool features and prices change often. Check each provider’s pricing page before you commit.

The Build Case

Want a second opinion before you spend?

A business case, a build, buy or wait call, a one-page lean MVP scope, a plan to sell it first, and a check of one dev quote. US$1,500 (A$2,300 in Australia), fixed price, 2 weeks. Your fee comes off the 3-month community if you join.

I work with Sazae, a Sydney software and IT firm. If you need a build, I may suggest them, and you're always free to use your own developer.