What it really costs to build a small business app with AI

The monthly plan is the small part. Your hours and the upkeep after launch are the big part. The formula is your hours × the value of your hour, plus plan fees, plus upkeep.

Road map ROI is made by Cookie Tech, which builds apps for clients with these tools. Weigh what we say with that in mind. We name no price for any builder, and we rank none.

Builder prices: checked October 2026. Credits and usage pricing change, so check the date.

To put numbers in, use the build, buy or rent calculator. For where each of its numbers comes from, see how to estimate software costs.

What the plans cost

We do not print a price table. Builders change plan names, credits and limits often, and a number here would be wrong within a quarter. Open the builder's own pricing page and read three things: the monthly price of the plan you would need after launch, what a credit or usage unit buys, and whether hosting and AI features inside your app draw on the same allowance. Write down the date you looked.

Source type: pricing page

Put the sum of your builder, AI assistant and hosting plans into the calculator as a monthly figure.

Credits and usage run out

Many plans give a fixed allowance. A long build with many fixes can use it up before the month ends. Find out on the pricing page what happens then: you wait, you buy more, or you move up a plan. Use a busy month, not an average one, as your figure.

Source type: pricing page

For how to enter this, see AI tool and builder credits.

Your own hours are the main cost

Write your honest guess. The calculator counts double that, and four times it in the worst case. Both multiples are our assumptions, not measured rules. First guesses leave out testing and fixing. Flyvbjerg and Budzier studied 1,471 IT projects: the average cost overrun was 27%, and about one in six overran by 200% on average (paper). The reasoning is on hours to build with AI.

Source type: estimate

Two real builds

Cookie Tech's own time tracker recorded these hours, building with Lovable and Claude. Both are for small businesses. We name no client.

• A production-planning app for a small food business: about 47 measured hours over about 10 weeks, in 2026.
• A business-management app for a small business: about 45 measured hours over about 5 weeks, in 2026.

Read these limits before you use them:

• The builder is experienced and does this for a living. A first-timer should expect more hours.
• The hours cover building, fixing and changes. Meetings are not in them.
• Two cases are not a statistic. Do not average them into a rule.

Source type: own records

What one study found

METR ran a randomized study of 16 developers working on mature open-source projects they had worked on for about 5 years on average. With AI tools, they took about 19% longer to finish tasks (arXiv 2507.09089). It is a preprint about early-2025 tools. It did not test a small business app, and it did not test people new to coding. Read it as a warning against assuming AI makes every job faster, not as a measure of your app.

Source type: published preprint

What your hour is worth

Use what you charge a client for an hour, or your monthly income divided by the hours you work. An hour spent building is an hour not spent selling. The calculator prices your hours at the rate you enter. The three methods are on what your own hour is worth.

Source type: own records

What AI-built apps still need

Ask these before launch, and price the answers:

• Where does it run, and what does that cost each month?
• Who checks the logins and who can see which data?
• Where are the backups, and has anyone tested restoring one?
• Who fixes it when it breaks on a busy day?

The reason to ask: Veracode gave more than 100 AI models 80 coding tasks, and the code had a known security flaw in 45% of cases (Veracode, 2025; Veracode sells security scanning). An AI coding assistant once deleted a live production database (OECD AI Incidents Monitor). That is one documented case, not a rate. The calculator does not price this work.

Source type: vendor study and one documented incident

Upkeep is your time after launch: fixing what breaks, small changes, questions from your team. The calculator starts at 5 hours a month and uses 10 in the worst case. Both are our assumptions. Count more if customers use the app. See yearly upkeep.

A worked example

Example (invented numbers): You guess 40 hours and your hour is worth $30. The AI tools cost $45 a month. You use the app for 3 years, the calculator's default.

• Build hours, likely: 40 × 2 = 80 hours × $30 = $2,400.
• Build hours, worst case: 40 × 4 = 160 hours × $30 = $4,800.
• Tools: $45 × 36 months = $1,620.
• Upkeep, likely: 5 hours × $30 × 36 months = $5,400.
• Upkeep, worst case: 10 hours × $30 × 36 months = $10,800.

Likely total: $2,400 + $1,620 + $5,400 = $9,420. Worst case: $4,800 + $1,620 + $10,800 = $17,220. The build is the smaller share. Upkeep hours are the larger one.

Compare with the real cases: both landed near 45 hours for an experienced builder. That is above a 40-hour guess and below the 80 hours we count. Two cases do not prove the doubling is right or wrong.

When paying someone is cheaper

Compare your likely and worst totals with the middle of a written quote range plus its yearly upkeep. If your worst case is more than the quote, the quote is the safer choice. If your likely case is far below it, building may pay, but only if you have the hours. Read what a developer charges for how to get a quote, and custom vs off-the-shelf software for the per-person comparison.

Sources

• Cookie Tech internal time tracker, 2026: our own measured hours on two builds. Two cases, no client named.
• METR, arXiv 2507.09089: randomized study, 16 developers on mature projects they knew well, about 19% longer with AI tools. A preprint.
• Flyvbjerg and Budzier (arXiv): published paper, 1,471 IT projects.
• Veracode, 30 July 2025: vendor study, 80 coding tasks.
• OECD AI Incidents Monitor: one documented incident.
• Each builder's own pricing page, checked October 2026. We print no builder prices.

Anything not listed here is our assumption or plain advice, and is labeled that way. This is not financial, tax or legal advice.