You Can Build a Cost Ledger for a Small AI Service

Costs start after you build it for free

You can use free tools, with help from AI, to make the first screen of a small web service. When you combine several tools from different providers, website hosting, data storage, and AI usage can each be billed by a different company.

Café24 AI Space (카페24 AI Space) starts at KRW 4,900 per month based on its public pricing as of October 2026. Its database storage is not allocated separately; it shares storage with app files, logs, and container storage.

Combining overseas tools can raise the starting cost. Vercel Pro, a website hosting service, starts at USD 20 per month, while Supabase Pro, a service for storing customer records, starts at USD 25 per month. Together, that is about KRW 61,000. Continue using a USD 20-per-month creation tool, and the total becomes about KRW 88,000. That does not yet include AI call charges, international payment fees, or value-added tax.

Look at costs in three parts: monthly fixed costs, variable costs per customer, and taxes and payment structures. Creation tools and baseline operating fees are paid even with no customers. Usage charges rise whenever customers summarize documents or generate images. For some services, the second cost rises even when the additional users are free users.

Assume a service has KRW 80,000 in monthly fixed costs. If a customer pays KRW 11,000 including value-added tax, payment costs are KRW 363, and AI usage costs KRW 1,500 per customer, about KRW 8,137 remains to cover fixed costs. You need ten paying customers just to barely cover the tool costs. Add the maker's labor and refund costs, and break-even moves further away.

How one operator's ledger changes

Consider a non-technical operator who built a service that summarizes recordings of after-hours consultations at private academies. An academy director uploads a recording, receives a consultation summary and a suggested date for the next contact, and the operator charges KRW 11,000 per month.

Today, the operator checks a creation-tool subscription in a card statement, checks this month's usage in the website hosting dashboard, and separately finds AI call charges in an AI provider's dashboard. They count customers again in the payment dashboard, then manually copy numbers from four places once a month.

The problem is not the total cost but the broken connection between the numbers. The operator cannot immediately tell which academy uploads enough recordings to create a loss, how much was spent on free-trial customers, or whether storage will exceed its limit next month. They discover the price was too low only after the bill rises.

Add a cost ledger, and the first screen can show this month's fixed costs, average usage cost per customer, current number of paying customers, and the expected date of the next pricing-tier upgrade. Place each customer's processed recording volume beside the fee they paid, and the operator can find customers who are creating losses.

Pricing no longer has to be set by instinct. Compare what remains when KRW 11,000 includes 300 minutes of recordings with what remains when a KRW 7,700 base fee includes 100 minutes. Preview the outcome with ten, thirty, or one hundred customers. You can also alert both the operator and the customer before usage reaches 80% of its limit.

Some work still stays with people. Ask how much time customers are actually saving. Explain a price change to customers who upload overly long recordings. Give customers confidence that their records are stored safely. A calculator cannot decide the right price for you.

Anyone can rebuild a similar calculation screen. What is harder to copy is the accumulated record of bills Korean operators have actually paid and customer-level usage, trust that cost alerts are accurate, and historical records that continue even when payment or hosting providers change. That data and trust can last longer than the technology.

How services elsewhere connect costs and pricing

CoverLetterGPT in Germany

CoverLetterGPT creates cover letters from résumés and job listings. Its operator, Vince Canger, hosted the web service and customer records on Railway and connected OpenAI. He disclosed direct monthly operating costs of roughly KRW 22,000. Payment fees were higher, at roughly KRW 63,000.

The operator sold subscriptions and generation credits, and disclosed roughly KRW 770,000 in monthly recurring revenue and roughly KRW 580,000 in average monthly profit. He said maintenance took about one hour per month, but this is self-reported and needs separate verification. The case shows that for a service producing short text, payment costs and sales channels can matter more than AI usage charges.

Audionotes in India

Audionotes transcribes and summarizes voice notes and meetings. The operating team started by connecting speech-to-text and text-generation features to Bubble, a visual app-building tool, and sold monthly subscriptions to individuals and knowledge workers.

For voice services, costs rise with usage time. The operator publicly discussed how unlimited lifetime access could create losses, and tried to address the cost through subscriptions and usage limits. Its official introduction states that it has more than 200,000 users and more than 1 million generated notes, but these are self-reported figures.

Weckr in Sweden

Weckr shows how much an AI-feature service earns or loses for each customer. It records usage costs by customer and feature, matches them with payment revenue, and finds customers whose costs exceed their subscription fees.

Its free tier tracks up to 50,000 events per month, and its paid tier up to 1 million. It offers spending caps and unusual-usage alerts. The publicly reported software download count is more than 3,800, but its revenue and number of paying customers are not known. It can be viewed as a move from simply viewing costs to preventing losses in advance.

Things you can build now

1. A ledger that gathers a month's costs

• What it does: Upload card statements or bills, and it categorizes monthly spending into creation costs, website hosting, customer-record storage, and AI usage.
• Who uses it: A solo operator who built an initial paid web service with AI and receives bills from three or more providers.
• Why now: The point at which free allowances run out differs by tool, so it is easy to notice costs later than revenue.
• First screen: Show this month's fixed costs, usage-driven costs, next month's forecast, and creation tools that can be cancelled.

The defensibility of this service comes not from the calculation screen but from real bill-classification data. If operators using the same tools accumulate records of the point at which they raised prices, the service can give more accurate alerts.

2. A profit-and-loss alert by customer count

• What it does: Enter a fee and cost per customer, and it shows how many customers are needed before money remains and which customers create a loss.
• Who uses it: A non-technical operator planning to sell KRW 10,000-to-KRW 50,000 monthly services for document summaries, consultation notes, or promotional copy.
• Why now: Even a low monthly fee can leave less than expected after payment costs and AI usage charges.
• First screen: Show two adjustable bars for price and target customer count, the break-even customer count, and the result after adding the operator's labor cost.

To be trustworthy in Korea, it needs to reflect domestic value-added tax and actual payment contracts. If data on refunds and customer-support time by industry also accumulates, it becomes harder to copy than a simple calculator.

3. A usage-pricing checklist for voice and image services

• What it does: Sets a base fee, included usage, and overage fees based on recording minutes or the number of images generated.
• Who uses it: A solo maker trying to price a meeting-recording summary or product-photo generation service before selling it.
• Why now: Customer usage varies widely in these services, so unlimited pricing can lead to losses.
• First screen: Show monthly cost for a typical customer and a high-usage customer, recommended included usage, and a cost-cap alert.

The lasting asset here is the real usage distribution. You need to know how many minutes or images most customers use before proposing an included allowance that is not excessive.

4. A migration guide for leaving free tools

• What it does: Guides services nearing free storage or operating limits through the order of backup, paid conversion, and migration to another provider.
• Who uses it: Local businesses and service agencies that built booking, membership, or application services with free tools and are beginning to accumulate real customer records.
• Why now: Changing a price plan may be easy, but safely moving customer records and login information can still be difficult.
• First screen: Select the current tool, storage used, and monthly visitors to see the remaining free period and a migration task list.

This service will not last by collecting only the latest documentation. Its advantage comes from records of errors during real migrations, relationships with domestic production agencies, and the trust that someone will answer when problems arise.

Check this in 30 minutes today

Ask one person who runs a small web service to look at last month's billing screens with you. If they receive bills from three or more providers and cannot answer, within ten minutes, how much one additional customer costs or how many customers they need to break even, a cost ledger or profit-and-loss alert is worth building first.

Why this matters where you are

Check whether a small service in your market receives bills from separate providers for hosting, storage, AI usage, and payments. Tax and payment structures will differ from the Korean examples, but the need to connect customer revenue with customer-level usage remains. Start by collecting real billing records and identifying the usage that creates losses before setting an unlimited price.