AI model development services
「Credentials set the price」
The floor is high. So is the door — degrees and job history are doing the trust work here, and buyers read them.
Select and implement a model for a client's problem. The highest price band on this list, and the highest barrier — what sells is a research and engineering track record, not tool operation.
- Difficulty
- 5/5
- To first revenue
- A single engagement runs about two weeks. How long until you win one is not something we estimate.
- Startup cost
- ₩100,000~1,000,000
- Coding
- a lot
- Sales
- some
- Automation
- 2/5
We have zero cases with a verified earnings figure, so we do not invent a revenue range. What follows instead is arithmetic on published prices and official fee rates.
Where the money goes
One sale, start to finish — Single package
This is also the figure you report as revenue — not the amount that reaches your account.
- Service fee (16.4% / 9.4% / 4.4% 누진 구간)−₩124,200
- Payment network fee (3.3%)−₩26,400
- VAT on those fees (위 두 항목의 10%)−₩15,060
Output VAT ₩72,727 − input VAT on fees ₩15,060−₩57,667
⚠️ Tool costs, your own labour and income tax have not been taken out yet. Individual (non-business) accounts handle VAT differently and cannot issue tax invoices at all.
The ceiling you cannot pass in a month
Published delivery days divided into the working month. Not what you can earn — the line you cannot cross alone.
| Tier | List price | Days | Max/mo | Revenue ceiling | Operating profit |
|---|---|---|---|---|---|
| Single package | ₩800,000 | 14d | 1 | ₩800,000 | ₩576,673₩28,834/day |
↔ Swipe sideways
What these numbers are, and are not
- Nobody earned this. Not one listed case has a verified revenue figure.
- List prices and delivery days come off the seller's page; fee rates come from the platform's own documentation.
- “Max/mo” is division — 20 working days divided by the published delivery time. It does not mean orders arrive that steadily. In practice they do not.
- Operating profit here is before tool costs, your own labour and income tax. Rework and refunds are not in it either.
Who you sell to
- Companies needing an AI capability
- Research and prototype teams
How the money comes in
- Project engagements
- Surcharges by data type
Steps
- Write down the data types you have actually worked with — this route starts from experience
- Frame the portfolio as problem → chosen method → validation, not as output
- Fix scope and handover items (source, accounts, access) in the contract
- Buyers are companies, so get the invoicing requirements in place first
Tools used
Claude API · Cursor · GitHub Copilot
Linked tools are ones whose pricing we checked and listed.
Risks
- Lowest replicability here. Degrees and track record are effectively the product
- Engagements are long, which makes running this alongside a job hard
- Preparing training data eats most of the schedule — price it in or you lose money
Verified cases using this model
BMostly confirmed81 transactions prove demand — but this one is not a solo side business
StellaDev — AI model selection, design, implementation and validation · KR · Revenue not disclosed
Compare with
- Level 2Website production from one template「Build once, sell repeatedly」
- Level 3AI automation agency「Taking the repetition nobody wants」
- Level 2AI chatbot and GPT builds「Selling the setup, not the AI」
Do this today
- Write down the data types you have actually worked with
- List only the services in this category with a three-day turnaround or less
- In one sentence, write what separates the expensive listings from the cheap ones