Build with AI. Earn. Share. Find the next project.
Deliver useful landing pages, websites and small automations. Find projects on CrowdWorks, package services on Coconala, and earn repeat work through testing and clear instructions.
Checked September 23, 2026 · Freelancing notes based on public projects
Studio Toriumi's public projects provide a basis for testing web and automation services. That is an assessment of the work on display. The bilingual sites, information collection and scheduled updates, and browser tools in the portfolio show what can be built. Personal projects and paid client work are different kinds of evidence: this page does not claim completed commissions or achieved earnings.
FIRST / WEB
Landing page fixes and web improvements
Evidence: This portal's Japanese and English pages, plus browser interfaces such as the communication template tool.
First service: Fix mobile layout, spacing, buttons or copy on one existing page. Agree on the specific changes and supported browsers first.
Follow-up work: Improve another page, replace campaign content or handle monthly updates.
NEXT / AUTOMATION
Automate one recurring task
Evidence: Public information collection and scheduled update projects such as BCI WATCH.
First service: A small workflow with defined inputs and outputs: clean a CSV, aggregate a public API or produce a recurring report. Build a personal demo first for any untested environment, including Google Apps Script.
Follow-up work: Add data fields, failure alerts or operational support.
First service: A prototype or feature with a limited number of screens and data fields. Quote discovery separately from implementation when authentication, payments or integrations expand the scope.
Follow-up work: Production implementation after validation, maintenance and handover.
Clients buy a usable result, not just access to AI. Demonstrate experience through public projects. Do not reuse employer documents, private career information, or client code and data in sales material.
02
Find projects on CrowdWorks. Package services on Coconala.
Options for service providers. Fees checked September 23, 2026.
Sort listings by newest first. Search for landing page fixes, website updates, Python automation or manuals within work you can verify.
20% on the first ¥100,000, 10% on the portion over ¥100,000 up to ¥200,000, and 5% above ¥200,000. Consumption tax is added to the fee. Official fees ↗
Compare web and workflow improvement projects. Include communication, setup support and maintenance when judging the scope.
The freelancer fee is 16.5%, including tax. A lower fee alone does not make work easier to win. Official FAQ ↗
Own site and referrals
New enquiries through your own publishing or referrals. Build evidence of your work, clear explanations and trusted recommendations.
Even with no marketplace fee, marketing, contracts, invoicing and payments take time and may cost money. Do not move clients met on a platform outside it.
A 20-minute daily routine for new listings
10 minutes: read new listings and shortlist up to three good matches.
Check the budget, available materials, deadline, client reviews and payment status, AI permissions, and what repeat work would actually involve. Being first to apply does not guarantee an advantage.
10 minutes: write a specific proposal for the best match.
Briefly show one relevant personal project, the first questions to resolve, the deliverables, draft date and acceptance process. Avoid sending the same generic proposal in bulk.
Before starting: confirm the scope and funded payment.
Do not produce a complete deliverable as an unpaid test. Use a small paid trial with defined scope and acceptance criteria. Treat possible repeat work as a possibility, not a guaranteed order.
Keep communication and portfolios within each platform's rules. On Coconala, use the built-in portfolio rather than directing clients to external contact details or your own site with an enquiry form. CrowdWorks approval for necessary external communication does not permit direct contracts or external payments. Check the official rules.
Public listings show why demand and profitability differ
These few examples are not market-rate statistics or evidence of our earnings. The ¥44,000 landing page listing was closed when checked; listing status can change.
Landing page coding: ¥44,000
A listing published September 7, 2026 requested static HTML and mobile support, offered ¥44,000 including tax, set a one-week deadline and mentioned repeat work. The estimated amount after platform fees alone is ¥34,320. Profitability depends on supplied copy and design, revisions and total hours.
Landing page design: ¥3,000
A listing published September 10, 2026 offered ¥300 for a selection test plus ¥2,700 for production. Even the full ¥3,000 leaves an estimated ¥2,340 after platform fees. Over two total hours, that is ¥1,170 per hour before expenses. “Repeat work available” is not enough to make a project worthwhile.
For automation, a Power Automate Desktop project for PDF processing listed a ¥50,000–¥100,000 budget and requested setup support and ongoing improvements. An advertised budget is not profit. Listing status and displayed details can change.
03
Calculate earnings from fees and total hours
A ¥50,000 project is not automatically profitable because you use AI. On CrowdWorks, an estimated ¥39,000 after platform fees leaves ¥36,000 after ¥3,000 in expenses. If sales, meetings, revisions and testing take 20 hours, that is ¥1,800 per hour. At 10 hours, it is ¥3,600.
Estimated platform fee
¥11,000
Amount after platform fee
¥39,000
After expenses, before income taxes
¥36,000
Effective hourly earnings, before income taxes
¥1,800
Calculations run on your device. Inputs are neither stored nor transmitted. The CrowdWorks estimate assumes a fixed-price contract and uses tiered fees plus 10% consumption tax on the fee; standard Coconala services use 22%, and Lancers uses 16.5%. The estimated total fee is rounded to the nearest yen. Withholding, income taxes, bank transfer fees and payment processing fees are excluded. Actual amounts may differ by contract type and rounding; confirm them on the platform's quote screen. Allocate expenses such as APIs, paid AI tools, hosting and assets to the project.
Look at what remains, not just sales. These hypothetical examples do not guarantee projects or earnings.
Scenario
Sales
After expenses, before income taxes
Total hours → hourly earnings
No projects won in the first month
¥0
A loss if you incur expenses
Sales time is also unpaid
One ¥50,000 CrowdWorks project
¥50,000
¥36,000 (¥3,000 expenses)
20 hours → ¥1,800
Two ¥50,000 CW projects + three ¥10,000 maintenance contracts
¥130,000
¥95,400 (22% on each contract; ¥6,000 expenses)
34 hours → about ¥2,806
Set your minimum price first
Where the platform fee is 22%, a useful price floor is (target hourly earnings × total hours + expenses) ÷ 0.78. A ¥3,000 hourly target, 10 total hours and ¥3,000 in expenses give approximately ¥42,308. If the scope grows, adjust the price, scope or deadline.
Small offers to test: A limited fix to an existing landing page at ¥10,000–¥30,000, a paid automation prototype at ¥30,000–¥50,000, or maintenance at ¥10,000–¥30,000 per month. These are possible starting offers to test for profitability, not market rates or confirmed prices. Define maintenance tasks and an hours cap. A complete new landing page or production system belongs in a separate quote using the standard project pricing.
04
Share an 80% draft early. Deliver after testing.
“80%” describes an early draft that makes the direction clear. It does not mean delivering with agreed requirements missing. For a fast turnaround, narrow the scope and schedule the review before starting.
Before work: agree on a one-page specification.
Document the purpose, target pages, supplied materials, inputs and outputs, supported environment, deliverables, revisions, acceptance criteria, AI permissions and extra charges. Start production once payment is funded and the required materials are ready.
Early review: for a small fix, aim for a first draft as soon as the next business day.
Agree on timing based on availability and materials. Label the preview “for review, not yet accepted” and show the main screens and behavior. Do not run real payments or use live data in a prototype.
Testing: check normal use, errors and mobile screens.
Check broken links, empty or invalid inputs, form submission, duplicate actions, permissions, API failures and narrow screens. For automation, check failure handling, stopping and rerunning, duplicates and backups. Record why any test does not apply.
Handover: deliver the work, test results and instructions.
Include source files, setup steps, a short walkthrough video or illustrated guide, and known limitations. Publish to production after client acceptance. Keep secret values out of documentation and transfer them through an agreed secure method.
Repeat work: separate defect fixes from additions.
Agree on a period for fixes within the original scope. Quote new features, updates and maintenance separately. Define contact hours, a monthly work limit and incident support coverage.
A minimum handover guide
What the tool does, and its limits
How to open it, enter data, run it and check results
What errors mean, and how to stop or undo an operation
Where to update it, running costs and who manages access
Tested environment, test date and known limitations
A proposal template
To address (the client's problem), I propose improving (defined scope).
A relevant example is (an approved client example or a personal project).
I expect to share a draft on (date after receiving materials) and deliver on (date including testing).
The handover includes (deliverables), operating instructions and test results, with (number and scope) of revisions.
Please confirm whether AI assistance is permitted and clarify (key question).
05
Deliver, learn and share what you build
Build a collection of knowledge and components you are allowed to reuse. Feed each project's problem, fix and verification back into your skills notes, checklist and estimating process. Learn from the structure, effort and mistakes in past examples without copying client deliverables or paid templates without permission.
BUILD
Use a consistent case study format
The problem, what you built, what you tested, total time and the next improvement. Label personal projects clearly and never invent results or sales.
SHARE
Publish a short project note
With the client's permission, explain improvements through screenshots, an article or a short video. Without permission, create your own demo with fictional data. Let your published work attract new enquiries.
REPEAT
Ask what still needs doing
At handover, ask about tasks that recur each month. Suggest maintenance only for useful work such as content updates, reporting or error checks. Avoid unnecessary monthly subscriptions.
Think of “free AI side hustles” as a starting point for learning and prototypes
You can begin with an existing computer, editor and free tiers to build samples and test demand. There is no need to buy courses or multiple AI subscriptions first. Production APIs, domains, hosting, assets and maintenance may still cost money. Check free-tier limits and commercial-use terms, then agree whether ongoing costs are included in your price or paid by the client.
AI can help with drafts, starter code, test ideas and manual outlines. A person remains responsible for requirements, rights, confidentiality, behavior and delivery. Before entering client data into AI, confirm permission and the environment being used. The CrowdWorks AI policy also calls for advance communication and meeting the agreed quality standard.
06
Run a small, 30-day experiment
Week 1: two offers and two samples.
Focus on a limited landing page fix and a small automation. Produce a walkthrough, test results and a one-page guide, then define each offer and its minimum price.
Week 2: review new listings and apply to suitable projects.
Aim for five individually considered proposals in total. Prepare a Coconala service with clear quoting conditions. Do not send the template above unchanged in bulk.
Week 3: if you win a paid trial, record the work through testing.
Track draft turnaround, revisions, all working time and the amount received. If no project is won, review the proposal, samples and target projects.
Week 4: use the numbers to choose what to continue.
Use around 15 well-matched proposals in total as a practical cap for the experiment. Compare response rate, contracts, effective hourly earnings and repeat enquiries. If there is no response, change the offer and target client rather than relying only on lower prices.
The first goal is one profitable project you can repeat, before a monthly income target. Use existing project experience to test sales and delivery on a small scale, then focus on the work whose numbers make sense.
Checked September 23, 2026. These are reference links, not affiliate links. Calculations, service ideas and workflows are this site's suggestions. Listings, fees and terms can change; check again before applying or contracting. Platform sources are in Japanese.