Staffing automation: timesheets, agreement wages, refunds
Client timesheets, yearly checks of agreement wages against benchmark wages, and placement retention and refunds, automated with AI and human review.
- Runs on your own PCs and servers
- Use Azure OpenAI, Amazon Bedrock, Google Vertex AI, or an in-house model
- Its engine is Descarty's open-source Dagu (GPL-3.0)

Common problems
Hours arrive in each client's own format or system and are gathered again before payroll and billing close
Benchmark wages change every year, so agreement wages are checked again by job and region, even mid-agreement
Agencies ask employers whether placed hires left within six months, then work out refunds and the figures they must publish
Use cases in this industry
Write job postings for each job board
From a client's job request, AI drafts postings in each job board's format. Staff check them, then publish through each board's official route.
Check and track dispatch period limits
Site limits come from the client's notice; per-person limits run from the worker's start in the unit. As either nears, AI drafts a check that staff review.
Total client timesheets and watch overtime caps
Collect dispatched workers' hours from each client's attendance system and find people near the overtime cap. AI drafts the note to the client; staff check it.
Check agreement wages against new benchmarks
Load the benchmark wages the labor ministry sets each year and compare them with agreement wages and each worker's pay by job, region, and experience.
Import hours reported by clients
AI reads hours clients email as Excel files, PDFs, or scans and checks them against contracts. Approved hours go to the dispatch system for payroll and billing.
Track placement retention and fee refunds
Confirm with employers at set points up to six months after hire that placed people stayed. For early leavers, draft the refund and log the six-month count.
Dispatch and placement paperwork grows with every client and employer
As of June 1, 2025, Japan’s worker dispatch reports covered 44,110 offices and about 2.01 million dispatched workers. In fiscal 2024, dispatching offices reported about 860,000 clients (counted per office). Among fee-charging placement agencies, 30,561 offices reported for fiscal 2024, with 889,000 placements into jobs of four months or longer.
Dispatch management systems, the core systems of Japanese staffing agencies, manage applicants, staff, contracts, and timesheets, and produce invoices and payroll from timesheets. More job boards now deliver applicants through official integrations. Applications made with Indeed Apply can be viewed in applicant tracking systems (ATS) integrated with Indeed, and the HRMOS ATS automatically imports applicants from 18 boards, including doda and Mynavi Tenshoku (mid-career edition, October 2023). One staffing agency that started using several boards found that the same person often applied through more than one board, and brought its applicants together with a tool from dip, a job board operator.
What stays manual is where information comes in from outside: from clients, from the Ministry of Health, Labour and Welfare (MHLW), and from employers.
| Work | Screens and systems used | What stays manual |
|---|---|---|
| Applicants | Job boards, ATS | Applications from boards without an official integration, and checking duplicates across boards |
| Contracts | Dispatch management system, web services clients designate | Entering the same contract again in the service a client designates |
| Timesheets | Clients’ attendance systems, Excel files and PDFs sent by email | Gathering hours in each client’s format and typing them into the dispatch management system |
| Agreement wages | MHLW notices and Excel tables, dispatch management system | Comparing agreement wages with the benchmark by job and region every time it is revised |
| Required disclosures | The agency’s own website | Updating the margin rate, the agreement’s end date, and other items every year |
| Placement retention and refunds | Core system, email | Confirming continued employment with employers, and totaling refunds and early leavers |
Hours arrive in each client’s own way
Under Japan’s worker dispatch regulations, the client reports each dispatched worker’s days worked, start and end times, and breaks to the agency at least once a month, on a set date. The report may be on paper, by fax, or by email.
Some products let workers enter their hours in a web timesheet the agency provides, for the client’s staff to approve online. The staffing edition of e-navi Timesheet, for example, is aimed at agencies that want to stop collecting timesheets every month, and offers online approval by client staff and CSV export to payroll software.
Clients may also designate the service to use. The Japan External Trade Organization (JETRO) asks its staffing agencies to use e-staffing: the agency enters contracts on its screens, and dispatched staff enter their hours in Web TimeCard. Signing up requires the agency’s fixed global IP address.
Every new client adds screens and formats for receiving hours. Before payroll and billing close, staff still gather them and type them into the dispatch management system. Collecting hours from clients’ attendance systems to watch overtime caps is covered in “Total client timesheets and watch overtime caps.”
Benchmark wages change every year, and agreement wages are checked again
Agencies set dispatched workers’ treatment either to match the client’s regular employees or through a labor-management agreement. Under the agreement method, agreement wages must be at or above the average wage of ordinary workers doing the same kind of work in the same area (the benchmark wage). As of June 1, 2025, 1.897 million of about 2.01 million dispatched workers were covered by such agreements.
The benchmark wages are set in a notice from MHLW’s Employment Security Bureau, which says they are updated every fiscal year based on the latest statistical surveys. The figures for April 2027 to March 2028 were issued on August 28, 2026; the previous year’s came on August 25, 2025. The tables of base values by job and the regional index are also published as Excel files. The figures move both ways: the benchmark commuting allowance, for example, fell from 79 yen an hour in fiscal 2026 to 67 yen in fiscal 2027.
Even when the benchmark changes partway through an agreement, the agency must check whether agreement wages are at or above the new figures. If they are and wages stay the same, the agency attaches a note confirming this to the agreement. If wages change, the agreement is concluded again.
The margins are thin. In an MHLW review of randomly sampled agreements (fiscal 2025), the average agreement wage for general office, secretarial, and reception work (no experience, lowest hourly rate) was 1,120 yen, on average 19 yen above the benchmark. In the same sample, 88.1% of agreements ran for one year.
MHLW publishes a wage comparison tool for checking that agreement wages are at or above the benchmark, and released the version for fiscal 2027 on October 2, 2026. What remains is mapping the agreement’s jobs and the clients’ regions to the ministry’s tables and checking them against each worker’s job, work location, and hourly pay. An example is “Check agreement wages against new benchmarks.”
The agreement also feeds required disclosures. Agencies publish, on the internet or by other means, their margin rate, whether they have an agreement, which workers it covers, and when it ends. Before a dispatch contract is signed, the client must give the agency information on how its own employees are treated; for contracts limited to workers under an agreement, this is the client’s training and welfare facilities. Without this information, the agency must not conclude the contract. Since October 1, 2026, the terms agencies state when hiring and dispatching a worker must also say that the worker can ask for an explanation of differences in treatment and the reasons for them.
Placement continues after the hire: retention checks and refunds
Private placement agencies earned about 983.5 billion yen in fees in fiscal 2024, about 1.03 million yen per placement into a job of four months or longer.
Under a refund policy, the agency returns all or part of its fee when a placed person leaves early. Fee-charging agencies must state their refund policy to employers and publish it with their fee schedule, by posting it at the office, online, or by other means. MHLW notes that publishing the fee schedule and refund policy on its staffing services portal is still required.
Agencies must also publish online how many people placed in permanent jobs left within six months, not counting dismissals. To confirm that number, the agency must survey the employers. An agency with a refund policy that counts the people who triggered a fee waiver need not survey them. The annual business report is due by April 30.
The more employers an agency places with, the more checks of continued employment up to six months, and refund calculations, are left to staff. An example is “Track placement retention and fee refunds.”
Taking job orders changed too. Since April 2024, the working conditions stated when recruiting must include the possible scope of changes to duties and workplace, and the criteria for renewing fixed-term contracts, including any cap. MHLW asks placement agencies to confirm that employers have given them this information properly. Drafting job postings is covered in “Write job postings for each job board.”
What to check before automating
- Terms of boards and clients: check terms and contracts to see whether job boards’ admin screens and the services clients designate may be operated automatically. Job board examples are in “Check the terms before you automate a website”
- Whether existing tools are enough: where ATS integrations with job boards, agency timesheet products, or MHLW’s wage comparison tool cover the work, use them
- Connection requirements: some services clients designate require registering a fixed global IP address. Choose the computer that runs workflows to fit
- Personal information: Japan’s Employment Security Act requires that job seekers’ personal information be collected and used only as needed for the purpose of the business. Decide first which items go to AI and which AI model is used
- Who decides: agreement wages are decided through labor-management talks. Staff decide refunds and what clients and employers are told. AI reads, compares, and drafts
What Kitewell does
Descarty’s Kitewell is integration and automation middleware that connects systems through APIs, files, databases, and screens.
- Operates screens: AI reads and operates web systems and desktop apps that have no API. Operations that worked are remembered and repeated the same way on later runs
- Keeps going when screens change: when a remembered operation no longer fits the screen, AI reads the screen again, does the step, and remembers the new operation. If it still cannot proceed, the run stops, and AI looks into the cause and suggests the fix
- Starts from email or a schedule: Gmail, Microsoft 365, and IMAP mailboxes are checked at set intervals, and attachments are saved. Workflows can also run at set times, such as weekdays at 8:00
- Reads documents: AI steps read hours sent as Excel files, PDFs, or images, and the ministry’s tables. The AI model is chosen per step, including in-house and local models
- Fits older import screens: Excel files can be converted to Shift_JIS CSV and uploaded to the dispatch management system’s import screen
- Runs from an Excel list: register a list of covered workers or placements as a batch sheet, and it runs row by row and writes each result back to its row
- People check where it matters: approval can be set on any step. Requests reach Slack, Microsoft Teams, and email, and sending one back restarts the run from the chosen step
- Passwords stay out of AI’s hands: passwords are referenced by name and typed in without showing the value to the AI. When a one-time password is required, the run pauses until staff enter it
- Records and alerts: each run keeps its logs, screenshots, and approval decisions. Failures are sent to Slack, Microsoft Teams, email, and more
You can start from one piece of work, such as importing hours from each client through handing them to payroll and billing. A demo shows how it runs, a free connection test confirms Kitewell reaches your systems, and a one-month PoC measures the result against success criteria agreed up front, before you license it for production. See From evaluation to production for details.
Sources
- MHLW: Status of the worker dispatch business as of June 1, 2025 (preliminary) (March 31, 2026; in Japanese)
- MHLW: Worker dispatch business reports for fiscal 2024 (preliminary) (March 31, 2026; in Japanese)
- MHLW: Employment placement business reports for fiscal 2024 (preliminary) (March 31, 2026; in Japanese)
- ASPIC: Staffing management systems (in Japanese)
- Indeed: Indeed Apply (in Japanese)
- HRMOS: HRMOS Recruiting now integrates with YOUTRUST (October 4, 2023; in Japanese)
- dip: Seeing duplicate applicants by managing job boards in one place (May 22, 2023; in Japanese)
- e-Gov: Worker Dispatch Act and its enforcement regulations (in Japanese)
- e-Creation: e-navi Timesheet web attendance system (in Japanese)
- JETRO: e-staffing setup checklist for staffing agencies (May 8, 2023; in Japanese)
- MHLW: Equal pay for equal work for dispatched workers (in Japanese)
- MHLW: Notice on benchmark wages for fiscal 2027 (August 28, 2026; in Japanese) and for fiscal 2026 (August 25, 2025; in Japanese)
- MHLW: Q&A on the revision of the fiscal 2027 benchmark wage notice (August 28, 2026; in Japanese)
- MHLW: Wages and other terms stated in labor-management agreements (sample, fiscal 2025) (January 27, 2026; in Japanese)
- MHLW: Overview of the revised equal pay rules for dispatched workers (in Japanese)
- e-Gov: Employment Security Act and its enforcement regulations (in Japanese)
- MHLW: Leaflet for placement agencies on the April 2024 rules for stating working conditions (in Japanese)
- MHLW: Items to be stated when recruiting, added from April 2024 (in Japanese)
- Indeed: Indeed Terms of Service (updated September 30, 2026; in Japanese)
Frequently asked questions
- May a tool operate client systems or job boards' admin screens?
- It depends on your contracts with clients and each system's or board's terms. Some boards, such as Indeed, prohibit accessing the site with automated systems, including AI, without their express written permission. The PoC checks each one, uses an official integration where one exists, and automates only what is allowed.
- Does AI decide whether pay falls below the benchmark?
- No. The benchmark is calculated from the ministry's tables and compared with agreement wages. AI checks how jobs map to the ministry's categories and summarizes the results. Agreement wages are decided through talks between the employer and the union or worker representative.
- How does this differ from a dispatch management system?
- A dispatch management system manages staff, contracts, and timesheets, and produces invoices and payroll from timesheets. Where it covers the work, use it. Kitewell is for bringing in what arrives from outside it: hours sent by clients, the benchmark wages the ministry publishes, and replies from employers.
Use cases for any industry
Read invoices into the accounting system
AI reads incoming invoices, checks them against purchase orders, and records them in the accounting system. Invoices whose amounts differ wait for approval.
Create invoices and send them by email
Invoices are filled into your template from an Excel billing list and emailed once a person approves. AI first points out amounts far from last month's.
Research a list of prospects
For an Excel list of hundreds of companies, each one's business, locations, and size are collected from its website into its row, with a priority for sales.
Find missing clock-ins
Collect yesterday's clock-ins from the attendance system and, once HR approves, ask employees with missing entries to fix them in Teams.
Take in emailed invoices and draft the entries
Find unread email with an invoice attached, have AI read the amount, and record a draft ledger entry. Uncertain entries wait for accounting to confirm.
Review SaaS accounts
A SaaS account review matches each service's users to current staff. Collect each console's users, match them, and disable leavers only after approval.
Partner registration that starts on approval
A kintone approval starts the workflow by webhook. It looks up the tax registration number, AI matches it to the request, and a match enters the sales system.
Collect KPIs from SaaS tools into a weekly report
Collect figures from advertising, sales, support, and other SaaS tools, and produce a weekly report with week-on-week comparisons and likely causes of change.
Check supplier invoice registration numbers
Check your supplier master's registration numbers monthly through the NTA's Web-API, and list suppliers that were revoked, lapsed, or differ in name.
Prove Kitewell on your own systems
Try it free for 14 days on your own PC. To see it working first, ask for a 30-minute demo. There is also a one-month onboarding (¥500,000 before tax), where we build your first job together.
What the onboarding checks
- Which method connects each target system: API, file, database, or screen
- Whether Kitewell can work a system with no API through its screens, as a person would
- Whether AI relearns a step and processing continues when a screen changes
- Whether errors are handled as agreed: stop, notify, or retry
- Whether it runs in your environment, with the AI model you choose
- Whether processing time and AI usage cost fit daily operation
How the onboarding runs
- Week 1Requirements and success criteria
- Weeks 2 to 3Build
- Week 4Verify on production screens
- Final dayReport
One-month onboarding
- Length
- One month
- Cost
- ¥500,000 (tax excluded). Your staff build, with the developer beside them
- If you go on to a license
- The onboarding fee is credited to the first year's contract
- Without the onboarding
- At Team scale, start on Team monthly (¥14,800 per computer a month, tax excluded), and cancel in any month
You can try it first on your own PC, in the free 14-day trial.