SEISEI SERVICES — AI IMPLEMENTATION
Where should a business start when bringing AI into its operations?
Updated
The short route is to write down the workflow, the decision points and the points where a person confirms the result before adding AI, then build small, go live, and improve while operating. SEISEI INC. (心享勢成株式会社, Toranomon, Tokyo) provides AI implementation that covers this whole path, from requirements to build, go-live and operation, in Japanese, Chinese and English.
Which problems is this for?
- Receiving, checking and re-keying documents takes staff time, and quality depends on who does it
- Work is split across Excel, paper, fax and an old in-house database, so nobody sees the overall status
- Enquiries arrive in several languages (Japanese, Chinese, English and others) and handling depends on individuals
- You have tried AI, but it has not reached the production workflow or your internal checking procedure
What do we deliver?
The work runs in four stages. You review the deliverables at the end of each stage.
| Stage | What we do | Deliverables |
|---|---|---|
| Requirements | Map the workflow, the decision points, what AI does and what people confirm, and which data is handled and where it is stored | Workflow diagram, requirements list, AI-versus-human role table |
| Build | Build the business system and put AI (text reading, classification, drafting and similar) inside the workflow | Working system, tests, operating guide |
| Go-live | Prepare production, fix the backup and rollback procedure, then switch over | Production environment, backups, switch-over and rollback procedure |
| Operation and improvement | Improve from staff feedback and records, and update for changes in law and business | Changes, operation log, periodic review |
Which principles do we build by?
AI proposes, a person confirms
AI readings and drafts are shown as candidates. Staff check them before anything enters the record. Judgements and conclusions are not left to AI.
Decide where data lives first
For personal data or accounting records, we fix the storage location, the AI service and its region, and how consent is obtained during requirements.
Everything leaves a record
Who confirmed or sent what, and when, is kept, so the history can be traced later.
Ship small, then improve
We do not build everything at once. The workflow with the largest effect goes live first, and the scope widens during operation.
Multilingual
Screens and notices can be provided in Japanese, Chinese, English and other languages. Meetings can be held in Chinese.
How does an engagement run?
- First conversation: we hear about your current work and what is not working.
- Workflow review: you show us the documents, screens and steps, and we draft the scope for AI.
- Proposal and quote: we set out scope, stages and team, and quote individually.
- Build and go-live: you review at each stage.
- Operation and improvement: after go-live we continue changes and periodic reviews, if you wish.
Are there case studies?
We publish the systems we built, to the extent they can be made public. Only numbers that can be checked are stated.
- Residence application portal: document upload, AI document reading, draft renewal reasons, fee calculator
- Report Desk: from reading receipts to trial balance, BS and PL output
- A Tokyo inbound tour operator: arrangements, hotel notices, fax sending and billing in one system
When is this not the right service?
If configuring an off-the-shelf SaaS is enough, or if you want to pick a tool before the workflow is settled, our support is more than you need. Neither SEISEI nor the systems we build compute tax amounts, prepare tax returns or answer questions on tax judgements.
Frequently asked questions
What does it cost?
It is quoted individually. Cost varies widely with the scope of work, the data handled and the stages required (requirements only, or through to operation), so there is no fixed price list.
Which AI do you use?
We choose by the work and the nature of the data. The published cases use Anthropic Claude, Google Gemini and Qwen on Alibaba Cloud (Tokyo region).
What if the AI makes a mistake?
AI output is shown as a candidate and confirmed by staff, so errors can be corrected before anything is final.
Can we work in Chinese?
Yes. Meetings, documents and screens are available in Japanese, Chinese and English.
Can this be combined with financial architecture work?
Yes. Having the same person look at back-office systems and at the structure of the company and its assets is what distinguishes SEISEI.
Sources you can check
Related pages
Get in touch
Tell us about your current workflow and the tasks that are not working. Write to ceo@seisei.tokyo.