One process in production, measured in hours saved.
An AI implementation partner for Vietnamese and Japanese SMEs. We pick one job your staff do by hand today, such as keying in orders, sorting email, or translating reports, and have AI do it inside the tools you already use. A person signs off wherever a mistake is not allowed, and we measure before and after with the same ruler.
iPlus is hiring
iPlus helps companies in Vietnam and Japan hire for a wide range of AI roles, with competitive pay and benefits.
Quick assessment of one workflow
Describe one task your staff still do by hand. We reply in writing: which part AI can take, where a person still needs to approve, and how many hours it saves.
Within 2 business days
No obligation. The result is yours to keep.
You send
- What the manual task is and who does it
- The tools in use: Excel, kintone, Google Workspace, Microsoft 365…
- Roughly how many hours a week it takes
You receive
- A one-page assessment in English, Japanese or Vietnamese
- What AI can do and where a person still approves
- Estimated hours saved and the data risks
- How it would run inside the tools you already use
Next step, if it fits: The AI Workflow Assessment (2 weeks).
Choose the description closest to your company. We will point to the right first step.
Not every company should start in the same place. Sometimes the right answer is “not yet”, and we will say so.
AI Workflow Assessment
This is the right starting point for most companies. In two weeks and two working sessions you receive a ranked list of the 3–5 jobs most worth handing to AI, with hours, ROI and data risk for each.
- Know which job is worth doing first and which is not
- Real hours, not gut feeling
- A roadmap you can follow on your own if you wish
Three steps. Each one leaves you with something in hand.
Each step is a fixed package, never billed by the hour. The durations below are typical, counted from when you grant access to the data and name a contact person; the exact schedule is set in the contract. The result of one step is your basis for deciding whether to take the next.
AI Workflow Assessment
The most important step, because it decides whether spending on AI makes sense at all.
- 2 working sessions with the people who do the work, online or at your office
- The process redrawn, with real hours counted instead of estimated
- A 5–8 page report in Vietnamese, Japanese or English: the 3–5 most time-consuming processes, hours recoverable, ROI, roadmap, data risk
- One presentation of the findings to your management
A clear conclusion — even when the conclusion is “this should not go to AI yet”.
Workflow Automation
Or an internal-knowledge assistant (RAG): it answers from your own documents, with citations.
- One process running daily on real data, inside your kintone, Microsoft 365, Google Workspace, Slack or Excel
- Stop points for human review at the steps that must not be wrong
- A log of every step: what the system read, what it decided, who approved
- Before-and-after figures, measured the same way as in the assessment
A process that has actually run, with numbers — not a demo on sample data.
AI Advisory (monthly)
AI is not install-once-and-done. Models, data and processes all change.
- Monitoring of deployed AI workflows, with fixes when they start to drift
- One improvement every month
- A monthly “hours saved” report, with the errors caught
- Q&A, and model updates when a better model ships
Keeping what already runs running correctly, and adding one more job each month.
AI Usage Policy1–2 weeksstandalone or with step 2
An AI usage policy and data risk assessment in Japanese, built on the ISMS / JAMA framework we drafted for a Japanese company. For firms that want staff to use AI without the risk of a client contract being pasted into a free chatbot.
Five jobs every company has, done by hand by someone.
Each scenario states what AI does, what a person must check, and how the result is measured.
Two office staff key 40–80 orders a day by hand from fax into kintone.
A 1–2% error rate sounds small, until an order with the wrong part number ships.
Claude Code workshop for a 30-person software company
Not a process but a way of working: 2 days on the company's own codebase (CLAUDE.md, skills, regression tests, review), an AI usage policy for writing code, then 4 weeks of support over Slack. For SES and software firms whose clients are asking about hands-on AI experience.
AI is not 100% right. So we design the stop points.
The two biggest fears about AI are wrong answers and leaked data. We do not promise that neither will ever happen. We design the system so that when one does, a person sees it before your customer does.
Every AI result carries a confidence score. You decide below which level a person must look.
High threshold: safer, more for reviewers to do. Low threshold: faster, riskier. We set the threshold with you for each kind of job. Large-value orders and emails to customers always pass through a person, however confident the AI is.
Before we startcontract
An NDA is signed before you share any data, or before the first conversation if you prefer. Which data may be given to AI and which may not is agreed and written into the contract.
During designarchitecture
Stop points for human review are drawn from the start. Data stays in systems you control: AI can run inside your Microsoft 365, Google Workspace or cloud. We use only business APIs, where submitted data is not used to train models.
In operationmonitoring
A log of every step, role-based access, and a monthly report that lists the mistakes AI made and the ones that were caught.
We have drafted a security rulebook on the ISMS and JAMA frameworks, including L3-level procedures and 14 registers, for a Japanese company. The same experience goes into your AI usage policy, written in Japanese, in a form Japanese legal and compliance teams can read.
Ask about securityBuilt on the tools you already have. Nobody has to switch software.
The technical scope is fixed in advance, so no project starts from zero. That is how your first process can be running after 4–6 weeks.
Jobs with a ready frame
- Reading and sorting email and orders
- Extracting PDFs (invoices, quotes, purchase orders) into kintone or Excel
- RAG over internal rules and manuals
- Drafting quotes and shift reports automatically
- CV screening
- Translating and summarising Japanese, Vietnamese and English documents
Saying “no” early is cheaper than “sorry” late.
Not because we cannot. These jobs rarely bring value to an SME, and we do not want you paying for them.
Building our own model, fine-tuning
Existing models are good enough for almost every office job. Spend on the process, not on GPUs.
Projects over 3 months without AI Advisory alongside
An AI system nobody watches drifts. We do not hand over something we know will break.
Systems that need someone on-site all the time
We work remotely, and meet in person when a working session or handover needs it.
Promising “AI replaces the department”
AI gives hours back to your people. Whoever promises more than that should be asked who is accountable when it is wrong.
What managing directors ask in the first 5 minutes.
Your question is not here? Ask it directly in the 30-minute call.
Will our data be used to train AI, or leak?
It is not used for training. We use only business APIs, where submitted data is not used to train models, and AI can run inside your Microsoft 365, Google Workspace or cloud. An NDA is signed before you share any data, and the scope of data given to AI is written into the contract. We do not promise zero risk; we reduce it with access control, logs and the smallest data scope that works.
What if the AI answers wrongly?
It sometimes will, which is why every process has stop points. Results below the confidence threshold, or in the kinds of work that touch customers and money, always wait for a person. Every step is logged so it can be traced, and the monthly report lists the errors that were caught.
How do you check the quality of what you build?
Every code change is reviewed by a senior engineer other than its author before it is merged. Automated tests and static analysis run on every change to catch regressions. Before handover, the workflow runs on your real data and is signed off against acceptance criteria agreed at the start, including the before-and-after measurements.
Who owns the code and workflows you build?
Your company does, from day one. An IP assignment clause is signed together with the NDA before we write the first line of code. Source code, agent configuration, prompts and documentation are delivered to your own accounts and repositories. We never hold code hostage: your IT team or another vendor can take over at any time.
Roughly how much does it cost?
We quote after the 30-minute call, because scope differs from company to company. Prices are fixed per package, never by the hour. The diagnosis step is designed so a managing director can approve it alone. Model API costs are stated in the operations contract.
We have no IT team. Can we still do this?
Yes. Your users only see results in tools they already know: a record in kintone, a draft in Outlook, a message in Slack. We do not ask you to change software. The operation behind it is our job under AI Advisory.
What language do you work in?
Vietnamese, Japanese or English, your choice. The assessment report and handover documents are written in the language you choose; the AI handles all three, including the Vietnamese–Japanese pair. For companies in Japan: Hanoi (GMT+7) is only two hours behind Tokyo, so meetings and day-to-day communication happen within your working hours.
What if we want to stop halfway?
After each step you may stop and keep what you received — the report, the process, the logs. AI Advisory has a 3-month minimum, month-to-month after that; when you want your own IT team to take over, we have a handover package.
Can we get a subsidy such as Japan’s IT Implementation Subsidy (IT導入補助金)?
Possibly, but we do not promise it. It is a Japanese programme for companies in Japan, and applications must go through a certified support provider, so in some cases we work through a partner. Whether you qualify depends on the kind of tool and your size; approval is not in our hands. We will tell you the realistic chances in the 30-minute call.
How do we start?
Book a 30-minute call, online, in Vietnamese, Japanese or English. You describe one job your staff do by hand; we tell you what AI can take on, which step fits, and whether it is worth doing at all. If AI cannot help, we will say so, and you lose nothing but 30 minutes.
Tell us one job your staff do by hand every day.
Vietnamese, Japanese or English, whichever suits you. If AI cannot help with it, we will say so, and you lose nothing but 30 minutes.
Fixed price per package, quoted after the call — never by the hour.