We find where the work gets stuck, then fix it the way your team works
We work with companies of 10–100 people with process-heavy, repetitive work, with a priority on professional services: accounting and bookkeeping, legal and land-registration services, consulting and insurance, and medical clinics. Automation, AI and custom software are tools we use, but every engagement starts with finding the problem.
A 30-minute call first, then a look at the workflow
You don't need to know what to fix yet. Thirty minutes won't find the real problem, but it is enough to judge whether a closer look is worthwhile. The real problem-finding happens in the next step.
Book a 30-minute intro call- 1
30-minute intro call
Free. You describe how the work gets done today. Kai Wu judges whether it is worth a closer look and where to start, and says plainly which parts don't need doing yet.
- 2
A look at the actual workflow
This is where the real problem-finding happens. We watch the work run once: which tools you use, where the data comes from, which step only one person knows how to do. A requirements document helps, so bring one if you have it. Before changing anything, we write down the acceptance checklist and which numbers to compare while old and new run side by side. Whether to continue after that is your decision.
Problems we find and fix
These five come up most often. The fix depends on the problem: sometimes a process change or cleaner data, and only sometimes code or AI.
Someone spends hours every week building the same report, and when the numbers don't match, nobody can say how they were calculated
Report automation
We first write down which fields and rules each number comes from, then turn the report into a single button. Old and new run side by side, the numbers are compared cell by cell, and the old process is switched off only after you sign off. An operating manual and a rollback plan are included.
Example: a 31-KPI weekly report for a medical group with multiple clinics, all 7 steps automated.Numbers from multiple branches and sources live in different places, and seeing the whole picture means consolidating them by hand
Data dashboards
We bring the data into one place in one format, shown as status lights and trends. Every number has data lineage: hover over it to see where it came from and which rules were applied.
Example: finance and operations figures for 35 branches, visible to head office on a single screen.You spend money on ads but can't tell which ones actually lead to a sale
Ad data integration and attribution
Google, Meta and LINE (the dominant messaging app in Taiwan) combined on one screen, with every metric defined the same way. We connect click → conversation → booking → sale to calculate true ROAS, not the conversions each platform reports for itself.
Example: an ad agency's client accounts sync automatically every day, and each client sees only its own data.Repetitive manual work takes up staff time, and the data is locked in legacy systems or behind an intranet
Workflow automation
Customer service, CRM, billing and payments, pulling data out of legacy systems: repetitive manual work handed over to the system. We have automated legacy systems that could only be viewed over a VPN, without changing the existing equipment.
Example: National Health Insurance claim status for 33 branches, moved from manual branch-by-branch checks to a scheduled automatic summary.Professional content has to keep coming, and every piece has to clear banned regulatory terms and search engines' standards for professional content
AI content pipeline
AI writes the draft, the system checks it for banned regulatory terms, and after human review it is published in one click. Quality over volume, meeting search engines' high bar for professional content.
Example: a content pipeline for a medical website: AI draft → check for terms banned under Taiwan's Medical Care Act → human review → one-click publish.
Three principles for every delivery
Replace safely, not recklessly
We don't hand over a tool and walk away. A parallel run, a rollback plan and handover documentation are all part of the delivery. Before an old process is retired, the data has to show that the new one can be trusted.
Solid data first, then AI
If the data can't be trusted, AI only reaches the wrong conclusions faster. We integrate, standardize and confirm the data is reliable before we talk about AI analysis and automated decisions.
Legacy systems and intranet data can be fixed too
We have automated legacy systems that can't be modified and data locked behind a VPN. Leaving your existing equipment as it is is our default assumption.
Tools we use
- Supabase
- Odoo
- n8n (self-hosted)
- Docker
- Next.js / Vercel
- Playwright
- Claude / AI agents
We pick the tools after we've seen the problem.