If you buy an iPhone or a Mac in Taiwan, there is one reseller besides Apple's own stores with shops all over the island: STUDIO A. It opened its first store in 2007. Behind the storefronts is STUDIO A INC. (晶實科技股份有限公司), part of the Foxlink Group. Today it runs nearly 60 locations across Taiwan and, beyond retail, serves campuses, government tenders and corporate purchasing.
Starting in March 2026, the company put Claude Code in the hands of 76 employees across 12 business units, in waves, and let them rebuild their own work with AI. Six months on, much of the data pulling, cross-checking and report assembly has been handed to AI, and some employees have gained back half a day or more every week.
- A Training Department staff member cut her weekly statistics report from 78 minutes to 4. Together with five other routine tasks, her weekly time on them went from 361 minutes to 28.
- A sales rep in the Education Business Division used to open nearly 280 tender notices one by one every day, spending more than two hours on it. A system now pulls them in daily and AI screens them first; the number of tenders that need a full specification download and a close read has dropped by 80%.
- Two members of the Procurement Department turned 17 routine tasks into fixed workflows in two months. Daily work such as arrival notices, serial-number conversion and education-price consolidation now runs automatically every day.
Challenge: The business outgrew its systems
Every day, STUDIO A's head office runs a set of routine work: weekly sales reports, education-price checks, government tender tracking, enrollment confirmation for every course. The data for all of it lives in the company's systems.
The membership system is built on Salesforce, which sends out LINE notifications, marketing emails and course registrations. Another set of internal IT systems was designed when the company was much smaller and is still in use today. Over the years the business has diversified and the numbers people need have changed, but adding a field or a report to a system means waiting for a slot on the vendor's or IT's schedule.
Whatever can't get a slot falls to people. Take the Finance & Accounting Department's weekly report: all the numbers are in the ERP, yet every week the finance team first has to export dozens of Excel files, reconcile them region by region, and then roll them up into company-wide figures. If a single file is pulled with the wrong date range, or one column is pasted in the wrong place, the numbers no longer tie out.
Finance is not an exception. Before the rollout, the company asked every department to take stock of its core work. The two difficulties checked most often were "compiling data takes too long" and "information is scattered and hard to organize." Asked which step goes wrong most often, departments gave similar answers: pasting into the wrong column, formulas not carried over, pulling the wrong category.
Approach: Let AI grow inside everyone's work
The year before, the company had tried an AI rollout, and nothing stuck; the output ended up scattered across different AI tools. Early this year, some employees installed OpenClaw on their own, got it running, and then stalled, not knowing what to connect it to. Looking back, Kat Chang, the Training Manager who led this rollout, puts the problem this way: AI never became part of people's everyday work; it "stopped at using a tool, one you could even skip." So this year she decided to work department by department, around each one's own needs.
This time there was one rule: everyone brings their own work and connects the whole thing, from data coming in to results going out. It only counts once it is connected end to end. A connected workflow stays, and the next time the same work comes up, it simply runs again.
As for how to connect it, every department and every person works in different steps, and head office can't know in advance which way works best. That is why the choice was Claude Code: it doesn't prescribe how it is used, so each person can fit it into their own work, following their own steps. Practices grow on their own first; the good ones are then brought into the company's systems.
In March, the Training Department, which runs the company's internal training, went first. The staff member mentioned above used to carry every monthly internal course from enrollment to completion by hand: download the list, check each trainee against several required courses to see who was eligible, email regional managers to confirm, then search her inbox for their replies one by one; before class, send notices and prepare sign-in sheets; after class, collect the sheets and log everything in the training records. She connected the entire chain with Claude Code. Now the lists, the eligibility checks and a draft of every email are ready for her; she reviews them, then hits send. She isn't an engineer, and she isn't in IT. The Training Department's experience became the starting point for everyone else. Over the following months, departments joined in waves. Who joined together was decided by the work, not the org chart: people whose work touches the same data were placed in the same wave. The Sales Department's weekly report is also used by Finance & Accounting, so staff from both departments joined in the same wave.
Confluence Partners designed the exercises for each wave, all drawn from that department's own work, with 6 to 12 hours of rollout per wave depending on need. The roles were clear. Confluence Partners set up the AI working environment, built a working understanding of Claude, and guided hands-on building and integration in the sessions. Claude Code did the step-by-step work: pulling data, cross-checking, drafting. The employees decided what to build and connected their own work end to end. From late August, the company began consolidating employees' accounts onto Claude Team.
Results: Individual practices grew first, then became the company's system
The first thing each person's workflows save is their own time. A few examples from the departments' results reports:
| Task | Department | Before → After | Time saved |
|---|---|---|---|
| Assessment process | Training | 91 min/week → 2 min | ~98% |
| Device serial-number lookup | Education Business | 300 records: 40–50 min → 2 min 40 sec | ~94% |
| Weekly reseller sales reconciliation | Education Business | 45–60 min → about 5–10 min (self-estimated), with automatic cross-checks | ~86% |
| Arrival notices | Procurement | 5–10 min each → about 2 min, runs daily | ~73% |
| Reading and logging project order emails | Education Business | 60 min/day → 30 min, now scheduled to run automatically | 50% |
Every person: improving their own work
These numbers are the employees' own work. Sales and finance aren't programming jobs, but what the work is each day, and which step takes the longest, is something the person doing it knows best, along with what needs to change.
The Education Business Division handles school and government tenders. One of its sales reps used to visit the government procurement website every day, search by keyword, and open the results one at a time. Over nearly two months, he built a tender-tracking system with Claude Code: it pulls in new notices automatically every day, AI flags the ones likely to be relevant, and he has only two decisions to make: whether to download the specifications, and whether to pass the tender to the rep responsible. Each time he finished a piece, he handed it to colleagues to use and kept improving it as they went. Tenders used to be passed along by word of mouth; now every one is recorded, and he can look back and see how many of the tenders passed on were eventually won.
Another colleague in the same division often gets questions from school teachers such as "I can't log in to the device management platform." He used to answer from memory, and when he couldn't remember, he had to ask a senior colleague. He turned the division's accumulated Q&A into a searchable web page: type "can't log in," and the standard answer comes up, ready to copy and send. New colleagues now look things up in the same place.
The Sales Department's weekly report draws on data in the ERP. The ERP hasn't changed and still has no API; a colleague still downloads the data as a CSV. The difference is what happens after the download. It used to be only the first step, with sorting, calculating and charting all done by hand; now the file goes into a dashboard, and the numbers and charts come out in one go. The finance colleague responsible for the company's P&L has also built her own analysis pages for budget, pricing and sales volume. She imports the data herself, and when the database's fields and structure need to change, she makes those changes herself too.
With Claude Code, they have gained a new capability: when they think of a better way, they can build it themselves, without waiting for IT or anyone else, and when something doesn't work smoothly, they fix it themselves.
"The biggest difference is that a lot of what people imagined for our systems or reports no longer feels 'impossible.' Whatever it is, ask Claude first."
Kat Chang, Training Manager, STUDIO A
The whole company: connections no one planned
As more grew out of individual work, new connections began to emerge at the company level. This August, the Education Business Division laid its members' results reports side by side and discovered that a tender-tracking system built by one sales rep and the bid proposals another colleague was producing with AI were in fact two halves of the same job: one finds the tenders, the other writes the proposals. When the two of them started building, each was thinking only of their own work; head office could not have drawn a connection like this in advance. Once they saw it, the division itself proposed joining the two halves into one flow: notices come in, AI assesses them and drafts a proposal, and the sales rep is notified.
The Sales Department and Finance & Accounting have gone further. Their numbers sit upstream and downstream on the same line; the two departments used to work separately and confirm figures back and forth by email and phone. Now both sides' numbers are connected in the same dashboard they built themselves, and the figures everyone accepts are the ones on it. Operations from each region up to the whole company are visible the moment the dashboard is opened; when management discusses strategy and makes decisions, this is the version they look at.
To this day, Kat Chang keeps receiving messages that colleagues send on their own initiative. Some say what they built has helped them a great deal, especially with the hardest parts of their work; others say that when they shared what they built, colleagues who used it found it useful too.
"Acceptance is high now, applications are blossoming everywhere in people's work, and the quality has gone up. From the company's point of view, the next step is to converge and keep the best. Beyond that, it's putting the time we've saved into more projects, so that AI actually brings revenue to the company."
Kat Chang, Training Manager, STUDIO A