Craft Growth & AI Joint Consulting Co., Ltd.
We bring over 20 years of experience in traditional manufacturing management, including 10 years as a general manager. Not an outside consultant telling you what to do, but someone who has actually worked on the production floor. We diagnose your site first, make process-improvement recommendations, build custom management software when needed, and stay with you until the new system is fully adopted.
20 minutes to identify your factory’s three highest-priority areas for improvement.
Employees tick off inspections on their phones, the supervisor’s dashboard updates in real time, and overdue items or tasks that exceed the allotted hours are flagged automatically.
Screens show sample data, not real customer data. Mandarin voiceover with English subtitles; some on-screen interface labels remain in Chinese.
No jargon. We start on the shop floor and build only what you need.
We walk the production floor, watch the process, talk to the people doing the work, and find where things are really stuck.
Drawing on 20+ years of manufacturing experience, we turn the diagnosis into practical, actionable recommendations — not armchair theory.
Once the process is straightened out, we build to your needs. Not every company needs it, and we won't push off-the-shelf software.
We work alongside your floor staff so your veteran craftsmen are willing to use it every day, and we keep tracking the results.
70%+
of Taiwan's SMEs lack even a basic digital system — operating data is scattered across Excel and paper, or lives only in veteran workers' heads and verbal handoffs.
It's not that owners are unwilling to change. No one has told them, in language and processes the floor understands, where to start and how to take it one step at a time.
Our name tells you how we work — five letters, five stages.
No questionnaires, no workshops. We walk the production floor ourselves to see the real processes and pain points — the details only a 20+-year manufacturing veteran can spot.
We confirm with management the business problem this change must solve and the goal it must achieve, then — drawing on 20+ years of manufacturing experience — recommend process improvements you can actually implement. No “AI for AI’s sake” projects.
Once the process is straightened out, we build only what's needed — knowledge bases, report automation, cost and quality analysis — not a generic off-the-shelf package forced on you.
We work alongside your floor staff. With the right method, people used to the old way are willing to change gradually — we don't drop off a tool and a document and leave.
After rollout we keep tracking results, so the habits are still there and the numbers still hold six months later — not a flash in the pan.
These cases are drawn from internal manufacturing implementations personally led by our founder. All customer and confidential information has been anonymized; only verifiable technical facts and figures are kept.
Knowledge Loss
A 20-year technical note platform was being discontinued, putting all of its content at risk of disappearing — and the text inside scanned attachments couldn't be searched at all.
We built a local full-text search system (including Chinese OCR) so keywords, dates, and attachments are all searchable.
11 GB of data imported in one pass, in just 15–40 minutes.
Labor Drain
Thousands of business cards collected in the field had to be keyed into the computer one by one, and contacts couldn't be found when traveling.
Take a photo or paste an email signature and AI automatically recognizes and structures the fields, then syncs them to the address book every day.
Recognition costs only NT$0.18–0.35 per card.
Scattered Systems
Labor hours, complaints, and quality-improvement records were three separate ERP exports; tracing one root cause meant cross-checking them line by line by hand.
We merged them into a single dataset with natural-language search — just ask, “Why did this model run over hours?”
Historical data for 2,789 components, connected in one pass.
Verbal Reporting
Every quarter, board presentation scripts were written from scratch, and finding how the same topic had been explained before meant digging through years of files.
AI compares past meetings' slides with their transcripts and drafts a first version of the new script in the established tone.
Compared 13 past meetings and 617 slides (98.2% coverage); a draft script for a new deck takes 5–8 minutes.
Knowledge Loss
Twenty years of email and multiple archive files were scattered everywhere. The built-in search couldn't find anything, and tracing an old quote or the root of a complaint meant hours of digging.
We built a searchable email index with advanced query syntax (fields, date ranges, and more) that updates automatically on a schedule.
The first index covered several hundred thousand emails; daily mail is refreshed every 15 minutes.
Labor Drain
Every morning and evening, a status report had to be compiled by hand from the calendar and email; verbal promises in supplier and customer emails often went untracked and unfulfilled.
It automatically compiles the day's calendar and email highlights into a report, and keeps a “commitments” list: who promised what, and whether the deadline has passed.
Two summary reports are generated automatically every day, all year round.
Labor Drain
Cases in progress (complaints, negotiations, projects) required someone to watch the emails and update the status by hand — and when things got busy, they were missed.
Every day the system scans related emails, updates the tracking record automatically when there's news, and stays quiet when there isn't.
Scans at a fixed time every day; sends a notification only when something has actually changed.
Scattered Systems
Material usage, labor hours by trade, and defect records lived in separate exports, so finding out “which process step caused this rework cost?” meant manual matching.
We merged material, labor, and defect data into one dataset and broke it down by trade, so you can query the rework cost of any specific process step directly.
Uses the same database as the complaint/quality system, so results can be cross-checked across systems.
Scattered Systems
Past quotation documents were scattered across folders by year, so when estimating there was no way to find pricing for similar past jobs.
We full-text indexed every historical quote and derived unit-cost metrics by material, so estimators can look up historical pricing as a reference while quoting.
Consolidated 549 quotation files into nearly 3,700 unit-cost metrics; re-indexing everything takes only about 3 minutes.
Scattered Systems
Each year's gross-margin settlement reports were stored separately, so comparing margins across years, customers, or “sister products” built from the same mold meant flipping through spreadsheets by hand.
We connected multi-year margin data, automatically grouped sister products made from the same mold, and broke down margins into six cost categories to identify the sources of the differences.
The platform's output was checked against 1,143 records line by line and matches the original spreadsheets 100%.
Verbal Reporting
Quality self-inspection was still mostly filled in on paper or by hand in the system, so no one had real-time visibility into which items were overdue and which unit was responsible.
Every day it automatically compares the production schedule with self-inspection records. Overdue or missing items are grouped by responsible unit, and notifications are sent directly to team leaders' and section managers' phones. These notifications are escalated to higher management as the items remain overdue for longer periods. Weekends and public holidays are automatically excluded.
Syncs automatically early every working day; runs on a cloud platform within the free-tier quota, with no need to build your own server.
Verbal Reporting
Process progress for each component was scattered across the scheduling system; items past their due date without a completion report were only chased one by one in the weekly meeting — often when they were already several days late.
We reused the same alerting architecture, changing only the data source, to create a live progress-tracking dashboard. Overdue days are calculated excluding public holidays, overdue items are grouped by responsible team, and alerts are escalated to supervisors once the overdue period exceeds a specified threshold.
A single inventory run before launch surfaced 86 components and 342 overdue progress codes (estimated as of the reference date).
Verbal Reporting
The production schedule was managed in a static Excel Gantt chart, so whenever one task slipped, every task after it had to be pushed back by hand.
We switched to an interactive scheduling tool that supports task dependencies and has Taiwan's public holidays built in for working-day calculation; changing one task automatically reschedules everything downstream.
Iterated on real usage feedback and refined through four versions.
Verbal Reporting
A new component is about to start, but its standard hours haven’t been estimated, so the floor staffs the job by gut feel. Who should estimate it, and by when, depends on someone asking verbally.
Based on the schedule, the responsible engineer is reminded to start the estimate 5 working days before the start date, and the engineer and section manager are reminded to complete it 3 working days before. Working days automatically skip weekends and public holidays.
The schedule is checked automatically every working day, early morning and midday, with two-stage reminders — no one has to remember the dates.
Scattered Systems
Standard hours live in the scheduling system and actual hours in a separate time-logging system, so checking whether a process step has run over means comparing records by hand — and when hours keep accumulating after a step is marked complete, no one notices.
Standard and actual hours are matched by component and progress code. Only steps currently in work are listed, visible by responsible team; if hours are added after a step is marked complete, the relevant supervisors are notified automatically.
The first inventory covered 92 in-progress progress codes, 18 of which already had actual hours above standard (Oct 9, 2026).
A typical traditional manufacturer's process looks roughly like this, and every stage can be digitized step by step — this is the scope of what we do.
Today: Quoting relies on veteran staff's memory and verbal experience, and newcomers take a long time to get a feel for it.
Digitized as: A historical quote database plus a unit-cost metrics lookup tool.
Today: Drawings and process cards circulate on paper, and when a version changes, departments easily fall out of sync.
Digitized as: Online drawing and process version control, with changes automatically notified to the departments concerned.
Today: Inventory is judged by eye on the floor and counted by hand; shortages are often discovered only when the material is needed.
Digitized as: Electronic logging of material in and out, with automatic alerts when stock runs low.
Today: Schedules are managed on whiteboards or paper Gantt charts, and any change means redrawing everything.
Digitized as: An interactive scheduling system where dependent tasks update automatically.
Today: Paper work orders and paper self-inspection sheets are signed first and keyed in later; nobody knows right away when something is overdue.
Digitized as: Electronic forms entered in real time, with overdue items rolled up into a dashboard automatically.
Today: Problems are scattered across emails and verbal handoffs, so root causes are hard to trace afterward.
Digitized as: An integrated quality-event database that lets any department look up root causes directly.
Today: Delivery and warranty records are found by digging through folders by hand, so answering a customer takes time.
Digitized as: Systematic records of shipment and warranty history, with instant lookup for replies.
Today: Margins can only be worked out after several reports are compiled by hand at month-end, by which time it's too late to catch problems.
Digitized as: A cross-system gross-margin analytics platform that automatically consolidates data and allows users to access the results at any time.
“Knowing what to change and being able to change it are two different things — and that is exactly what sets us apart from typical consulting firms.”
Ph.D. in Engineering
A 20-minute diagnostic conversation. No hard sell, no empty promises — only the real problems on your floor.
Email Us to Book a Diagnostic CallOr write to us directly: contact@cgajc.com
