Broad Link Advisory · Hong Kong

Your business runs on WhatsApp
and spreadsheets. It doesn't have to.

Broad Link builds AI-powered operations systems for SMEs and growth companies — intake automation, live dashboards, inventory systems. Built and shipped personally by Justin Chiu, who ran a Series B AI company as COO for seven years. You see it working in weeks, not quarters.

See before & after justin.chiu@broadlink-advisory.com
4 AI systems running in
real businesses today
Weeks from first call to a
working system — not months
10× faster than traditional
development builds
7 yrs running operations as COO
of a Series B AI company

Client work · before & after

What actually changes
when AI runs the busywork.

Not slideware — these systems run in real Hong Kong businesses today. Here's what each operation looked like before, and what it looks like now.

Flagship Photo-booth operator · HK & Macau

Machine breakdowns: from group-chat chaos to a 24/7 AI dispatcher

Before
部機壞咗呀，印唔到相 🙏09:41
📷 photo of machine screen 錯誤代碼：409209:42
好急！今日仲有 event!!09:45
邊個跟開呢部機？？11:03
✗ No task✗ No owner✗ Lost once the chat scrolls past
After
WhatsApp+ Slack→ AI parser + photo OCR→ ClickUp
維修事故 · 機 #042 — 錯誤代碼 4092
緊急 High Owner: 技術組 Due: today 📎 customer photo
  • 到場前確認現場狀態及錯誤代碼
  • 檢查相機、線材、驅動
  • 重啟並完成拍攝測試
Reported 09:41 → Acknowledged 09:48 → Resolved ✓
✓ Auto-created in seconds✓ Deduplicated✓ SLA tracked

What was built

An LLM reads every breakdown report — mixed Cantonese/English, photos included — and extracts machine number, urgency, work type, and a repair checklist. Vision OCR pulls error codes straight off customers' photos. Tasks are created, deduplicated, and closed-out automatically, with the WhatsApp label cleaned up when work is done.

Value created

Zero manual triage, around the clock. Nothing gets lost in chat history. And for the first time, management can see response and resolution speed per machine — metrics that simply didn't exist before.

n8nOpenAI GPT + VisionWhatsApp / SleekFlowSlack APIClickUp API
Photo-booth operator · HK & Macau

Sales visibility: from portal-clicking ritual to an 08:30 inbox digest

Before
  1. Log in to the vendor portal
  2. Click through yesterday's reports, booth by booth
  3. Copy numbers into a spreadsheet
  4. Compare against last month — by hand
  5. Repeat everything for the second currency
≈45 minonly when someone remembers
After
✓ Zero clicks✓ Both currencies✓ Never forgotten

What was built

A scripted pipeline that pulls sales data directly from the booth platform's API, compares month-to-date against last month and last year, and emails leadership every morning — with self-healing authentication so it runs unattended, even overnight. Plus downtime and error-log analytics that turned parts-replacement planning from gut feel into per-model evidence.

Value created

Leadership starts every day with yesterday's numbers and trend context already in their inbox. Maintenance budget decisions are now backed by error-rate data per machine model.

Node.jsGraphQLOAuth2 / JWTGmail APIScheduled agents
Fragrance & personal-care manufacturer · HK

Production records: from one fragile Excel file to a real system of record

Before
ABCD
PF1606…=VLOOKUP(B2,'Produc…3.5kg=IF(ISNA(MATCH(…
Tea Rose#REF!=C4*'In House'!$D$2??
HandWash=SUMIF('Diary'!A:A,…#VALUE!=B5-C5
DiaryProduct ListIngredient OH+5 sheets
8,757 formulasone person understands itno history
After
Finished Products155 records
Raw Materials184 records
Formula Lines1,253 records
Batch Runs + Inventoryfully traceable
📱 Scan barcode→ Match 460-SKU catalog→ Shared live count
✓ Every batch traceable✓ Stocktake = scan & count✓ Team-readable guide

What was built

A scripted migration parsed the entire workbook — 8,700+ formula cells — and normalized it into a relational Airtable base: products, raw materials, formula versions, batch runs, and inventory movements. On top: a phone-friendly barcode inventory app and a plain-language user guide the production team adopted without training sessions.

Value created

Tribal knowledge became a queryable system of record. Every batch traces back to its formula version and ingredient consumption. Stocktakes went from list-matching on paper to scan-and-count on a phone.

PythonAirtable APIReact + ViteSupabaseBarcode Detection API
Fragrance vending network · HK

Revenue truth: from month-end spreadsheet archaeology to a live dashboard

Before
📄eftpay_export_may.csv
📊device_records.xlsx
📊Fragrance per location.xlsx
📧venue contracts — buried in email
month-end matching by handmachines move venues — numbers break
After
Revenue▲
Transactions▲
Top venue
Daily revenue trend · per-venue variance · DOW × hour demand heatmap
✓ Live, login-gated✓ Contract-correct per venue✓ Daily summary email

What was built

A reconciliation pipeline that matches payment-provider records against device data — contract-aware, so when a machine moves between venues, revenue follows the contract that was active on that date. On top: a login-gated Next.js dashboard with KPI cards, trends, per-venue performance, and a demand heatmap, plus automated daily sales emails.

Value created

Month-end reconciliation became continuous and automatic. The team sees clean, contract-correct revenue per venue every day — and negotiates venue placements with heatmap data instead of hunches.

Next.jsReactSupabaseChart.jsAirtable

Approach

Senior operator judgment,
AI delivery speed.

You're not hiring an agency with a bench to fill. You work directly with one experienced operator who directs a team of AI agents — and who has sat in your chair.

Start with the P&L, not the tech

Every engagement starts from a cost or time question: what does this workflow burn each month, and what's it worth to fix? Builds are prioritized by ROI, then measured after shipping.

An AI team, directed by a human

Strategy and architecture, implementation, and QA each run on dedicated AI tooling with shared project memory. That's the 10× speed — without losing a single accountable owner: me.

Production discipline

Duplicate guards, dry-run modes, credential safety, fallback paths, and post-deploy verification on every workflow. I ran ISO 9001/27001 certifications as a COO; those habits ship with the code.

Built to hand over

Plain-language user guides, documented systems, and tooling your team actually owns. The goal is that operations run without me — not a retainer that never ends.

Daily toolbelt

Claude CodeCodexOpenAI APIn8nComposio / MCPNext.js / ReactPythonSupabaseAirtableClickUpSlack & WhatsApp APIsVercel

Who's behind this

Built by an operator,
not an agency.

Broad Link Advisory is run by me, Justin Chiu. Before this, I co-founded Apoidea, an AI fintech backed by Pacific Century Group, and ran it as COO for seven years — from a blank page to Series B, 120+ people across four countries. I owned HR, finance, and IT infrastructure; secured HK$40M+ in public funding; and passed ISO 9001 and 27001 audits. Before that, I priced risk as an actuary at AIG and Zurich.

I know what messy operations feel like from the inside — the WhatsApp groups, the spreadsheet nobody dares touch, the report that's always late. That's exactly what I build against.

Also open to conversations with teams who want this capability in-house full-time.

Co-founder & COO → Series B 120+ team · 4 countries US$13M raised HK$40M+ grants secured ISO 9001 / 27001 Actuarial roots — AIG, Zurich Full background on LinkedIn →

Contact

Have a workflow everyone hates?

Tell me about the spreadsheet nobody dares touch, or the WhatsApp group your business secretly runs on. A typical first engagement automates one painful workflow in weeks — you see it working before committing to anything bigger.

Describe your bottleneck LinkedIn

Hong Kong · English / 廣東話 / 普通話