ABOUT
I spent twenty years running the operations I now help other people fix.
TO BE SUPPLIED
For two decades I ran distribution operations — Bossard, Avnet, Arrow Electronics, DKSH, Zuellig Pharma. Healthcare distribution centres at 100,000 sq ft scale, cold chain and ambient. Multi-shift teams, high-volume outbound, and the GDP, GDPMD and ISO 13485 accountability that comes with moving medicine.
The numbers I’m proudest of came from process, not technology. On-time delivery from 55% to 98%. Inventory accuracy from 10.5% to 78% in a warehouse nobody wanted to touch. USD 7 million out of an operating cost base. Overtime from 45% to 5%.
But I also built things. A warehouse management system on SQL when the budget wouldn’t stretch to a real one. A shop-floor production app. E-approval systems that cut PO turnaround by sixty percent. Power BI dashboards, RPA, integrations. I’ve been the person who has to use the system at 6am, and the person who built it — a rarer combination than it should be.
What I learned across all of it: technology was never the hard part. Nobody had written down how the work actually happened. We’d spend six weeks drawing the process and two weeks building — and the six weeks were where the value was.
Then generative AI arrived. Large companies started spending enormous sums solving problems they hadn’t defined, while a sixty-person firm — where an admin task really is fifteen percent of someone’s job — got ignored, because consulting economics don’t work at their size.
I started Process Room to serve the second group with the discipline of the first. I read the releases, run the models and test the agent frameworks so you don’t have to. But I still open with the same question I opened with on the warehouse floor: show me how this actually works.
Being one person
Process Room is me. You deal with the person doing the work rather than an account manager, and I run no more than three engagements at once. Where one needs more hands, I bring in people I’ve worked with — same confidentiality and data terms, and I tell you before it happens.
It also means everything I build is documented to a standard where you’re never dependent on me. Your accounts, your credentials, your systems. If I disappeared tomorrow, your team could keep running it. That’s a design principle, not a reassurance.
Credentials
- Operations
- Two decades across pharmaceutical, medical device, FMCG and electronics distribution. Multi-country: Malaysia, Singapore, Thailand, Philippines.
- Regulated
- GDP · GDPMD · ISO 13485 · GMP · NPRA/MOH · ISO 27001 · TAPA. Audit-facing, CAPA-closing.
- Technical
- LLM systems and agents, RPA, Power Automate, n8n, Python, SAP, JDA WMS, Power BI, custom application development.
- Education
- Master of Business Administration, Universiti Sains Malaysia. Lean Six Sigma Green Belt.
- Based
- Klang Valley, Selangor. Working across Malaysia.
What I believe
- Most AI projects fail for process reasons, not technical ones. The model is rarely the bottleneck. Undocumented workflows are.
- The right amount of AI is usually less than expected. Plenty of “AI projects” should be a rule-based automation, an off-the-shelf tool, or a deleted step.
- Confidentiality-constrained firms get most of the value from none of the sensitive data. Intake, scheduling, documentation, reporting and internal knowledge are all outside the boundary and all high-value.
- An AI adoption that only works on today’s tools is a liability. Build the process so the model, the vendor and the hosting can all be replaced without starting over.
- The job is to become unnecessary on this problem and necessary on the next one.