Chief Operating Officer / Singapore / Southeast Asia
Choices that cost money first and earned trust later
- Group revenue, 2025
- S$1.06 bn
- 71 per cent from semiconductor-equipment makers
- Colleagues in five plants
- 6,700
- Penang, Johor, Bac Ninh, Singapore and Suzhou
- Recruitment fees repaid to workers
- RM 15.8 m
- To 1,412 migrant workers by March 2022
Decisions & consequences
01Repaying recruitment fees that Kelvorne never received
In June 2021 an audit commissioned by one of our customers interviewed 120 migrant workers at our Johor plant. Kelvorne had prohibited recruitment fees since 2019, but the rule bound our Malaysian agencies, not their partners in Nepal and Bangladesh. Our own interviews with 300 workers in Penang and Johor found that 78 per cent had paid, a median of about RM 10,900 each, often borrowed. Debt of that kind is a recognised indicator of forced labour, US customs had detained goods from several Malaysian manufacturers on such grounds, and the customer expected a remediation plan within ninety days.
I proposed repaying every worker still employed who had paid a fee, and every traceable worker who had left in the previous two years. Irene Phua, our chief financial officer, argued for applying the employer-pays rule to new hires only: the fees had gone to third parties before employment, we had no legal liability, and the bill could reach RM 20 million in a year that had not budgeted for it. Ravindran Muniandy, who ran Johor, feared that repaid workers would resign and that changing agencies during an order surge would leave his lines short. Both concerns were reasonable. The chief executive backed repayment, and the board approved it after the Audit and Risk Committee reviewed the method.
An independent social-compliance firm interviewed workers in Nepali, Bengali and Burmese. We set amounts from each worker's account, tested against known fee levels in each country, and paid in three instalments into workers' own bank accounts. We dropped two agencies and began direct recruitment in Nepal. Hiring slowed for five months, and about S$14 million of revenue slipped from late 2021 into 2022.
By March 2022 we had repaid RM 15.8 million, about S$5.1 million, to 1,412 workers. Voluntary attrition among migrant workers fell from 19 per cent to 11 per cent within a year, and the customer's follow-up audit closed the finding. Recruitment cost, about RM 6,800 a hire, now sits in our labour standard cost.
I would do two things differently. I would have asked workers to elect representatives to check our calculation before the first payment; 160 disputed their amounts, and some were right. And I would have published the payment timetable on the first day, because rumour filled the weeks we spent perfecting it.
Sending Suzhou's weldment lines to Penang instead of Vietnam
In November 2022 Varunda Systems, our largest customer at about 31 per cent of revenue, asked its suppliers to make parts for tools bound for the United States outside China, amid tightening restrictions on chip-making equipment. Our Suzhou plant made 212 of those part numbers, gas-line weldments, frames and cable assemblies worth about S$170 million a year.
Martin Oei, our chief executive, and two directors favoured moving all of it to Bac Ninh. Labour there cost about 35 per cent less than in Penang, the site had room, and Varunda had tied a 4 per cent price reduction to a Vietnamese move. I recommended a split: frames and cable assemblies to Bac Ninh, and the 138 ultra-high-purity weldment part numbers to Penang, where we already held Varunda's qualification for orbital welding, an ISO Class 5 cleanroom and certified welders. Building that capability from nothing while copying a process exactly seemed to me two risks taken at once. The board accepted the split. We conceded 2.5 per cent on price rather than 4, and reduced Suzhou from 740 people to 400, with S$6.8 million of severance.
We planned nine months and took fourteen. Forty-eight Suzhou technicians accepted secondments of six months, each part family ran in both countries until Varunda approved the change. The Penang day shift qualified on schedule. In August 2024, after the Suzhou technicians had gone home, night-shift first-pass yield on weldments fell from 97.6 to 91.2 per cent. Only three night-shift welders could diagnose a purge or electrode fault, against eleven on days. We brought four technicians back for ten weeks, and the transfer cost S$19.4 million against a budget of S$13 million.
By the first quarter of 2025 weldment first-pass yield in Penang was 98.4 per cent on both shifts. Varunda awarded Penang the gas-line work for its next-generation platform, about S$60 million a year from 2026. Suzhou now serves Chinese customers, with revenue of S$58 million in 2025, and every Kelvorne transfer carries a night-shift exit criterion.
I would certify night-shift welders alongside the day shift, not after it. And I would have told the Suzhou team about the reductions in the first month rather than the third, by which time they had heard it from a customer's supplier.
Opening Bac Ninh five months late to train its own supervisors
In June 2019 Kelvorne's board approved a S$52 million greenfield plant in Bac Ninh province for frames, cable harnesses and sub-assemblies. The plan, which my team had prepared and I had supported, opened the plant in March 2020 with 30 supervisors and technicians seconded from Penang and Johor for its first year. Two customers had scheduled qualification builds there for the second quarter of 2020.
By September 2019 I had changed my mind. The secondments would cost about S$4 million in the first year, and a plant run by visitors teaches its own people to wait for instructions. I proposed opening in August instead, after training 46 Vietnamese supervisors and technicians in Penang for sixteen weeks. Martin Oei objected that a delay would cost us credibility with both customers and leave a leased building empty, and one customer's purchasing director told us plainly that he was disappointed. We moved his qualification builds to Penang at higher cost, forgoing about S$2.3 million of margin on S$9 million of revenue, and paid S$0.8 million of rent on an empty building.
The cohort trained in Penang from October 2019 to February 2020 and went home weeks before Malaysia closed its borders in March 2020. Once the pandemic stopped travel, the secondment plan would have been impossible anyway. Bac Ninh opened in August 2020 with four expatriates, reached its planned output in March 2021, seven months after opening rather than the twelve in the original plan, and passed both customers' qualifications at the first attempt. In 2025 it employed 1,250 people and earned S$150 million of revenue. Nguyen Thi Bich Ngoc, a member of that first cohort, became its plant director in 2024.
I should be honest about luck. The pandemic made a sound decision look much wiser than it was, and I cannot claim the five months were the reason Bac Ninh coped with closed borders. What the decision did change was our default: new Kelvorne sites are now led by local supervisors from the first day, with visitors in support.
What I would change is the scope of the training. We trained production supervisors and technicians but not the people who would run payroll, customs clearance and the warehouse. Those functions struggled for six months, and two customer shipments were held at the border over paperwork we should have rehearsed.
Halting chamber-lid shipments rather than shipping on a deviation
In October 2015, two and a half years into running the Penang plant, I learnt that Varunda's integration factory had found 23 of the 410 aluminium chamber lids we had shipped that quarter failing its helium leak test. The cause was porosity deep in the aluminium plate, invisible to our dimensional checks and missed by the pressure-decay test we then used. Lids were worth about RM 140 million a year of our Varunda business, and its integration line was building tools for a chip maker's new fab.
Varunda's buyer proposed that we keep shipping under a deviation while his technicians retested every lid at integration. Our regional sales director supported him: it kept revenue flowing, and it was what the customer had asked for. I disagreed. A retest at the customer moved our defect into its factory, and we did not yet know how many plates were affected. I stopped lid shipments for nine days, quarantined 600 plates and 180 finished lids, and bought two helium leak detectors for RM 1.4 million so that we could test every lid ourselves.
The cost was RM 7.2 million, including scrap, expediting and a RM 0.9 million charge from Varunda for six days of slower integration. Our quarter missed its plan, and the sales director's frustration was justified; he had to explain the gap to a customer he had spent years winning. We found porosity in 41 of the quarantined plates, moved to a plate supplier that inspected ultrasonically, and added in-house leak testing for every vacuum part the plant made.
There were no leak escapes from Penang in 2016 or 2017. In 2016 Varunda chose Penang for the chamber bodies and lids of its next platform, and our chamber revenue reached about RM 180 million a year by 2018. People in the supplier-quality team I had once worked in told me later that the stop, more than the recovery, had decided the award, though price and capacity counted too.
I would change one thing. I told Varunda's buyer on the first day and let him carry the news to his own quality organisation. Graham Pettifer, its vice president for supplier quality and my former manager, heard two days later. I should have rung him myself that afternoon, as I would have wanted when I sat in his chair.