The six problems a machinery CIO cannot delegate

Our seven Chinese plants grew at different times and with different habits. The Changzhou excavator plants ran the European ERP suite installed in 2009 and an execution system built in-house; Hefei, acquired in 2012, still printed paper routing cards; Chengdu, opened in 2020, used a local planning tool. I do not believe a common system should erase every difference. It should make the differences visible and deliberate. Before each cutover we write down how the plant actually releases orders, issues material, records welds and closes work, and we compare that account with the template. Of 214 process definitions written since 2021, about a fifth carry an approved plant variant, each with an owner and a review date; the rest are common. The plant director, not my team, signs the cutover readiness check, because the plant carries the consequences. Six plants have moved since 2023, and Chengdu moves in November 2026.

Policy in China encourages state-linked companies and important industries to move core software, databases, operating systems and servers to domestically developed products, a programme usually described as IT application innovation. Our largest shareholder is a provincial state-owned investment company, so the expectation reaches us directly. I support the direction and dispute only the sequencing that a single deadline invites. We grade every application by how well its process is understood, how mature the domestic alternatives are in plants like ours and what a failure would stop. Finance and procurement moved first because their processes were well documented. Plant systems followed one at a time, in order of readiness. Overseas entities remain on the foreign suite until at least 2027, connected through an integration layer, because their auditors, tax filings and dealer links need a different answer. The approach missed a single-platform milestone our shareholder had hoped to report for 2025, and I explained the reasons to the board in writing.

Ruoshan's overseas revenue depends on 214 independent dealers who own the relationship with the customer. My team runs the portal through which they order parts, register machines, file warranty claims and see service history, and a hosted dealer management system for the 61 dealers that wanted one. The work is less about software than about agreements: what a dealer must share for a warranty claim to be paid, what we may see of their customers, and what happens to the records if a dealer leaves the network. Parts availability is the measure dealers care about most. First-pass fill from our Rotterdam, Dubai and Singapore hubs rose from 81 per cent to 92 per cent between 2020 and 2025, mainly because the portal showed us dealer stock and machine hours, not because we added inventory. Small dealers remain the hard case: 23 that adopted the hosted system abandoned it within two years.

Engineering drawings, routings and quality records flow from Changzhou to the Rayong plant and to overseas suppliers. Machine data, warranty evidence and dealer orders flow back. China's Personal Information Protection Law and Data Security Law, and the assessment and filing routes the Cyberspace Administration of China operates for outbound transfers, mean that each outbound flow needs a basis and some need approval. Requirements for ordinary business flows were eased in 2024, but anything that could be treated as important data still needs care, and location data from machines working on Chinese sites is a category I treat conservatively. We keep an inventory of 63 recurring cross-border flows, each with a business owner, a classification and a legal basis, reviewed with our legal team twice a year. Data collected abroad is governed first by the law of the country where it was collected, which is one reason our dealers' customer records stay on regional platforms.

A machinery plant's network carries traffic most IT security tools were not built for: robot controllers, line PLCs, hydraulic pump test benches, paint-shop controls and the engineering workstations that program them. Since the Hefei incident in 2023 we have divided each plant into zones by line, put a monitored boundary between every plant and the corporate network and removed every permanent remote connection. Fifty-two equipment vendors and integrators now reach our machines through a brokered service that records each session and closes it when the task is done. We hold verified offline copies of 4,100 robot and controller programs and test restoring a line every quarter. China's multi-level protection scheme sets baseline expectations for classified systems, including industrial control systems, and I treat it as the floor rather than the design. Each plant director owns the plant's risk measures and reports them monthly; my team provides the means and the monitoring.

Since 2022 we have run eleven AI pilots in plants, parts and service. Three have been scaled, two survive as narrower tools and six have been stopped. The weld-monitoring model at our Huai'an structural plant, which flags boom and arm welds for ultrasonic inspection from robot current and voltage traces, cut weld rework by 18 per cent and pays for itself. A parts forecast for dealers, built on machine hours from telematics, improved first-pass fill. The service assistant for overseas technicians became useful only after we restricted it to cited answers. The stopped pilots included a paint-thickness predictor whose sensors drifted and a failure predictor for hydraulic test benches that had too few failures to learn from. Every pilot now starts with a written stop condition and a named business owner from outside my department, and I report the stopped ones to the board's strategy committee alongside those we scaled.