A domestic ERP will not fix a process we never wrote down

Most Chinese manufacturers with a state shareholder, and many without one, are now planning to replace foreign core software with domestically developed products. The policy, usually described as IT application innovation, reaches enterprise resource planning, databases, operating systems, office software and servers. I support its direction. A machinery maker whose finance, procurement and production planning depend on a foreign vendor's support calendar carries a risk that has nothing to do with politics: the vendor can end maintenance for a version, change its licensing or withdraw from a market, and the customer has little say.

My argument is narrower and more practical. Most substitution plans I have seen, including the first draft of our own, are organised around a date. Every entity and every plant moves by the end of a given year, and progress is reported as the share of systems that have switched. That measure misses the main risk. A core system is a record of how a company works. Where the company has never written down how it works, the new system will be configured from guesses, and the guesses will be tested on the first busy day after cutover.

What we were really replacing

When we began planning Ruoshan's move in 2021, we asked each team to list the custom programs it relied on in the foreign suite installed in 2009. There were 1,180. Forty-three per cent had not run in two years. Of those still in use, many existed because a plant had once done something differently from the template and nobody had decided whether the difference was worth keeping. One plant issued steel plate against a planned cutting pattern; another issued it against the actual nesting results from its cutting machines. Both recorded the issue in the same field, and month-end variance reports compared the two as though they meant the same thing.

We then asked each plant to describe, in writing and in its own words, how it released a work order, issued material, recorded a weld, handled rework and closed an order. Several could not do it without a week of argument among their own engineers. That was the most useful week of the programme. It showed that what we were replacing was not a software product. It was twelve years of habits, some sensible and some accidental, that the old system had absorbed without anyone choosing them.

A domestic system configured on top of that would have inherited the same confusion, probably with more of it, because a new vendor's consultants learn a client's processes from the client's documents. Where there are no documents, they learn from whoever is free in the workshop that afternoon. I spent four years as a delivery director at a software company, and I know how much of a configuration is decided in exactly that way.

Sequence by readiness, not by calendar

We graded every application on three questions. How well is its process written down and agreed? How mature is the domestic product that would replace it, judged by references in plants like ours rather than by its brochure? And what stops if it fails: a report, a shipment or a production line? Finance and procurement scored well on the first two and moved first, in January 2023. Plants followed one at a time, in the order in which they finished their process definitions, not in the order that made the best presentation. By December 2025 six of seven plants had moved, and the seventh moves this November.

The approach had costs that I will not pretend away. We missed the single-platform milestone our largest shareholder had hoped to report at the end of 2025. Running the foreign suite for overseas subsidiaries alongside the domestic core costs about RMB 36 million a year until we decide their future. And one cutover failed even so. At our Huai'an structural plant in August 2024, converted bills of materials for 3,400 weld assemblies contained errors that our rehearsal, run on a sample of orders, had not exposed. The plant issued material by manual pick list for nine days and 310 excavators were delayed. That lesson concerned rehearsal rather than sequence, and we now rehearse every cutover with a full month of real orders.

Readiness also meant being honest about where domestic products were stronger. The execution system we chose handles Chinese quality-record formats, local tax invoices and the shift patterns of our plants more naturally than the foreign system it replaced. The ERP core needed more work on multi-currency consolidation for our H-share reporting, and we built that with the vendor rather than assuming it would arrive in the next release. A substitution plan that treats every domestic product as equally ready is as careless as one that treats none of them as ready.

Questions for a board approving the plan

Boards in China will approve many of these programmes over the next few years. Directors need not judge software. They can ask questions that reveal whether a plan is built on the process or on the date.

First, ask what share of the processes the new system will run has been written down and agreed by the people who perform them. If management cannot answer, the cutover dates are estimates of optimism.

Second, ask how many custom programs exist in the current system and how many will be carried forward. A large carry-forward means the company is moving its accidents into new software.

Third, ask for the order of migration and the reason for it. An order chosen by readiness will look uneven on a slide. An order chosen for reporting will look tidy.

Fourth, ask how each cutover will be rehearsed, with what volume of real transactions, and who holds the authority to postpone it. At Ruoshan the plant director signs the readiness check, because the plant bears the consequence of a bad weekend.

Fifth, ask what will happen to overseas entities, whose auditors, tax authorities and dealers may need a different answer, and when that decision will be taken.

None of these questions opposes substitution. They make it more likely to succeed. The companies that will regret their programmes are not the ones that moved carefully. They are the ones that moved on schedule with processes nobody could describe, and then spent two years discovering those processes in production.