A factory digital transformation plan: four questions, three starting points, one year

A factory digital transformation plan most often turns into a list of equipment and software, and at year end nobody can say which item changed what. A plan that lands answers four questions first: what is the goal, where do we start, in what order, and how is it accepted. Only then does the purchase list matter.

Four questions

Goal. Not "become digital" but a specific problem: stock record accuracy, on-time delivery rate, hours spent on manual consolidation, downtime response time. A goal needs a metric and a current value, or progress cannot be judged.

Starting point. One of three — data, process or equipment. Running all three at once dilutes both budget and attention.

Sequence. State the precondition of every step. Scheduling needs usable operation and hour data; costing needs accurate issue, return and work-in-progress records.

Acceptance. Each stage has deliverables and pass criteria that a third party can re-check, not "everyone feels it works well".

Three starting points

Data. Symptoms: reports built by hand, one metric reported three ways, month-end closing runs into overtime. First move: a metric dictionary and consistent documents, before any system.

Process. Symptoms: the same task done several ways, a different person bringing a different routine, exceptions handled verbally. First move: draw the key process, fix a single entry point and one owner.

Equipment. Symptoms: frequent stoppages with no records, unclear equipment efficiency, maintenance by experience. First move: status and reason collection on three critical machines.

Choosing a starting point

Pick by whichever symptom most affects delivery or cost. The three depend on each other: equipment data only matters when it reaches a process, and a process only becomes measurable when it lands on documents. The starting point fixes the first step, not the destination.

Where the money comes from

In 2023 the Ministry of Finance and MIIT issued the notice on pilot cities for SME digital transformation (Caijian document 2023 No. 117), setting up three batches of city pilots from 2023 to 2025. On 21 June 2024 the two ministries published the second batch, 36 cities and districts. Pilot cities support local SME upgrades, with the eligible industries, thresholds and support ratios published by each city.

So before a plan says "apply for policy support", three checks are needed: is the city on the pilot list, what does this year's local document require, and does the planned work align with its acceptance criteria. Those criteria tend to be concrete — equipment connectivity rates, CNC coverage on key operations — and a plan whose goals map onto them is easier to review.

The SME Digital Transformation Guide, issued by the MIIT General Office on 3 November 2022 (MIIT document 2022 No. 33), sets out a path across building transformation capability, improving supply and increasing policy support. It is a useful frame, not a template: it addresses the national SME population, not one plant.

Do the arithmetic before the systems

Two sums: the economic impact of the problem, and the boundary of the investment. The first says whether it is worth doing, the second says where to stop.

Impact can be estimated coarsely but must hold up: how much capital sits in a one-point stock discrepancy, how many hours go into monthly consolidation, how much output one stoppage costs. These numbers need not be precise to the decimal, but they must come from your own records rather than industry averages.

Write the boundary into the plan: budget ceiling, time ceiling, and what counts as sufficient. Projects without boundaries expand until nothing is finished.

A one-year roadmap

Months 0-3: establish the facts. Metric dictionary, consistent documents, one process map, plus collection on three machines or two operations. The deliverable is data that reconciles, not a system.

Months 3-6: connect data to action. A dashboard of at most ten metrics, alerts on thresholds, alerts converted into tasks with owners. Success is measured as the time from an exception to someone acting on it.

Months 6-12: extend and institutionalise. Copy what works to a second line or workshop and write definitions, processes and owners into policy. The usual failure here is not technical: people leave and the rules leave with them.

Three failing plans, and when to bring in help

1. Everything at once: three to five modules planned, each half-built. Fix: one goal, one starting point for the first year.
2. Software without process change: the old routine inside the new system, producing worse data. Fix: settle the process, then choose the tool.
3. No acceptance metric: the year-end review can only say a system went live. Fix: one re-checkable metric per stage.

Bring in outside help in three cases: nobody internally can drive it full time; a choice between options lacks a basis for judgement; or cross-department definitions need a third party to chair. Otherwise, finish what you can do yourself first — it is cheaper.

The Action Plan on AI plus Manufacturing, issued by eight departments (MIIT joint document 2025 No. 279, 25 December 2025), places data management and scenario selection ahead of model deployment in its annex — the same order as the roadmap above. Reverse it and the model simply runs faster on wrong numbers.

If you are writing this year's plan, send us the goal, the budget and one current-state process map (contact) and we will check whether the starting point is right. To see a transformed plant on one screen, look at the T001 manufacturing AI service centre.

In one line: the first page should say which metric moves, from which starting point, and how success is measured — worth more than ten pages of features.