The short version: how manufacturing digital-transformation projects get started is changing generations. The old pattern: the boss hears a talk and buys a system. The new pattern spreading across regions: "the enterprise writes the problem, government assists, platforms answer, the shop floor accepts, the market judges" — needs are written as problem statements, solutions are solicited openly, results are accepted on the shop floor. Good news for SMEs: free diagnostics, more buyer leverage, acceptance criteria defined up front. But you must first learn to write a good problem statement.
Policy is moving fast: some districts keep publishing AI-for-manufacturing scene lists and openly solicit technical solutions against real production pain points; one province runs the "enterprise proposes, government assists, platform answers, workshop verifies, market evaluates" model; cities send experts for free digital diagnostics, with hundreds of above-scale enterprises assessed online. The shared logic: turn application opportunities into identifiable, participable, evaluable projects, so the first yuan spent already has acceptance criteria attached.
1 · Writing the problem well: three elements, no exceptions
- Digitize the pain point. "Warehouse management is messy" is not a problem statement; "a full count takes 3 people 2 days, inventory accuracy is 85%, 20 mis-ships per month" is. Count today's hours, error rates and losses — the act of counting often reorders your project priorities by itself.
- Draw the boundary. Where the equipment vendor stops, where the software team starts, what stays in-house — written down first. Unclear boundaries are the most common way line projects die.
- Put the acceptor and the metrics up front. Who signs off? Measured by whose definition? How many times, before and after? Write all three into the requirements document and every later step has a referee.
2 · How to split the first budget
Once the diagnostic report is in hand, split the budget with our usual cost-module method (full breakdown in what custom software costs →):
- Narrow phase one. One scenario only — inspection or inventory, say. Keep the word "platform" out of the phase-one contract. A phase one that gets accepted and computes an ROI gives phase two its confidence.
- Budget data preparation separately. Cleaning history, unifying codes, keying base data — often a third of total effort. Hidden inside an "implementation fee" it breeds disputes; listed separately it stays honest.
- Reserve a discovery margin of 20–30%. However detailed the spec, the shop floor will surface gaps. A reserve for incremental requirements beats halting the project to renegotiate.
A practical way to judge a vendor: will they write acceptance metrics and definitions into the contract? A quote that only promises "satisfaction" parks all the risk at your end.
3 · The three usual traps
- The principal never enters acceptance. Leadership approves at kickoff; nobody signs at acceptance; the finished project has no owner. Decide "who accepts" during problem writing.
- Metrics without a baseline. If you never measured "how many today", nobody can say "how much better after". Two days of baseline measurement pay for themselves.
- Buying a big platform first. The most common black hole in transformation budgets. The scene-list model exists precisely to start small, accept each step — follow the rhythm the policy gives you; it is the stable one.
4 · The rhythm we recommend
- Weeks 1–2: sign up for the government's free diagnostic (most cities have one) while writing the digitized pain-point list internally.
- Weeks 3–4: take the list to 2–3 vendors for scope talks. Compare not the quote totals but who writes out the acceptance definitions first.
- Months 2–3: launch phase one, measure the baseline before and after, accept on the shop floor.
- Then: replicate or pause on ROI, review quarterly.