Every Shenzhen CNC supplier website says the same things: precision machining, experienced team, quality first. None of it distinguishes a 200-machine factory from a five-machine shop with a good photographer.
The way to tell them apart is to ask for numbers that a real plant has on hand and a small shop does not. Here are eight, in the order that matters, plus what a good answer looks like.
1. Machine count by type, not total
"220 machines" means nothing without the breakdown, because a plant with 220 three-axis mills cannot do a five-face part in one setup.
Ask for: five-axis machining centres, CNC lathes, Swiss-type lathes, and grinding as separate numbers.
A real answer sounds like: "100 five-axis, 50 CNC lathes, 70 Swiss-type." A vague answer — "we have many advanced machines" — is the answer of a trading company.
2. The five-axis envelope, in millimetres
Five-axis capability has a size limit, and the limit is usually smaller than buyers assume. A 1000 × 600 × 500 mm envelope covers most automation and robotics work; a 400 mm part that needs simultaneous five-axis on all faces may not fit.
Ask for: the largest part envelope on a five-axis machine, and whether that machine does simultaneous or 3+2.
3. The tolerance they will actually put on the drawing
"±0.005 mm" is a shop standard, not a guarantee on your feature. The useful question is which tolerance on your specific drawing they will commit to, and on what instrument they will verify it.
Ask for: the guaranteed tolerance on your tightest feature, and the inspection instrument.
Anyone who answers "±0.001 mm on everything" is either not understanding the question or not planning to measure it.
4. CMM measurement uncertainty
This one is rarely asked and it separates the serious from the hopeful. A tolerance of 0.005 mm inspected on an instrument with 0.003 mm uncertainty is not really being inspected — the rule of thumb is a 4:1 ratio between tolerance and uncertainty.
Ask for: the CMM model, its stated uncertainty, and the date of the last calibration certificate.
5. CpK on a comparable part
Inspection proves a part was in tolerance. CpK proves the process stays there. A shop that has never calculated CpK on a machined feature does not run SPC, which means it detects drift at final inspection rather than at the machine.
Ask for: a CpK value on one critical dimension from a recent production run, ideally on a similar material.
6. On-time delivery, and its distribution
Quoted lead time is an average. What ruins programmes is the tail.
Ask for: the 50th and 90th percentile of actual delivery time over the last quarter, not the quoted figure. A shop that ships in 12 days every time is worth more than one that averages 9 with a 25-day tail.
7. First-article lead time, separately from production
These are different processes and they are often conflated.
Ask for: days from drawing to first article, and days from PO to batch shipment. At a well-run plant these are roughly 3–7 and 7–15 days. If both are quoted as "about two weeks", the first article is not being treated as a milestone.
8. What happens when a dimension is out
This is the only question that is not a number, and it is the one that predicts the relationship.
Ask for: the last time a first article failed, and what the plant did. A real answer describes containment, root cause, a process change and a re-submitted FAI. A weak answer describes a rework.
The remote audit artefacts
You do not need to fly to Shenzhen to verify most of this. Four artefacts cover it:
- A live video walkthrough showing machine nameplates, not a stock photo.
- One real FAI report with measured values visible (drawing number can be redacted).
- A CMM calibration certificate with a recent date.
- A written DFM response to one of your own drawings.
That last one is the highest-signal item on the list. The quality of a DFM response predicts the quality of the whole programme better than any certificate.
Red flags worth taking seriously
- A quality page with no named inspection equipment.
- A quotation that arrives without any manufacturability comment. A serious RFQ response flags tolerance and finish decisions, because those are where cost lives.
- Absolute quality promises. Reputable plants state inspection scope and process capability, not guarantees of zero defects.
- Prices well below the market with no explanation of machine park or routing.
- No willingness to share an FAI report before the batch ships.
The short version
- Machine count by type.
- Five-axis envelope in mm.
- The tolerance they will commit to, and on what instrument.
- CMM uncertainty and calibration date.
- CpK on a comparable part.
- p50 and p90 delivery, not the average.
- First-article lead time separately from production.
- What happened the last time an FAI failed.
Background on how our own plant answers these questions is on our Shenzhen CNC machining manufacturer page — 220 machines in Bao'an, ISO 9001:2015 and ISO 14001:2015, ±0.005 mm, prototypes in 3–7 days, production in 7–15 days, MOQ from 1 piece.
If you have a supplier answer you are not sure how to read, paste the wording in the comments and I will say what it does and does not tell you.
