How to Choose a CNC Machining Supplier: A 7-Step Guide for Buyers and Engineers

Home/Buyer's Guide/How to Choose a CNC Machining Supplier: A 7-Step Guide

Published 2026-10-08 ｜ Last updated 2026-10-08 ｜ 16 min read ｜ about 4,500 words

With a weighted supplier scorecard, three gate checklists and 6 costly mistakes

You send the same drawing to three shops. The quotes come back at 200, 500 and 1,500 yuan. How do you choose a CNC machining supplier? Most people take the cheapest one — and that is usually where the trouble starts.

An old rule of thumb in this trade: rework typically costs 3 to 8 times the machining cost of a single part. It is not a precise figure, but the direction is right. A part that saved you 300 yuan in machining, if a hole is found 0.08 mm off at the assembly station, costs you rework, freight, a stopped line and the production slot you lost. The 300 yuan you saved can end up a 2,000-yuan hole.

Choosing a CNC machining supplier is not a price comparison. It is pricing a risk. We are Shenzhen Goldencube, a precision machining shop founded in 2012, 14 years in the trade. From the factory floor side, this guide breaks down the cost structure of a quotation and gives you a 7-step screening method, a weighted scorecard you can fill in today, three gate checklists from prototype to mass production, and 6 post-mortems of costly mistakes. There is also an 11-item RFQ checklist to prepare before you ever send a drawing out.

1. Define the need first: you need a shop that matches your stage, not one that "can do it"

Most disputes we have seen come down to this: a mass-production requirement taken to a shop that is good at prototypes, or the reverse. The test is simple — ask how long their last order above 500 pieces took, then ask where a 5-piece order sits in the queue.

The first step is not finding a shop, it is characterising your order. One axis is where the information comes from: build-to-print is about process capability and inspection; build-to-sample (including reverse engineering) needs one extra question — will you send the measured drawing back for sign-off? Free-issue material needs material yield responsibility agreed in advance. We handle all three; see Products & Services for the capability boundary.

The other axis is how many pieces. This one is ignored more often, and mismatched more often.

Prototype (1–20 pcs), low volume (20–500 pcs) and mass production (500 pcs+) look like a difference in quantity only, but the scheduling logic is entirely different. A production-oriented shop runs long orders; your 5 pieces get squeezed into a gap — not because they cannot make them, but because the priority is naturally lower. A shop good at mass production is often bad at prototypes, and vice versa.

For projects that cross stages: screen against the standard of your final production stage, then check whether they will take your prototype order seriously.

2. What is inside a quotation: the cost structure of a CNC machining quote

You cannot judge whether a price is reasonable until you know which blocks it is built from. Without that step you can only guess from "what others charge" — which is exactly how you get led around.

A CNC quote usually contains six blocks: material (including cutting loss and nesting yield), machine hour rate, programming and fixture amortisation (one-off costs), tooling and fixture wear (rises sharply with stainless and titanium), outsourced finishing (by area or per piece, plus return freight), and inspection + overhead + margin. Full-dimension inspection and CMM reporting take time — they are a cost, not a free extra.

Typical industry split: machining accounts for 40%–70% of total cost (CNC lathes about 40%–60%, machining centres about 50%–70%). ⚠️ These are common industry ranges that vary by region, machine model and batch size — not a quotation commitment.

2.1 Machine hour rates: why published ranges are too wide to copy

For the same "vertical machining centre", three sources range from 35 to 150 CNY — more than four times apart. The data is not wrong; the definitions are simply not comparable. Some calculate from depreciation plus labour, some from market rate; tier-one cities differ from industrial towns; loose-tolerance high volume and tight-tolerance low volume are not the same job at all.

How we do it: our quotation separates hours and hourly rate — "estimated hours × hourly rate" — and states whether the part runs on a 3-axis, 4-axis or 5-axis machine. You can check it directly: on the same drawing, if two shops quote hours that differ by 100%, one of them has a far more convoluted process route.

In one line: do not use hour rates to "check the market" — use them to demand a breakdown.

2.2 The tolerance lever and the quantity lever

Moving a tolerance from IT10 to IT7 can add 30% or more to machining time. The extra time is extra operations — semi-finishing and finishing have to be split, cutting parameters come down, in-process measuring and tool compensation happen more often.

One layer fewer people talk about: tighter tolerances also raise the reject and scrap risk, and that is usually already priced into the unit price. Part of what looks "expensive" is insurance. For the inspection capability behind each tolerance grade, see the CMM, 2.5D vision and height gauge setup listed in our inspection section.

Actionable advice: put tight tolerances only on functional faces and leave the rest at the general tolerance grade of ISO 2768-1 (GB/T 1804-2000 in China). This single change usually cuts a quotation by 10%–30% without losing one bit of function.

Quantity works the same way: programming, fixture building and first-article setup are one-off costs, diluted by quantity only. The industry rule of thumb is 15%–25% lower unit cost beyond 100 pieces. A shop willing to quote 50 / 200 / 500 tiers is more credible — its cost model is calculated, not copied from a competitor.

2.3 A good quote vs a dangerous quote, item by item

Screenshot this table and review quotes against it with your colleagues. One extra question flushes out a lot of padding: "Beyond this quote, what other charges could arise?"

3. The 7-step screening method: narrow it down to 3 suppliers

Each step has the same three parts: what to look at / what to ask / red flag. Short on time? Do steps 1 and 6 — they eliminate the most.

Step 1 | Credentials and physical verification: real factory or a trading front?

• Look at: does the business licence scope include "manufacturing/production" (verify scope and status on the National Enterprise Credit Information Publicity System)? Can they name machine brands, models and commissioning years?
• Ask: "Can we do a live workshop video this week? I want to see which machine this part runs on."
• Red flag: "The factory is far, not convenient to film" or "we take orders, production is at a partner plant". The first usually has no factory; the second is a broker. Brokers are not unusable — just know you are paying two margins.

Step 2 | Equipment and process fit: do they have the machine your part needs?

• Look at: does the travel cover your part size, tool magazine capacity, is there an on-machine probe, can multiple faces be done in one setup (this directly drives geometric tolerances and datum consistency)?
• Ask: "How is this part fixtured? How many setups? How do you keep the datum consistent?"
• Red flag: says "we have 5-axis" but cannot name the model; cannot say whether the part should be milled or turned.

Step 3 | Inspection capability and reports: having a machine is not having a system

• Look at: are CMM, 2D vision, roughness tester and hardness tester in place; are critical dimensions inspected 100% or sampled, and against what acceptance rule?
• Ask: "Are critical dimensions inspected 100% or sampled? How long are records kept? Can a full-dimension report ship with the parts?"
• Red flag: do not relax when you hear "we have a CMM". Follow up with the sampling ratio and the record retention period. Buying a CMM is a purchase; turning it into a system also takes documented procedures, a calibration plan and trained people. Many shops only did the first part.

Step 4 | Similar cases in your industry

• Look at: have they made similar parts for your industry. Medical involves ISO 13485, automotive involves IATF 16949 and a PPAP package, export parts involve material certificates and RoHS/REACH. The common baseline is ISO 9001.
• Ask: "Have you made a similar structure? Can I see a redacted inspection report sample?" Our own work is shown in Products & Services — precision hardware parts, liquid cooling components, HIFI metal earphone shells, optical communication components, humanoid robot parts and precision fixtures.
• Red flag: do not be pleased by a shop that shows you another customer's full drawings. That does not prove competence; it proves they cannot control drawings. The next one sent out will be yours.

The first four steps test "can they make it". Steps 5 onward test "can they make it consistently". Most people stop at step 4, because by then price is on the table. The real gap is in the last three steps.

Step 5 | How structured is the quotation (use section 2 here)

• Look at: itemised or lump sum.
• Ask: quote the same drawing at 50 / 200 / 500 pcs and see whether a quantity-price curve comes back.
• Red flag: three quantities at almost the same unit price — the quote was not calculated.

Step 6 | Engineering response and DFM feedback

• Look at: what questions they raise after receiving the drawing. The quality of DFM (design for manufacturability) feedback is the cheapest litmus test of engineering level. Asking only "material, quantity" is routine quoting. Asking whether a bore is gauged 100% with go/no-go, how many hours of salt spray, what the burr standard is, whether tool marks may remain at the bottom of a counterbore — that is someone who actually read the drawing.
• Ask: "In this revision, which dimension do you think does not need to be this tight?"
• Red flag: a quote in 30 minutes with zero questions. That is not efficiency; it means the drawing was not read.

A typical case (not one specific customer): a drawing arrives with 12 dimensions, all at ±0.01. Our process engineer looks it over and asks: "Of these 12, how many are assembly faces?" The engineer thinks and says two. The other 10 are released to general tolerance, and the quote drops by roughly 20% — consistent with the range given in §2.2.

One sentence our process team repeats: "The first question when quoting is not quantity — it is which dimensions are functional. That decides most of the cost and almost all of the risk."

Step 7 | Outsourcing and lead-time control

• Look at: is finishing in-house or outsourced, is the subcontractor fixed, who is liable when colour or film thickness is off; can they give an hour estimate (only meaningful within 10%).
• Ask: "Is anodising done in-house or outsourced? If it is late or off-colour, who takes responsibility?"
• Red flag: "We cannot control the subcontractor" or "lead time is hard to say". Delay by a subcontractor is the CNC supplier's full liability, not force majeure — you signed with them; subcontracting further is their internal matter.

After 7 steps you should have 2–3 suppliers left. The next move is not comparing prices — it is scoring.

4. A weighted scorecard for CNC machining suppliers (use it as-is)

At this point most people make the same mistake: letting unit price decide everything. The scorecard exists to prevent that — it gets purchasing, engineering and quality speaking the same language on one sheet.

How to use it: score each dimension 1–5, multiply by the weight, sum for a 0–5 weighted total. Any red line is immediate elimination, no scoring — a veto binds harder than a score.

Adjusting weights: for medical or automotive parts, raise traceability to 25% and quality system to 30%. Hardware teams doing functional validation should add 5% each to engineering support and lead time.

One boundary condition: with fewer than three candidates the scorecard tells you little — there is nothing to compare. The answer then is not scoring, it is building a bigger candidate pool.

Pace: screen 5 → document review and scoring 2–4 working days → narrow to 2 for prototyping → 20–100 pcs trial run → approve as a qualified supplier.

Do not just look at a certificate copy: ISO 9001 certificates can be verified on the national certification information platform by certificate number or company name — search "Shenzhen Goldencube Technology Co., Ltd."; we are ISO 9001:2015 certified.

5. Prototype → low volume → mass production: three gates

One counter-intuitive conclusion first: making good samples and making good production runs are two different capabilities. A sample is coaxed out by a skilled operator — for five pieces he can re-set the tool, drop the feed and measure by hand three times. Production is repeatability: different operator, different shift, different tool batch, same result.

Always ask before releasing low volume: "After sample approval, will the low-volume run use the same program and the same tooling stock as the prototype?" If the answer is "we will re-schedule and re-set the machine", the prototype process was never frozen — and the low-volume result is unpredictable.

6. Six post-mortems: where they went wrong

These six are not the story of one customer; they are patterns that repeat in this trade. Change the shop or the buyer and the script is much the same. (No company names, amounts or dates.)

① The low-price trap: a beautiful sample, a cut-corner production run. Win the order well below market, make a beautiful sample; then, to cover the loss, compress cycle time, drop finishing passes and loosen sampling. The whole batch is found out of tolerance on the customer's assembly line. Rework follows — and with a line stop on top, it costs far more than the machining fee saved. → If a quote is 30% below the market average, ask how that is possible. There are only three ways to cover a loss: compress the cycle, drop operations, loosen inspection. The third is the most expensive.

② Overstated accuracy: verbal promises, no report. ±0.005 mm promised verbally and charged at a precision rate, then no inspection report with the delivery. When assembly interferes and you ask, the answer is "our sampling was fine" — the batch is scrapped with no recourse. → An accuracy promise must be written into the RFQ and the contract together with the inspection method and the report format.

③ Skipping the prototype: straight to a batch to hit the schedule. The milestone is pressing, purchasing asks "can we skip the prototype and make 300?", the shop says yes. It can — but the hole datums on the drawing were derived by design from the assembly relationship, while the shop locates from a different datum during fixturing, so the holes come out shifted as a group. It is only found at assembly; the whole batch goes back, and rework takes longer than prototyping from scratch — the milestone slips further. → The industry norm for prototyping is 3–7 working days (depending on complexity and outsourced finishing). Be wary of anyone promising "samples today".

④ Outsourcing blame: anodising colour drift. Parts come back from the anodiser visibly off-colour with the film too thin; the supplier says "we do not control the subcontractor" and asks the customer to pay the expedite fee for re-anodising. → Add one line to the contract: whoever performs the finishing, the main supplier carries full responsibility for quality and lead time. The subcontractor is their internal subcontracting arrangement, not your risk.

⑤ Blanket tolerances: ±0.01 on every dimension. An engineer, "to be safe", marks the whole drawing tight; the quote doubles, and most of those dimensions are cosmetic or non-mating. → Separate functional from non-functional faces and put non-functional ones on GB/T 1804 general tolerances. The cost drops immediately.

⑥ No change control: production to an old revision. A new drawing revision is sent, the supplier has no revision control, the line keeps running the old program, and by the time it is noticed several hundred pieces are made — scrapped in bulk with liability in dispute. → Put revision numbers and dates on drawings, get written sign-back from the supplier, and route every revision through a documented change review.

7. Before you send it out: a complete RFQ package (11 items)

In our records, the two most common gaps in incoming RFQs are: no 3D model, or no film thickness and colour code for the finishing. Miss either one and the shop has to guess — and every guess comes back later as an "additional charge".

Unsure about material selection? See our materials section: aluminium, stainless steel, titanium, brass, PEEK and carbon fibre, all sourced from major mills and traceable.

The single biggest saving: mark the critical dimensions separately and leave the rest at general tolerance. It is the fastest way to cut a quotation, and it costs no function at all — you simply stop paying for "looking tight".

8. Four clauses that must be in the contract

However good the work before, without these four clauses you still have no leverage when something goes wrong.

1. Quality standard clause. Turn the technical requirements of the drawing into contract terms instead of a vague "acceptance per drawing". State the rejection criteria (which drawing revision, which tolerance grade), who decides, who pays for rework, and how lead time is extended after rework.

2. Lead time and penalty clause. ① Start point defined as "drawing frozen + production slot locked", not the enquiry date or deposit date; ② how delay penalties are calculated and capped; ③ a weekly progress report system — penalties are an outcome constraint, progress reports are the process control.

3. Subcontractor liability clause. Delay and quality problems in outsourced operations (finishing, heat treatment) are the main supplier's full responsibility; if a subcontractor overruns by X days, an alternative channel must be activated at their own cost. Without this clause, on the bad day you will hear "that is the subcontractor's problem".

4. Tooling and drawing ownership clause. Fixtures and dedicated tooling belong to the buyer (even if designed and built by the supplier); agree the NDA scope, drawing retention period, and how drawings are destroyed or returned at project end. Especially important for build-to-print work.

9. Five cases where we suggest you do not work with us

This section may send some readers away, and we are still writing it. Long-term cooperation requires fit, not just a signed order.

① One piece, needed today. Precision machining needs stable fixturing, necessary measurement and a sensible cycle. Twenty minutes of stable fixturing and fifteen minutes of first-article measurement cannot be skipped. Anything delivered at the absolute limit will be "roughly right". If you need it today, a local job shop where you can wait on site is a better fit.

② Very small annual spend but extremely lead-time sensitive. The fixed cost of logistics, communication and scheduling coordination eats the price difference. A local supplier in the Pearl River or Yangtze River Delta you can drive to is cheaper overall.

③ Projects requiring a resident quality engineer on site. We are a 45-person precision machining shop with no capacity for resident staff. Such projects should go to larger suppliers with a resident service system.

④ Materials under export control or special compliance that we are not qualified for. Medical implant grade, aerospace grade and defence-related materials need full qualifications and a complete traceability chain. Not having them means not having them; taking the job anyway would be irresponsible to you.

⑤ Drawings still changing frequently, not frozen. Making samples at this stage wastes your money. Freeze at least one revision before sending it out; DFM advice can help, but it cannot replace design finalisation.

We are not writing these five to look honest. When we take on work we cannot do well, it usually ends badly for both sides: you lose the schedule, we lose production slots.

For those five cases we will tell you directly who is a better fit. If you are not in any of them, send drawings whenever you like (even an unfrozen revision) — start with a free DFM analysis. No order required, and no need to decide on a quotation first.

Frequently asked questions

In closing

Back to the opening question: how to choose a CNC machining supplier. The answer is short — find a shop that will still give you serious DFM feedback on a 10-piece order. A shop that does that almost certainly has frozen processes and an inspection system. The reverse does not hold: a shop that only pays attention at 5,000 pieces will most likely squeeze your 10 pieces into a gap in the schedule.

Four things you can take away as they are: the 7-step method (every question is quotable), the weighted scorecard, the three-stage gate list, and the four contract clauses. The last one is skipped most often, and it is the one that leaves you speechless on the bad day.

Get the selection right and what you save is not the machining fee — it is rework, line stops, and the three months spent finding another supplier.