Two quotes for the same drawing often differ by more than the machine hour rate, because the drawing itself decides how much tolerance, setup and inspection the shop must buy. A ±0.05 mm callout and a ±0.2 mm callout on the same 20 mm feature can look equally reasonable to a designer, yet they imply very different process capability, fixturing, gauging time and yield. This guide covers the decisions that actually move tight CNC tolerance cost, so a buyer can see which tolerances are paying for function and which are paying for nothing.
Start With Functional Need, Not With a Habit
A tolerance is a permission slip: it states how much variation the assembly can absorb before something stops working. When that reasoning is written down, many features turn out to be looser than the drawing implies. When it is not, tight values get copied from a reference part or a template nobody has revisited.
Ask what the feature does (a sealing groove, bearing seat and locating dowel hole have very different sensitivity to variation), what happens if it drifts (a gasket gap or a connector that will not seat is a function; “it looks more professional” is a preference), and who consumes the tolerance, since in a stack-up several parts share one budget. A useful discipline is to label each tolerance functional (the assembly fails outside it), process-driven (needed for a later operation or datum) or convenience (a default never challenged). Convenience tolerances are the first candidates for loosening; functional ones are the last.
Process Capability Decides Whether a Tolerance Is Cheap or Expensive
A tolerance is only achievable if the process distribution fits inside it with room to spare. NIST’s engineering statistics handbook defines process capability as a comparison of an in-control process output against specification limits, using indices such as Cp and Cpk built on six process standard deviations. Two details matter for sourcing:
- The distribution must be stable and roughly normal for the indices to behave as intended. A process that drifts from tool wear, thermal growth or inconsistent blanks is not in statistical control, whatever a single sample suggests.
- NIST notes that capability estimates are generally only meaningful with a reasonably large sample, cited at about 50 independent values. A short first-article run proves a feature was achieved once, not that it will hold over 5,000 pieces.
This is where tolerance cost originates. As a tolerance shrinks toward the natural spread of a process, scrap and rework rise sharply because the distribution tails cross the limits. The shop has three responses: buy a better process (finer machine, temperature control, better tooling), add process controls (probing, in-process gauging, more frequent offsets), or accept a higher reject rate and price it in. All three are legitimate; all three cost money.
The relationship is not linear. Going from ±0.2 mm to ±0.1 mm on a turned diameter is often routine on a modern CNC lathe. Going from ±0.02 mm to ±0.01 mm on the same diameter can require a different machine class, a controlled environment, different clamping or a secondary grinding or lapping operation. The cost curve steepens where the requirement crosses the capability boundary of a process family, and that boundary depends on feature size, length, stiffness and material, not on a number that applies to every shop.
General Tolerances Set the Baseline Before Any Tight Callout
ISO 2768-1 defines general tolerances for dimensions that carry no individual tolerance, selected by class letter: f (fine), m (medium), c (coarse) and v (very coarse). For linear dimensions, the medium class permits ±0.1 mm up to 6 mm basic size, ±0.2 mm from over 6 to 30 mm, ±0.3 mm from over 30 to 120 mm and ±0.5 mm from over 120 to 400 mm, with the fine class roughly a third to a half of that in the smaller ranges. The tables, as reproduced by reference sources such as AmesWeb, are in millimetres and apply only when the drawing invokes the standard near the title block.
A drawing marked ISO 2768-m already tolerances every untoleranced dimension at a level most general machining holds without special measures, while an individually toleranced dimension is a deliberate instruction. Ten tight callouts scattered across a part often force more inspection than one tight callout on the feature that matters.
Setup Dependence: The Same Tolerance Costs More in a Bad Orientation
Tolerance is not only about how precisely a tool cuts; it is about how many times the part must be re-referenced. A tight tolerance between two features machined in one setup is a machine accuracy problem. The same tolerance between a face from setup one and a bore from setup two becomes a fixturing, re-clamping and alignment problem, with errors adding.
This is setup dependence. Each setup brings its own locator wear, clamping deflection and operator judgment. Same setup and same axis is usually the cheapest tight tolerance; same setup on a different axis adds indexing error; a different setup adds fixture repeatability and datum transfer error; a different machine or supplier adds handling, re-datuming and a full first-article cycle.
Design changes that reduce setup dependence are often cheaper than tightening manufacturing. One primary datum, critical features reachable in one orientation, or a through-hole instead of a blind one that forces a flip can remove an operation. See CNC fixture datum strategy.
Inspection Burden: Tight Tolerances Are Paid For Twice
Tight tolerances are paid for at the spindle and again at the inspection bench. Features needing a CMM program, a dedicated gauge or temperature-controlled measurement add cost that never appears in a machining time estimate.
ISO 14253-1 makes this explicit commercially. Its decision rules require measurement uncertainty to be taken into account when proving conformance, so near a specification limit there is a band in which conformity cannot be proven. Nielsen’s explanation of the rules notes that tolerances are effectively reduced by measurement uncertainty when a supplier must prove conformance, and expanded by it when a buyer must prove non-conformance. A shop with ±0.005 mm uncertainty must therefore hold a ±0.010 mm tolerance with less provable margin than a ±0.015 mm one.
Buyers who specify tight tolerance without a measurement method or uncertainty requirement leave the supplier to protect itself, usually by choosing a more conservative process or adding inspection. Agreeing the gauging approach, acceptance rule and sampling level in advance makes the price more predictable. Inspection cost also scales with the number of toleranced features: 30 toleranced dimensions may need a full CMM program, while eight plus a general tolerance note may be covered by a few critical checks. See inspection sampling plans for CNC parts.
Where the Cost Actually Accumulates
Academic work supports treating tolerance values as an economic decision. A 2024 study in The International Journal of Advanced Manufacturing Technology compares optimised tolerance allocation with three heuristics (equal tolerances, a precision factor, and proportional to nominal size) and proposes a procedure to estimate the cost reduction. Its premise is telling: tolerance allocation strongly influences manufacturing choices, yet in practice is often done by heuristic rather than analysis.
For a buyer, the cheapest drawing is not the one with loose tolerances everywhere. It is the one where every tightened tolerance is justified by function, achievable without special measures, reachable in a reasonable number of setups and verifiable by an agreed method. Typical cost contributors:
- Extra operations added to reach a tolerance (grinding, honing, lapping, reaming), plus handling between them.
- Yield loss when a process distribution sits close to the limits.
- Inspection time, including programming, fixturing and reporting.
- Setup and fixture cost, which amortises poorly on small batches.
- Process control such as probing, in-process gauging and temperature control.
The first two interact: a tolerance may be cheaper via a secondary operation than via a marginal single-operation process, so the same callout can receive very different quotes from shops with different equipment.
Cost-Reduction Questions to Bring to the Next RFQ
Ask these before the quote is finalised. They shift the conversation from “can you do it cheaper” to “which requirement is driving the price”.
- Which three features dominate the price, and why? An estimator can usually name them; “all of them” means the drawing needs review.
- What general tolerance class applies, and does every individually toleranced dimension need to be? ISO 2768-m plus a handful of critical features often removes cost with no functional loss.
- Can the critical features be machined in one setup? If not, can a datum change or better tool access fix it?
- What measurement method and uncertainty will prove conformance, and must the drawing state an acceptance rule?
- Is this tolerance functional or inherited from a template? If nobody can name the failure it prevents, review it.
- Does batch size justify the fixture and inspection investment, or would a stock size or blank change reduce the per-part burden?
Related design-level reductions are covered in DFM for CNC machining and CNC machining drawing requirements, and datum conventions in datum schemes for CNC parts.
Common Mistakes That Inflate Tight Tolerance Cost
Tightening everything to be safe. Blanket tight tolerances transfer all risk to the supplier, who prices it back in. Selective tightening is cheaper and technically sound.
Ignoring the measurement side. A tolerance that cannot be proven with available gauging is not a useful specification; state the method and uncertainty expectation.
Late drawing changes. A tolerance change after quoting can invalidate the process plan, fixture and inspection program; see revision control for CNC parts.
Conclusion
Tight CNC tolerance cost is not a single number. It is the sum of process capability limits, setup and fixturing decisions, inspection and measurement-uncertainty requirements and yield effects, each depending on the feature, material, batch size and equipment. The tabulated values in ISO 2768-1 and the definitions in NIST’s handbook give a shared vocabulary, but they do not replace a process-specific capability assessment. The reliable way to control cost is to make the drawing say what it means: a general tolerance class, tight callouts only where function requires them, and an agreed inspection method.
Sources and Scope Notes
The ISO 2768-1 values quoted here follow a secondary tabulation and apply only when the standard is invoked on the drawing; the standard itself is authoritative, including angular tolerances and ISO 2768-2. Process capability definitions and the sample-size note follow NIST’s handbook and describe statistical properties, not the capability of any particular machine or supplier. The ISO 14253-1 discussion summarises a published conference paper and does not reproduce the standard. The tolerance-cost observations are qualitative and drawn from the cited journal article’s framing of tolerance allocation as an economic decision; no universal price ratios are claimed.
