Most make-or-buy decisions I have sat through were priced on unit price, spare capacity and capital avoidance. All three are legitimate, and all three ignore the variable that decides whether the decision ages well: how much control over the process you hand to an organisation whose incentives differ from yours. A machining cell in your own plant reports drift, tool wear and start-up scrap to you every shift. A supplier reports delivered parts.
The purchase order transfers the activity, never the accountability. When a plated fastener fractures in the field or a casting weeps at a customer's test bench, the complaint arrives at your door carrying your part number. You may recover something commercially through the contract; operationally you still own the containment, the root-cause work and the customer conversation that follows.
So the quality question is not simply whether they can make the part — most suppliers can, eventually — but how much visibility and influence you retain once the paperwork is signed. This article works through that question: how to assess capability before committing, where control actually leaks, what escapes genuinely cost, and which levers to hold back.
Capability Assessment on the Process That Will Make the Part
Capability evidence only means something when it comes from the serial process: production tooling, production fixtures, production operators, production cycle rate and production gauges. A toolroom sample or a soft-tooled prototype proves geometry and little else. Insist on a run at rate, then a dimensional layout on consecutive parts from that run, reported feature by feature on the critical characteristics rather than averaged into a single index across the whole drawing.
Before you believe any of those numbers, interrogate the measurement side. The long-standing expectation is that gauge uncertainty consumes only a small fraction of the tolerance; if the supplier's hand micrometer is taking readings on a bore with a few microns of total tolerance, the capability figure is fiction. Ask where the CMM actually sits — a lab held at twenty degrees Celsius, or a bay that swings with the seasons. That matters enormously when the parts are aluminium.
Special processes demand different evidence entirely. Heat treatment capability means furnace surveys and load patterns, not one hardness reading on a coupon somebody selected for you. Welding capability means a qualified procedure, a qualified welder and parameter records you can read. Plating capability means bath chemistry logs and hydrogen-embrittlement bake records for high-strength steels — delayed fracture from embrittlement is a classic escape that no incoming inspection will ever catch.
Ask about the edges of the process, too, because that is where scrap and escapes live: first-off and last-off practice, parts made immediately after a tool change, the first batch following a weekend idle, parts run while the coolant or bath was being topped up. A supplier who cannot describe their own start-up discipline will be discovering your part's failure modes on your production time.
Where Control Quietly Drains Away
The largest single leak is change control. Nothing on a standard drawing forbids the supplier from switching insert grade, raising cutting speed to rescue a delivery, changing coolant, rearranging the heat-treat load, or moving plating to a cheaper sub-tier. Each of those changes can shift a failure mode — tool wear rates, residual stress, hydrogen pickup, coating thickness — without moving a single dimension on the part.
Below your supplier sits a chain you rarely see. Your contract binds the company you pay; it does not bind the foundry, the plating house or the heat treater they use. In my experience few tier-one suppliers can list their own sub-tiers for a given part without a phone call. Traceability fragments accordingly: a lot number on a certificate that maps to nothing you can interrogate once a failure surfaces two years into production.
Information asymmetry does quieter damage. As the buyer you receive only the parts that passed; the supplier's internal filtering removes exactly the evidence of drift you would have acted on in-house. Sampling plans at goods-in are built to catch gross shifts and contamination, not gradual movement, and by the time drift becomes visible at receiving it has usually been running for weeks upstream.
The visibility gap between in-house and outsourced
In-house process
- Drift and tool wear visible every shift
- Start-up scrap counted and reviewed daily
- Change control governed by your own engineering
- Process knowledge on your payroll
Outsourced process
- Only conforming parts arrive at goods-in
- Internal filtering hides drift for weeks
- Changes made silently to rescue delivery
- Sub-tier process opaque to both parties
Knowledge leaves your building at the same time. Ten years after outsourcing a machined housing, nobody on your staff has run the process, so you cannot challenge a supplier's root-cause claim, cannot judge whether their cycle-time story is honest, and cannot bring the work home quickly if the relationship sours. Control loss compounds: every passing year makes both effective oversight and insourcing harder.
The Real Bill for Escapes
Unit price conceals the tail of the distribution. When a defective lot gets out, the immediate costs are sorting at your line or the customer's, containment stock, premium air freight and, at worst, a stopped line. Behind those sit the slower ones: warranty returns whose root cause is buried two tiers down, corrective-action loops that repeat because they address symptoms, and your own engineers spending weeks doing the supplier's problem-solving for them.
Few organisations book these costs against the part that caused them. Audit travel, scorecard administration, surveillance time and containment paperwork all land in overhead, which is how a part can look cheap on the quote comparison yet be expensive in truth. If you want the comparison to mean anything, charge the cost of managing a supplier against that supplier's part number.
The asymmetry deserves respect. What a failure costs scales with where the part sits in the product, not with what the part cost: a low-value seal can write off an engine assembly, and a mis-plated clip can trigger a field campaign. When I weigh a buy decision, I care less about the part's price than about the value of the hardware downstream of it and the number of places it is installed.
Model the tails rather than averaging them. A simple scenario exercise — one escape per year landing at the worst possible point, one process change slipping through unnotified, one sub-tier quietly disappearing — tells you more about true exposure than any blended figure. Run the same scenarios for the in-house option, because internal scrap is visible daily and flatters nothing, while external escapes get discounted simply because they are rare and late.
A supplier reports delivered parts. The complaint, two years later, arrives carrying your part number.
What You Should Never Sign Away
Accountability is the first thing you cannot delegate, and verification of safety and statutory characteristics is the second. Decide explicitly, characteristic by characteristic, who checks what and with which evidence: the supplier at source with records per lot, you at incoming, or both. For characteristics whose failure could endanger people, keep a verification point under your own control regardless of who cuts the metal.
Contract language becomes your control system once the work leaves. Require written notification and approval before any change to process, material, tooling location or sub-tier; right of access for process audits extending to sub-tier sites; lot-level process data delivered with each batch; containment obligations with a clock on them; and recognised accreditation for special processes such as welding, heat treatment and non-destructive testing — verified, not assumed from a certificate on file.
Tooling and data ownership decide whether the decision is reversible. If you own the tools, the fixtures and the current process documentation, a transfer to another source is a project; if the supplier owns them, it is a negotiation you will lose. The same logic applies to records: capability studies, layout reports and process data generated on your parts should be contractually yours, because they are the passport to a second source or a move back in-house.
Black-box supply deserves its own caution. When the supplier designs the part as well as making it, you have delegated the engineering too — design changes you cannot technically evaluate, validation runs you did not witness, failure modes nobody on your side has ever considered. For critical functions, keep design authority internal and buy the manufacturing, not the thinking.
The control layers a buy decision must preserve
- Design authorityKeep internal for critical functions; buy the manufacturing, not the thinking
- Verification of safety characteristicsA control point under your own roof, characteristic by characteristic
- Change and sub-tier approvalWritten notification and sign-off before any process, material or location change
- Tooling and data ownershipMakes the decision reversible; without it, a transfer becomes a lost negotiation
- Delivered partsAll the buyer sees by default — everything above must be built deliberately
Making the Buy Option Survivable
If the decision is to buy, staff it like an engineering function rather than a mailbox. Set surveillance cadence by risk — a new part, a new process or a new site gets frequent presence, while a mature part on a proven line gets trend review — and spend audit days watching the process run, not admiring the quality manual the evening before.
Build your own early-warning system instead of relying on goods-in. Review the lot data the contract obliges the supplier to send: tool-wear trends, parameter logs, hardness across the batch. Choose incoming checks for the features that actually drift, and verify your own gauge capability first — there is no irony quite like rejecting good parts, or accepting bad ones, because receiving inspection cannot measure the tolerance it is policing.
Plan the escape before it happens. Agree who contains, who sorts and who pays, in days rather than weeks; keep traceability granularity fine enough to bound the exposure, because a recall scoped by heat number is a different animal from one scoped by month of manufacture. When a failure surfaces in the field, lot definition is the single biggest lever on the size of the bill.
Finally, treat the decision as a standing review rather than a one-off. Capability decays, ownership changes, volumes move, and a supplier's competence on your part today says little about next year. Revisit the case at every design change and after every serious escape, and keep the insourcing route warm: current drawings, a documented process, an honest internal cost benchmark. The credible option to bring work home is itself a form of control — often the only one a supplier truly respects.
Deciding on the Right Axes
A make-or-buy case argued only on unit price, capacity and capital is a case argued on the axes that ignore consequence. The durable question is control: how much process visibility you keep, how fast you learn of drift, how binding your approval is over changes you never see, and how quickly you could reverse the decision if you had to. Price the part, then price the control you are giving up.
The method fits within existing systems rather than replacing them. A PPAP submission built on a genuine run at rate, a PFMEA that names the sub-tier processes, an AS9100 or IATF 16949 audit spent on the floor instead of the manual — each is a hook the standards already give you. The discipline is using them to defend the process, not just to approve the supplier.
Across two decades in automotive and aerospace, I have seen the same pattern: the organisations that outsource well are the ones that behave as if the purchase order changed nothing about their accountability — only their means of discharging it. The ones that outsource badly treated the signature as the moment responsibility left the building. It never does.
Keep the checklist short and hard: capability proven on the serial process, changes gated by contract, sub-tiers visible, escapes priced into the comparison, verification retained where failure harms people, and the route home kept open. That is the control that does not leave with the purchase order — because it was never in the purchase order to begin with.