The work comes home. The process knowledge didn't.

Here is an uncomfortable observation: in the space of one week, Honeywell Aerospace has said publicly that it is rethinking its outsourcing strategy and may pull work back in-house, while GE Aerospace has taken pains to insist its own deal is not a blueprint for broader integration. Two primes, one direction of travel. De-integration is now boardroom strategy in aerospace. Every analyst note I have read frames it as a capacity and supply-chain story – floors, machines, resilience. Not one asks the question that decides whether any of it works: who in your plant still knows how to run this process?

I have stood at both ends of the transfer. At SNOP I built the QA/QC department for a 900-plus employee greenfield plant from bare concrete, and that build gave me a price list for everything an insourcing plan assumes already exists. Before that I sat on the supplier side – at WITTE Automotive, and later through FOREAST, my consulting agency, running VDA 6.3 process audits at companies that had absorbed exactly the work OEMs no longer wanted. Both seats teach the same lesson from opposite directions. The work moves easily. The knowledge does not.

What outsourcing actually exported

The purchase order said work transfer. What actually left the building was a knowledge stack that took twenty years to build. Capability baselines expire. A Cpk study from 2015 describes a process that no longer exists: tooling has worn, furnaces have drifted, and somewhere along the way the supplier changed coolant, cycle time and torque values without telling anyone. The PFMEA you once owned has either been rewritten by engineers you never met or not touched at all. You will not know which until something fails.

VDA 6.3 audits taught me where that knowledge actually lives. I have opened machine cabinets and found the "real" setup sheet taped inside the door – handwritten, undated, no resemblance to the controlled document in the quality system. I have met the operator who could hear a spindle bearing going before the vibration sensor caught it. He retired. Nobody replaced him, because the process ran fine. That is the point. A supplier's plant runs on invisible compensations, and when the part comes home you inherit a decade of process drift with no baseline to measure it against.

A process you have not run in ten years is not dormant capability – it is somebody else's tribal knowledge wearing your part number.

Requalification is not paperwork, it is a cost stack

Building from bare concrete let me price what an insourcing business case treats as free. Before the first saleable part you need APQP from zero – process flow, PFMEA, control plan. Budget 200 to 400 engineer-hours per part family, and that assumes you can still find engineers who have run the process before.

Then the measurement side. Full MSA on every system, because the gauges that came back with the work arrived the way loaned tools always do: worn, modified, quietly out of calibration. Recalibration, new fixturing and replacement gauging for one family runs €40,000–80,000 before you measure a single part. Add capability studies and trial runs, with material scrapped at your expense while the process finds its feet – call it €30,000–80,000 per family. Then safe-launch containment: 100% inspection, extra inspectors per shift, sealed traceability for at least a quarter. First-article packages. Re-PPAP under AS9100 scrutiny. For a mid-complexity family, €250,000–500,000 of pure requalification cost before the part earns a cent. None of it appears in the relocation business case.

Treat every repatriated process like a new launch

The most expensive assumption in any insourcing plan is that day one looks like steady state. It never does. During early ramp at the greenfield plant we had an escape inside week six – a dimensional drift our supplier-era assumptions said could not happen, found by a customer, not by us. QRQC put containment on the dock within 24 hours. The A3 took three weeks to close, because the root cause sat in a parameter nobody had a baseline for. That single escape cost more than the control plan that would have prevented it.

So run repatriation the way you would run a launch, because that is what it is. Reset the PFMEA from a blank sheet – do not inherit the supplier's; you cannot audit its assumptions. Rebuild the control plan around your plant, not theirs. Apply first-article logic to every characteristic and expect customer re-PPAP on anything safety-relevant. And negotiate the data transfer with the supplier you are leaving before the final shipment: PFMEAs, control plans, calibration records, deviation history, the full 8D archive. They may hold the only valid process history in existence. The day after the last shipment, your leverage is gone.

The primes pulling work home are right that the old outsourcing ledger was incomplete – it priced the parts and never priced the knowledge. But the correction risks the same flaw in mirror image. Bring work back without funding requalification and you are not de-integrating; you are re-learning a process you once owned, one nonconformance at a time, with the customer as your detection system. Budget it like a launch. Staff it like a launch. Contain it like a launch. Anything less, and the first honest capability study you run will tell you exactly what the supplier knew all along.