A plant that finds defects only at end-of-line inspection usually finds them late, after the bad parts have shipped or been built into an assembly. That delay is often the gap between a bad quarter and a company-defining event. And when a recall does happen, weak traceability makes it more expensive, because a company that can't say which parts used the suspect material has to pull back every part that might have.
The cost is already bleeding before a recall ever happens
Quality-related costs (scrap, rework, warranty claims, inspection overhead) run as high as 15 to 20 percent of sales revenue in many organizations and up to 40 percent in some, according to ASQ. Scrap and rework alone cost the median manufacturer about 1% of sales in APQC’s benchmark of more than 1,000 organizations. Most of that isn't a single dramatic failure; it's the accumulated cost of not knowing something was wrong until the part had already moved three stations down the line.
Why end-of-line inspection structurally can't catch this
End-of-line inspection asks one question: is the finished part good or bad? It can't answer the question that actually prevents a recall: which specific batch of material, which welder, which procedure, produced this defect — and what else did that same combination touch? By the time a problem surfaces at final inspection, the traceable trail back to its root cause is already several stations old, and in most shops, it was never captured in the first place.
The bar is also rising from the standards side
AWS D1.1/D1.1M:2025 tightened NDT personnel certification requirements (Clause 8.14.6) and formally accepted digital radiography as an RT method — both point the same direction: inspection and its documentation are expected to be more rigorous and more digitally traceable, not less. A paper NDT log that satisfied an inspector in 2020 is a weaker defense in an audit today.
What inline traceability actually requires — not more inspectors, different capture
- Material certificates linked at intake, not retyped later. A filler-metal CMTR traceable to the lot actually consumed on a job is only useful if it's linked to the part before the part ships, not reconstructed from memory during an audit.
- Capture at the station, offline-capable. A welder recording who, what procedure, and which heat number at the end of a shift is recording memory, not fact. Shop floors rarely have reliable connectivity, so capture has to work offline and sync later — not require a live connection to be usable at all.
- Traceability in both directions. Forward, from a part number to its material certificate, welder, WPS, and NDT result. Backward, from a suspect material heat or a failed NDT result to every part it touched. The backward direction is what turns a potential fleet-wide recall into a contained, four-part correction.
The real test
If a suspect material certificate showed up tomorrow, could you list every part it touched in under an hour — or would that require phone calls, spreadsheets, and someone's memory of which project ran that week? That answer is the actual difference between correcting a handful of parts and recalling everything that might have touched that material.
This is exactly the gap Fidar MES closes for steel fabrication: BOM extracted straight from Tekla models, production data captured at the station and synced when the network allows, and full weld-to-part traceability in both directions — the audit trail an inspector or a claim actually needs, built as the work happens instead of reconstructed after.
The same heat and material records are what the EU's Digital Product Passport will ask steel supply chains for; we looked at what fabricators actually need to track. CBAM is the nearer deadline: it already covers fabricated steel structures, and EU importers will want mill and tonnage per shipment before the first paid declaration in September 2027 (see EU CBAM for Gulf steel fabricators).
Traceability only holds if every piece is identified at every station. Our guide to piece-mark barcode tracking shows how to do that without retyping marks.
The same piece-level data is what most AI projects on the shop floor end up missing, which we looked at in why over half of manufacturers aren't sure their systems are ready for AI.
