Inspection without stop authority isn't quality control. It's surveillance.

A vendor pitched me zero-downtime AI inspection last quarter. The demo was polished — high-speed cameras, real-time inference at 200 parts per minute, dashboards that made defect rates look like a stock ticker. I asked one question: can it stop the line? Silence. The technology could detect. That wasn't the gap. The architecture had no stop authority wired in. No PLC interlock, no escalation ladder, no mechanical hold. The pitch was about uninterrupted production, and that is precisely the problem. In two decades of running automotive and aerospace plants, I have never solved a quality problem by making production harder to interrupt.

Jidoka isn't decor — it's the stop authority your PFMEA assumes exists

Toyota built Jidoka around one principle: the operator — or the machine — that detects an abnormality must have the authority to stop. This is not a cultural nicety. It is the mechanism. Remove it and your entire quality architecture collapses, because every other tool, from 8D to PFMEA, was designed under the assumption that detection triggers response. Immediately.

When I built the QA/QC department at SNOP's 900+ employee greenfield plant — new lines, new people, zero institutional memory — the first culture battle was stop authority. Operators didn't want to stop the line. Supervisors didn't want them to either. The unspoken rule was: keep running, we'll sort it later. Later is where defects live.

We installed QRQC with hard trigger thresholds. Not guidelines. Thresholds. If a critical dimension drifted beyond control limits on three consecutive parts, the line stopped. Not after the shift. Not after the stand-up. Then. The pushback was predictable — false stops, lost throughput, operator anxiety. We ran the numbers anyway. The cost of a stop was a fraction of the cost of what it contained. We finished that year with 70% defect-cost reduction and 98% customer satisfaction. Not by making the line harder to interrupt — by making interruption structured.

Your PFMEA assumes this authority exists. Every risk priority number in that spreadsheet is calculated against a detection-escalation-response chain. Sever the response link — let detection run but never trigger containment — and the PFMEA becomes decorative.

The containment gap: faster data, slower response

Here is what happens when you bolt AI vision onto a line with no stop authority. The system flags defects at 200 parts per minute. Excellent. Where do those flags go? Into a dashboard. A database. A report reviewed at the next quality meeting — two shifts later, after 96,000 suspect parts have moved downstream into assembly, into logistics, into a customer's incoming inspection.

This is the containment gap: the distance between detection and response, measured in parts produced. AI vision without stop authority doesn't close that gap. It widens it, because you now possess more granular data about defects you still cannot contain in time. You haven't improved quality. You've improved documentation.

An inspection system that cannot stop the line isn't protecting your customer. It's documenting your failure in higher resolution.

Ford learned this publicly. They invested heavily in AI quality control, those systems fell short, and they rehired over 300 veteran engineers to fix what the algorithms couldn't hold. The lesson isn't that AI is useless. AI without operational authority is an expensive camera with opinions. Ford is also leading the industry in recalls this year — north of 770,000 vehicles — which tells you what happens when detection and containment aren't the same system.

The AI inspection market is projected to reach $6.29 billion by 2034. Inspection vendors now market "zero-downtime" as the headline feature. Almost none of that spend includes a line-stop relay or an escalation protocol. They sell throughput because throughput is what procurement teams want to hear. Nobody buys a system that stops their line. That is precisely what a quality system does.

Building inspection with teeth

At Airbus, the 97% reduction in internal lead time came from Routing Verification KPIs — and here is the detail that matters: those KPIs deliberately build checkpoint authority into the routing. Every verification gate has the power to hold parts until the condition is satisfied. The speed didn't come from removing checkpoints. It came from making checkpoints effective enough that the line didn't need rework loops, sort campaigns, or emergency containment teams downstream. Authority creates speed. The absence of authority creates rework.

Same principle applies to AI inspection. If you want it to generate value rather than data, give it teeth. Wire the detection output to the PLC so that threshold breaches trigger a fault code — not an email. Define QRQC trigger logic: consecutive defects, pattern detection, severity bands. The system doesn't need to be perfect. It needs to be authoritative within its defined parameters.

Set false-stop economics before deployment, not after. A €12,000 line stop from a false positive is annoying. A €120,000 customer escape is a crisis. The math favours the stop, every time. Give the escalation ladder a clock — if the system flags a defect and the response team doesn't engage within a defined window, the line stops. No waiting for the shift leader's permission.

False stops are the vendor's favourite scare story. What if it stops the line on a false positive? Yes. What if it does? You investigate, you reset, you lose twelve minutes. Compare that to the alternative: a defect escapes, reaches a customer, triggers a warranty claim, an audit finding, or a recall campaign that costs more than your entire inspection budget. The asymmetry is absurd. Yet procurement teams fixate on the false stop.

The best inspection system isn't the one that never stops your line. It's the one that stops it at exactly the right moment — and has the authority to hold it there until containment is real.