A mid-size automotive supplier installed an inline inspection system capable of detecting dimensional deviations down to two microns. The system featured real-time dashboards, automated email alerts, SMS notifications to shift supervisors, and a status tower visible from every workstation. In its first month it caught eleven deviations that would otherwise have escaped to the customer.
Eighteen months later, the same system detected a gradual drift in a critical bore diameter and sent forty-seven alerts over a three-week period. Not a single alert was acted upon. The result was a contained shipment of 12,000 housings, an eleven-hour customer line stoppage, and a corrective action request that required the CEO to fly to the customer's headquarters.
When the investigation team reconstructed the timeline, they found that the system had worked exactly as designed. The sensors were calibrated, the software was current, the thresholds were appropriate, and all forty-seven alerts had fired correctly. The technology performed flawlessly. The organisation had simply stopped listening.
This is Quality Feedback Decay: the progressive loss of organisational sensitivity to signals that were once meaningful. It is not a failure of measurement. It is a failure of human attention, and it destroys more quality systems than any single defect ever could. I have audited plants where six-figure inspection equipment functioned as expensive wall decoration because the alert architecture actively trained operators to ignore it.
The Mechanism of Alert Habituation
When a new alert system launches, every signal carries urgency. Supervisors respond within minutes, quality engineers investigate every anomaly, and management reviews daily reports. This hyper-responsiveness is driven by novelty and adrenaline, not by sustainable process. The organisation is treating every blip with equal intensity, which means it cannot distinguish between a genuine crisis and normal process variation.
Within two to four months, the team begins to calibrate. Operators learn which alerts indicate real problems and which are false positives. They start making contextual judgment calls: that alert always fires when running material from Supplier B, so it can be safely dismissed. This stage is genuinely useful — the organisation is developing pattern recognition and prioritising its attention.
The problem is that this healthy calibration rarely stabilises. Pattern recognition hardens into untested assumptions. The nuanced observation that an alert correlates with a specific supplier becomes the blanket conviction that the alert means nothing. The subtle drift from contextual judgment to categorical dismissal is almost invisible, and it sets the stage for systematic signal failure.
Four Drivers of Systemic Decay
The first driver is poor signal calibration. If a system generates alerts that frequently require no action, personnel rationally learn that the alerts are unreliable. When thresholds are too sensitive or criteria too broad, the system trains people to ignore it. The fix is not reduced sensitivity — it is tiered response architecture that separates advisory information from critical alarms.
The second driver is the no-consequence loop. When a signal fires and nothing bad happens — whether because it was a false positive or because someone caught the problem manually — the organisation learns that ignoring signals carries no cost. Every ignored alert that fails to produce a disaster reinforces the assumption that responses are optional. The organisation is confusing luck with adequacy.
The third driver is the context gap. Alerts typically arrive as a number, a colour, or a sound, stripped of meaning. The recipient does not know what changed, why it matters, or what action is expected. When people cannot form an intelligent response, they stop responding entirely. Effective signals are narratives, not naked data points: they specify the deviation, the timeline, the likely cause, and the recommended action.
The fourth driver is the ownership vacuum. If responsibility for responding to an alert rests with a shift supervisor who is simultaneously managing a personnel issue, a material shortage, and a production target — and the system has no escalation mechanism — the alert dies on arrival. Accountability must be built into the workflow, not layered on top as an afterthought.
The Consequence of Signal Death
By the time feedback decay reaches its final stage, the system is technically alive but functionally dead. Alerts fire into the void, dashboards update on walls nobody monitors, and the real-time responsiveness that justified the investment is gone. Operators glance at the red tower light the way they glance at a clock — registering it visually without processing its meaning.
The hidden cost is that this degradation is invisible on traditional quality metrics. Defect rates may remain stable because containment is happening downstream. Customer complaints may not spike immediately. Audit results may look acceptable because auditors check whether the system functions technically, not whether the organisation still responds to it behaviourally.
The most sophisticated detection system in the world is worthless if the organisation has learned to tune it out.
What is happening beneath the surface is the erosion of early warning capability. The organisation is no longer catching drift before it becomes deviation. It has shifted from a proactive posture — preventing problems — to a reactive one — discovering them after they escape. It has lost its quality immune system, and it will remain defenceless against the next process drift until that system is rebuilt.
Designing Alerts That Resist Decay
Preventing feedback decay requires structural design choices, not exhortations to be more disciplined. The first principle is ruthless calibration. Every alert should be earned. If a system generates 200 alerts per shift, it creates 200 opportunities for the organisation to learn that alerts are noise. If fewer than 70% of alerts lead to a documented response, the thresholds are wrong and must be tightened.
The second principle is visible loop closure. Every alert must have a documented resolution that is displayed alongside the alert itself. When a signal fires and someone responds, that response is recorded and shown on the same dashboard. This creates public accountability and reinforces the causal link between signal and action. Instances where an alert prevented a real defect should be highlighted, not buried.
Alert Architecture: Fragile vs Resilient
Fragile System
- Binary alerts with no severity tiers — every signal triggers the same alarm
- Naked data points stripped of context, trend, or recommended action
- No visible loop closure — alerts disappear without documented response
- Static delivery — same tone, same screen, same light until it becomes furniture
Resilient System
- Graduated levels — informational, advisory, and critical with distinct response rules
- Narrative alerts — deviation, timeline, likely cause, and assigned owner
- Signal-to-response tracking displayed publicly on the alert dashboard
- Thresholds reviewed quarterly against the actionable-to-noise ratio
Building Context Into Every Signal
An alert that arrives as a number without context is noise. An alert that arrives as a decision is much harder to ignore. Every signal should carry five elements: what changed, when it started, what it likely means, what action is recommended, and who is accountable for taking it. This transforms a beep into a workflow trigger.
Consider the difference. A bare alert reads: Bore diameter exceeded. A contextual alert reads: Bore diameter on Station 7 has drifted +3 microns over the last 200 parts, correlating with Tool Change 4 on the previous shift. The second version gives the operator a hypothesis to test and a specific action to take. It respects the recipient's time and intelligence.
This level of context requires investment in the alert logic itself — correlating SPC data with maintenance logs, tool-change records, and material batches. The technical effort is modest. The payoff is an order of magnitude reduction in alerts that are dismissed without investigation, because each alert now arrives pre-loaded with the information needed to act on it.
Auditing Behavioural Responsiveness
Most internal audit programs check whether quality systems function technically: Are sensors calibrated? Is software current? Are thresholds documented? These questions are necessary but insufficient. They miss the failure mode that actually destroys detection capability — the behavioural decay that renders a functioning system inert.
Feedback Decay Audit Cycle
- 01Extract response metricsPull alert-to-action ratio, mean response time, and auto-filtered alert count for the quarter.
- 02Test signal visibilityInterview operators and supervisors unannounced: what was the last alert, and what did you do?
- 03Review threshold adequacyFlag any alert category where fewer than 70% of signals triggered a documented response.
- 04Adjust and rotate deliveryChange alert tones, dashboard layouts, or escalation paths to break perceptual habituation.
- 05Report to leadershipPresent behavioural responsiveness data alongside Cpk and OEE in the monthly quality review.
A behavioural audit asks different questions. Are response times increasing quarter over quarter? Is the alert-to-action ratio declining? Have operators stopped noticing the visual signals? When was the last time an alert actually prevented a defect rather than merely confirming one that had already escaped? These questions reveal whether the organisation is still listening.
I introduced this audit methodology alongside Routing Verification KPIs at an aerospace manufacturer, and the combination cut internal lead time dramatically. The routing KPIs measured process efficiency; the responsiveness audit measured whether the process intelligence was actually being used. Together they exposed gaps that traditional ISO 9001 and AS9100 surveillance audits consistently missed — because those audits check system design, not system behaviour.
Add feedback decay metrics to your management review agenda. Track them with the same discipline you apply to Cpk, OEE, and on-time delivery. Response latency is a leading indicator of quality failure. If your mean alert response time is trending upward, a customer escape is already building — and your detection system will not save you, because the organisation has already stopped hearing it.
Attention as a Managed Resource
In two decades of quality leadership across automotive and aerospace, I have seen organisations spend heavily on detection technology and almost nothing on maintaining the human responsiveness that gives that technology value. The pattern is consistent: a new system launches with enthusiasm, delivers early wins, and then decays into background noise within twelve to eighteen months.
The best quality systems I have worked with do not have the most sensors or the most sophisticated analytics software. They have alert architectures designed around how humans actually process information over time. They tier their signals, rotate their delivery methods, close every loop visibly, and audit behavioural responsiveness with the same rigour they apply to gauge calibration.
Attention is a depleting resource. Manage it with the same structural discipline you apply to any other critical input to your quality system. Calibrate thresholds quarterly, build context into every signal, assign operational ownership, and audit whether the organisation is still responding — not just whether the equipment is still running. Your quality system is only as strong as the last person who acted on what it told them.