Most ROI models capture scrap at the point of detection. The defect is flagged. The part is rejected. The reject count goes into the denominator. But scrap cost in a high-volume production environment accumulates in three places, only one of which appears at the inspection station. The first is the visible scrap: parts that fail inspection and are removed from production. The second is the downstream scrap: parts that pass inspection but carry latent defects that produce field failures, warranty claims, and rework at the customer's facility. The third is the pattern accumulation cost: a defect that appears at a 2% rate for eight hours before a root cause is identified represents a different total loss than a defect that triggers an automatic process correction within minutes of first appearance. AI detection that only addresses the first category captures a fraction of the available value.

In most AI inspection deployments, the financial model assumes that detection equals action. The defect is flagged, the part is removed, the scrap cost is realized and contained. What the model does not account for is the human decision layer between detection and action. A defect flagged at 2 AM on the third shift requires a human to evaluate the alert, decide whether to halt production, escalate, or continue running -- and that decision takes time even in the best-case scenario. During that decision window, product continues to run. The accumulation of defective output between detection and response is invisible in most ROI models because it assumes detection is instantaneous and response is automatic. In alert-based systems, neither assumption holds.

A defect that first appears at 10 PM on a production run carrying a root cause traceable to a mold temperature deviation does not stay at 10 PM. If the inspection system flags the defect, generates an alert, and waits for human action, the mold temperature deviation continues for as long as the decision delay takes. In facilities with three-shift operations and supervisor handover gaps, a root cause that could have been corrected in minutes may accumulate scrap output across an entire run before a corrective action is confirmed. This accumulation cost -- the output produced between the first defect detection and the root cause correction -- is the largest line item in the true cost of a non-closed-loop quality system. It is also the line item most reliably absent from AI vendor ROI models.

A closed-loop quality AI system changes the P&L in three distinct ways that a detection-only system does not. First, scrap accumulation is bounded -- the system does not wait for human action to initiate a response. A defect pattern triggers automatic process adjustment or line halt within seconds of reaching a defined threshold. Second, root cause is captured at event time rather than reconstructed after the shift. The correlation between defect occurrence and upstream process variables -- material lot, temperature, cycle time, cavity identifier -- is built into the data record. Third, repeat defect cost is structurally lower. A defect that recurs because root cause was never identified at first occurrence does not recur when root cause is automatically diagnosed at every occurrence.

The Clause Most Procurement Teams Miss: Actuation Authority

The Clause Most Procurement Teams Miss: Actuation Authority



A closed-loop quality system requires that the AI layer have actuation authority -- the ability to signal the production line to change state without waiting for human confirmation on routine defects. Actuation authority is the specification requirement that most procurement teams fail to include in their AI vision evaluation criteria because it sounds like an integration detail rather than a capability requirement. It is not an integration detail. A system that detects defects and generates alerts is categorically different from a system that detects defects and closes the loop automatically. Only one of them delivers the ROI that appears in the original spreadsheet.

Detection accuracy is the metric that dominates AI quality system evaluation. It is not the metric that determines financial outcome. The unit of measurement that actually predicts ROI is cost per defect escaped: the total financial exposure from a defect that was detected but not acted upon in time to prevent downstream scrap, rework, or field failure. A system with 99% detection accuracy but a four-minute human response chain has a higher cost per defect escaped than a system with 97% detection accuracy and a five-second closed-loop response. When evaluating quality AI platforms, the accuracy specification should be a minimum threshold, not the primary decision criterion. The primary decision criterion is what happens in the seconds after detection.