The production monitoring deployment at Procter & Gamble and Maruti Suzuki, two of Jidoka Tech’s most referenced Nagare case studies, share a structural pattern that appears consistently in manufacturing AI monitoring implementations: the gap between what the production team believed they knew about their floor and what the monitoring data actually showed was significant enough to change the improvement priorities.
This is not a criticism of production teams. It reflects the inherent limitation of observation-based floor management: human supervisors observe a fraction of what happens on a complex production floor during a shift, and their mental model of where losses occur is shaped by the events they can see, remember, and discuss in shift handover meetings. The events that AI monitoring reveals most reliably are the ones that are hardest to see without continuous automated observation.
This guide covers what Nagare process optimization case study deployments demonstrate about the value that AI process monitoring generates in complex manufacturing environments.
What the P&G deployment revealed
The P&G deployment involved a high-speed FMCG production line running multiple product variants across three shifts. Before Nagare deployment, the plant tracked OEE through shift-based manual records, with supervisors entering downtime events and quality failures into a production tracking system at shift end.
The gap between manual records and AI monitoring data: the manual system captured planned downtime, major equipment failures, and end-of-line quality rejections. It missed micro-stoppages under two minutes, process compliance deviations where operators adapted standard procedures under time pressure, and the cumulative performance loss from lines running at 90-93% of their rated speed rather than 100%.
When Nagare monitoring was deployed, the machine-level data showed that micro-stoppages, each individually below the threshold that supervisors escalated, collectively accounted for 18% of available production time. The specific machines, the specific shift patterns, and the specific product variants that generated the highest micro-stoppage frequency were identified in the first two weeks of monitoring data.
The improvement program that followed targeted the root causes of those micro-stoppages: three specific machines with tooling wear patterns that generated increasing stoppage frequency as the shift progressed, and one product variant whose changeover sequence consistently ran 7 minutes longer than standard due to an instruction ambiguity.
What the Maruti Suzuki deployment revealed
The Maruti deployment was in an automotive component assembly environment with 22 assembly stations. The plant had a well-developed quality system with in-process inspection at three defined check points and end-of-line CMM verification. The in-process rejection rate was tracked and reported.
The gap that Nagare monitoring revealed: the defined inspection check points caught defects, but not the process deviations that created them. A sub-assembly that passed all three inspection gates and cleared end-of-line could have been assembled with sequence deviations that did not produce a measurable dimensional defect under normal conditions but created higher warranty failure rates in the field.
Nagare’s process compliance monitoring identified five assembly steps across three stations where operators consistently adapted the standard sequence under production time pressure. None of the five deviations produced a rejection at the current inspection gates. All five correlated with elevated warranty return rates for specific sub-assembly types, a connection that required cross-referencing warranty data against assembly monitoring data to establish.
The corrective action combined instruction redesign for two steps where the standard was ambiguous, and poka yoke additions for three steps where the sequence deviation was possible but not physically prevented.
The common pattern across both deployments
Both deployments illustrate the same principle: manual observation systems are calibrated to detect the events that are large enough, frequent enough, or visible enough to enter the manual reporting process. AI monitoring detects the events that fall below that threshold, which are often the events that cumulatively account for the largest fraction of quality and productivity loss.
The process improvement programs that follow AI monitoring deployment in these environments consistently differ from the programs that would have been designed from manual data alone. The improvement investment goes to different machines, different processes, and different operator training priorities because the data shows a different pattern of loss than the supervision team’s prior beliefs suggested.