A water utility compares AI that can recommend valve changes with FAQ chatbots. Which threshold practice best reflects decision criticality?

Decision criticality should drive residual-risk tolerance. AI that can influence valve changes at a water utility can affect pressure, contamination risk, and service outages; it warrants a lower risk tolerance than FAQ chatbots that answer low-stakes customer questions. Differentiated thresholds allocate stronger controls, tighter monitoring, and higher approval bars to operational AI where errors have physical consequences.

Setting a lower residual-risk tolerance for FAQ bots that answer outage queries than for valve-change AI inverts criticality and under-controls OT-adjacent AI where errors have physical consequences. Setting the risk tolerance for valve-change AI using historical FAQ bot accuracy and uptime data ignores the physical consequences of operational errors and misapplies low-stakes metrics to high-stakes decisions. Setting a lower risk tolerance for valve-change AI in offline simulation than for live FAQ bots misplaces criticality by tightening the safe test environment while leaving live valve-influence decisions under-controlled where errors have physical consequences.

Exam caveat: "influence valve changes" includes recommenders that operators routinely follow, not only fully autonomous actuators—human-in-the-loop does not automatically justify FAQ-level tolerance. Operational check: classify each water AI use case by decision criticality and confirm residual-risk thresholds and required control packages differ between valve-influence systems and FAQ bots.