In a recent piece for Automation World, Alan Raveling of Interstates Control Systems (a CSIA-certified integrator) writes artificial intelligence (AI) models are only as good as the instrumentation, historian data and process context feeding them. A model built on unreliable sensor readings, misaligned lab timestamps or poorly labeled operating modes will produce recommendations that can't be trusted—no matter how sophisticated the algorithm.

Raveling lays out a practical readiness path that starts with defining the operational decision to be improved (not the model itself), then building a data inventory, contextualizing tags to real process events like batches and CIP cycles, aligning timestamps across historians and lab systems, and establishing a secure, segmented OT-to-analytics data architecture. He recommends starting with advisory or diagnostic AI—flagging abnormal conditions or predicting equipment issues—rather than jumping straight to closed-loop control, since advisory models are far easier to validate and govern while trust is still being established.

For process control engineers, the article underscores that domain expertise—not data science—is the differentiating factor in whether AI initiatives succeed. Engineers who understand instrumentation reliability, process dynamics, and operating modes are best positioned to catch when a model is learning the wrong relationships, and their involvement early on is critical to building operator trust.

Read the full article at our sister publication, Automation World.