Plant operations are changing faster than the workforce that runs them. The World Economic Forum's Future of Jobs Report 2023 found that employers see skills gaps and difficulty attracting talent as the main barriers to transformation. According to McKinsey (2025), the share of U.S. manufacturing workers over 55 more than doubled in twenty years, and retirement rates have surged. At the same time, plants generate more data every year, and more of the day-to-day control work is moving to models that read that data.
The operator's role shifts with it. ARC Advisory Group describes the end state as "human-on-the-loop": the AI calculates and executes, and the operator moves from manual controller to high-level supervisor. That role asks for new skills built on top of the operator's knowledge of the unit. Seven of them matter most, and each builds on the one before, from real-time decision making at the board to planning conversations with the rest of the plant.
Data fluency
Plants produce large and growing streams of sensor readings, lab results and process variables, and making sense of that data is the harder part. The skill is separating meaningful signals from noise before acting on what a model recommends.
Take a furnace loop streaming hundreds of temperature tags. Before accepting a model-recommended fuel adjustment, the operator needs to spot and set aside thermocouples that spike intermittently, because they distort what the model sees. That habit of validating inputs becomes more important as models take on more of the routine decisions.
Plants can build this fluency deliberately. Short data-literacy modules fit into onboarding, and a weekly exercise in which crews identify the cleanest trends and the most reliable sensors keeps the habit alive. These basics prepare operators to work with systems that relate process variables to equipment performance while protecting plant reliability.
Systems thinking
A process model brings the whole unit, or a chain of units, into view at once. It can show how a small temperature rise in a preheater tightens a downstream separation constraint minutes later, and it can rank the variables that drive throughput, energy use and product quality. Operators who understand those connections spot a bottleneck as it forms instead of chasing isolated alarms.
One exercise builds the habit: trace a catalyst-aging event through each unit it touches. Feed variation shifts reactor selectivity, trims fractionator yield, starves a compressor and eventually erodes margin. Mapping that chain against the model's ranking of influential variables shows where a single intervention helps most, and where it could backfire.
A lighter version works every shift. Review one chart of the model's top drivers, ask why each of the top three matters, and compare the answer with operating procedures. Over time, operators move from adjusting individual setpoints to managing the behavior of the whole unit, a shift already visible in plants adopting AI in operations.
Working with the model
That systems view is the foundation for working alongside a model. Many plants start with the model making recommendations, and the operator validates each one before it is written to the distributed control system (DCS). Three questions keep that check safe and useful:
- Does the move respect current safety and product quality constraints?
- Will it improve the primary target for the unit, whether throughput, energy or giveaway?
- Is there enough context to explain the move to the next shift?
Once the model has earned trust, it moves to closed loop. It handles routine setpoint adjustments within limits the operations team sets, and the operator steps in on exceptions and owns the limits. That is the human-on-the-loop role in practice.
Trust builds faster when a "champion" role rotates across shifts, so ownership of model performance and feedback is shared. Champions annotate each accepted or rejected move, which creates an audit trail engineers can review and auditors can follow, and answers much of the concern about black-box logic. The partnership turns experience into documented practice and extends what an experienced crew can cover.
Adaptability and model health
Equipment ages, feed quality drifts and models are retrained, so the optimization problem keeps moving. When the process changes faster than the model adapts, model drift erodes the accuracy of every recommendation, so continuous validation becomes part of the front-line job. A simple routine catches trouble early:
- Track model error on a daily dashboard.
- Compare predictions with recent historian data to flag drift.
- Trigger retraining when error exceeds a predefined threshold.
- Align predictive maintenance windows with retraining needs, using the same sensor trends that forecast equipment issues.
Cross-training matters just as much. When a model relearns around a new feed blend, operators who understand several units can rebalance feeds, utilities and safety limits with confidence. It helps to treat every software update like a minor turnaround, and to run a monthly model-health drill so each crew practices diagnosing drift before it affects production.
Testing moves in simulation
A simulation of the plant built from its own historian data lets operators test a process change without touching live equipment, so the crew sees the result of a change before making it.
Load a new crude slate and watch the constraints carry through furnaces, compressors and product quality. Shift the steam and power split during peak pricing hours and look for the most profitable balance. Because the simulation draws on historian data and live sensor feeds, its answers reflect the constraints the crew faces. Plants that run regular simulation exercises, document the outcomes and share what they learn build operating intuition ahead of the next turnaround or market disruption.
Communicating what the model sees
Models surface patterns quickly, but the value appears only when those patterns reach the planners, engineers and managers who can act on them. Operators become translators between what the model sees and the business decision. Four steps keep the message focused:
- State the trend. Present the fact or change the model has surfaced.
- Add your analysis. Explain what it means and why it matters.
- Frame the trade-off. Put it in business terms, such as throughput against energy or risk against margin.
- Recommend an action. End with a clear next step or decision.
A daily five-minute huddle builds the habit. Pick one insight, such as an emerging fouling risk, and walk through the steps. For planners, emphasize schedule impact and production targets. For engineers, point to the specific sensor readings and equipment behavior. For leadership, express the risk avoided as margin preserved and downtime avoided. The hardest part is dropping jargon. Pairing with process engineers and plant managers to write plain-language summaries keeps everyone aligned on what the system sees and why it matters.
Steering energy and emissions from the console
Clear communication matters most when the subject is sustainability. Process models can weigh energy tariffs and environmental limits alongside production targets, adjusting fuel mix, steam loads and throughput so the plant meets its targets with less energy. Intensity metrics, such as kilograms of CO₂ per unit of product or megawatt-hours per tonne, give operators a running view of progress.
The operator's role moves from reconciling spreadsheets after the fact to steering margin and emissions together on the console. A margin-plus-emissions scoreboard in the control room, reviewed at every shift handoff, keeps both visible, and operators approve model updates when feed quality or utility pricing changes. The routine also surfaces hidden waste, such as excess flaring, minor leaks and poorly set recycle ratios, so crews can trace root causes and turn decarbonization targets into lasting operating practice.
How the skills fit together
Each skill supports the others. Data fluency sharpens systems thinking. Working with the model and checking its health keep its recommendations trustworthy. Simulation and clear communication turn what the model sees into decisions across the plant. Together they strengthen operator judgment and help the plant hold steady performance through market swings and workforce turnover.
How Imubit applies this
Imubit works with operating teams in refining, chemicals and polymers, cement, and mining and minerals. Map traces plan-to-plant gaps to the recurring decisions behind them. Model learns each unit's behavior from its own history with Deep Learning Process Control® (DLPC®). Control takes the critical decisions to closed loop on top of the existing APC and DCS, with operators on the loop as supervisors. Imubit has over 100 closed-loop controllers deployed worldwide. See the Imubit Platform.
