OEE is one of the most-discussed and least-understood metrics in manufacturing. Most factory floors carry a printed OEE chart somewhere visible. Most of those charts show roughly the same trend line. Most of those trend lines move sideways. The reason is that OEE improvement is treated as a single goal, when it is actually three different goals with three different paths to improvement, and only two of them respond to the kind of automation manufacturers typically buy.
The three levers, in order of automation responsiveness
Availability is the time the equipment is actually running versus scheduled to run. This is the lever most responsive to automation, specifically predictive maintenance and digital work management. McKinsey research finds predictive maintenance can reduce equipment downtime by 30 to 50 percent and lower maintenance costs by 10 to 40 percent [1]. Performance is the speed the equipment runs versus its design speed. This responds to a mix of automation and process redesign. Quality is the percentage of output meeting specification on first pass. This responds least to automation and most to upstream process control and workflow design. A factory pursuing OEE improvement by buying automation tools without diagnosing which lever is binding will spend money on the wrong layer.
A factory buying predictive maintenance because everyone is buying predictive maintenance is the dominant failure mode. Diagnose first. Automate second.
Diagnose first, automate second
The first question to answer before any automation: which of the three levers is your binding constraint? In our experience, the answer varies more than people expect. Discrete manufacturers with high product variety are typically performance-constrained, because changeovers eat available time. Continuous-process manufacturers are typically availability-constrained, because unplanned downtime cascades. Heavy industrial operations are quality-constrained, because rework consumes the headroom. Each diagnosis points to a different automation strategy. Skipping the diagnosis and buying predictive maintenance because everyone is buying predictive maintenance is the dominant failure mode. Aberdeen Group research, frequently cited in McKinsey work, puts the cost of unplanned downtime at an average of USD 260,000 per hour for industrial operations [2], which is why predictive maintenance gets the attention. But if your binding constraint is changeover speed, no amount of predictive maintenance investment will move the OEE number.
What automation actually does to OEE
For availability, the mechanism is well-documented. Sensor data feeds machine-learning models that detect signature patterns of approaching failure. Maintenance teams get advance warning, typically 5 to 7 days for critical components and 2 to 4 weeks for gradually-degrading systems. Repairs happen during planned downtime instead of unplanned. McKinsey case work cites organizations achieving 10:1 to 30:1 ROI ratios within 12 to 18 months of implementation [3]. For performance, the mechanism is less about prediction and more about real-time anomaly detection: spotting when equipment is running slower than design speed, and routing the alert to the right operator with enough context to act before the slowdown becomes the new normal. For quality, automation contributes through inline inspection and statistical process control, but the largest gains come from upstream workflow improvements that automation alone cannot deliver.
The headcount question
The framing of 'without headcount' matters because it forces the discussion away from labor displacement and toward capacity reallocation. A factory that improves OEE by 10 percentage points without adding people is not running with fewer people. It is producing more, or producing the same with more headroom for the things headcount was previously consumed by. In Belgian and Benelux manufacturing, where talent scarcity is acute and the OEE-to-headcount ratio is the binding metric for many operations directors, this framing matters more than the absolute productivity gain. Automation that improves OEE while freeing maintenance technicians from reactive firefighting is automation that solves both problems at once.
Roughly two-thirds of the OEE gap concentrates in fewer than five recurring loss patterns. The patterns are not glamorous. Naming them is the work.
The two-thirds rule
A useful heuristic: roughly two-thirds of the OEE gap on most factory floors is concentrated in fewer than five recurring loss patterns. Identifying those patterns through structured loss accounting is the highest-leverage analytical work a manufacturer can do. The patterns are usually some combination of one specific changeover that runs long, one specific tool that fails predictably, one specific quality defect that recurs, and one specific upstream feed issue that cascades. Each of those four has a different fix, and only one or two of them benefit meaningfully from machine-learning-based prediction. The rest benefit from rules, sensors, and process redesign, all of which are cheaper and more durable than ML-based approaches.
What gets measured
Three baseline measurements before any investment. First, OEE decomposed into the three component levers, calculated weekly for at least the preceding three months. Second, planned-to-actual variance on the top ten products by volume. Third, root-cause distribution of unplanned downtime events, ideally over six months. These three numbers tell you which lever to attack, which products are most exposed, and which failure modes deserve attention. Without them, automation investment is informed by vendor sales pitches rather than operational reality.
The role of the digital work management layer
McKinsey identifies digital work management as the second pillar alongside predictive maintenance, and the two compound. Digital work management captures field observations as structured data, routes work orders to the right technician with the right parts, and feeds the data captured during repairs back into the predictive maintenance model. Without the digital work management layer, predictive maintenance models degrade over time because the ground-truth labels needed to retrain them never make it back to the data layer. Most stalled predictive maintenance pilots stall here, not at the model. Building the work management layer first, even before the predictive maintenance model, is the right sequence for most mid-market manufacturers.
The Belgian and EU context
For Belgian manufacturers specifically, two contextual factors shape the OEE conversation. First, energy costs make availability improvements disproportionately valuable: every hour of avoided unplanned downtime saves not just labor and throughput but the energy cost of restart cycles, which for heavy industrial equipment can be substantial. Second, the EU AI Act's risk-based framework, with high-risk system rules entering application from 2 August 2026, means that any safety-critical automation needs governance designed in from the start [4]. This is not a blocker, but it is a factor in vendor selection and architecture choices that mid-market manufacturers often discover too late.
