Why Low OEE Doesn’t Always Mean the Machine Is the Problem

Overall equipment effectiveness combines availability, performance and quality. Availability reflects whether the process runs during the time it is expected to run. Performance compares its actual running rate with the defined ideal rate. Quality reflects the proportion of output that meets the required standard. Multiplying the three factors provides one measure of how much planned production capacity became good output at the expected rate.

That makes OEE useful. It exposes a gap between planned productive time and what the process delivered. Its components and underlying loss records can direct attention toward breakdowns, short stops, slow cycles and defects that would otherwise disappear inside a shift total.

The difficulty begins when the final percentage is treated as a diagnosis. A machine-level OEE result can contain time lost because of conditions outside that machine. Without understanding those conditions, a precise number can create a false sense that the location and ownership of the problem are already known.

A production line is a connected system. Each process depends on upstream operations to supply acceptable work and on downstream operations to accept completed work. A process can therefore stop producing even when it is mechanically healthy and capable of meeting its cycle time.

When upstream output is late or uneven, the process becomes starved. When downstream capacity, a buffer, inspection or material-handling route cannot accept its output, the process becomes blocked. Both conditions reduce productive time at the measured process, but neither necessarily originates there.

This is why ranking machines by OEE is a weak way to identify the production constraint. The lowest figure may belong to a capable machine that is repeatedly denied work. Meanwhile, a more highly rated upstream process may be unstable enough to govern the flow of the whole line. Bottleneck identification requires evidence about queues, starvation, blocking, recovery and finished output—not simply the lowest local percentage.

The first practical distinction is between loss originating in the measured process and loss imposed by the surrounding system. This is not an accounting exercise designed to make a machine’s result look better. It is a way to put investigation and improvement ownership in the right place.

Equipment-controllable loss points toward the asset, process method or immediate operating conditions. System-induced loss must be followed beyond the machine boundary. It may belong to material supply, another operation, production control, quality, line balance or a decision that prevents work from moving.

A worked multistage-line example described by Prof. He (Herman) Tang shows why this separation can matter. The conventional OEE for the measured process was 87.2%. A standalone OEE view, intended to distinguish the process’s own losses from losses imposed by surrounding operations, was 94%.

The 6.8 percentage-point difference represented external effects included in the conventional result: starvation, blocking and quality loss inherited from upstream. The conventional figure still described lost production opportunity at that point in the line. The standalone view answered a different diagnostic question: how was the process performing when losses outside its control were separated from its own performance?

This is an engineering example and diagnostic framework, not a documented FlowForge client result or proof of one universal OEE method. Its value is narrower and more practical. It demonstrates that one KPI can contain losses originating in several places, so the number must be decomposed before work is assigned.

When this machine isn’t producing, is it broken—or is it waiting? That question changes the investigation. If it is broken, slow or generating defects, the improvement belongs at the equipment or process. If it is waiting, ask what it is waiting for, why, and who owns that loss.

Observe a representative production period and categorise each non-producing interval. Useful categories include machine fault, minor stop, changeover, slow cycle, waiting for material, waiting for an upstream operation, downstream blocked, quality hold, and waiting for information, an operator or approval. Record start and finish times where practical rather than relying on a reason selected at the end of the shift.

A simple tally sheet can be more useful than immediately debating the monthly OEE number. It preserves the sequence of events: what stopped first, where work accumulated, which processes became starved and whether the line recovered. That sequence is often where the cause becomes visible.

Separating external loss does not make it disappear. If a healthy machine is starved for 45 minutes, the machine may not need fixing, but the production system does. The factory has still lost time and may still lose throughput, delivery performance or recovery capacity.

Follow starvation upstream. Establish whether material was unavailable, the previous process was stopped, output arrived in the wrong sequence or a quality issue prevented release. Follow blocking downstream. Look for a full buffer, an unstable next operation, inspection delay, handling restriction or schedule decision that stopped completed work moving.

Then test the effect at system level. Where is work accumulating? Which process cannot recover after disruption? When the suspected cause is removed, does finished output increase? If another process is the true constraint, protecting its time may matter more than raising the local OEE of the waiting machine.

OEE becomes unhelpful when comparison replaces investigation. Differences in product mix, planned time, ideal cycle definitions and loss coding can make a machine league table look authoritative while comparing unlike operating conditions. Even with consistent definitions, the lowest result does not establish the source of the loss.

An arbitrary target creates the same problem. Chasing a supposedly world-class percentage can direct effort toward a non-constraint with no effect on customer output. The better question is which recorded losses matter to the defined production system and which improvement can change its result.

Use the checklist beside the process during representative production. Base answers on observed time and events, then assign an owner to each significant category. Repeat the exercise when the product mix, shift pattern or operating conditions change materially.

A KPI is evidence. It is not the process. OEE can show that planned productive capacity did not become good output at the expected rate, but the monthly percentage cannot show every interaction that produced the loss.

Go to the shop floor. Observe when production stops. Ask what happened immediately before the stop and what happened elsewhere afterwards. Separate the process’s own losses from those imposed upon it, while keeping both visible in the system-level picture. Then follow the cause through material flow, adjacent operations, quality decisions and production control.

This discipline prevents a useful measure becoming a shortcut to the wrong action. See the process. Understand where the loss originates. Then decide what to improve.

An independent review can help when several machines appear to perform poorly, teams disagree about the true constraint, or OEE and actual throughput tell different stories. It is also useful when starvation and blocking are common but their causes and ownership remain unclear, or when repeated improvement activity has not increased shipped output.

A practical Production Review combines available data with direct observation of the production process. The aim is to separate equipment losses from system losses, test where the current constraint sits and give the team a defensible priority—not to produce another performance league table.