Which OEE Factor Is Actually Costing You the Most

A single OEE percentage tells you that a station is underperforming and nothing about what to do, because three independent losses multiply into it and the same overall figure can come from three completely different problems. Decomposing it names the intervention, the owner, and the price. EDGEBIC by User Solutions reports availability, performance, and quality separately per work center, which is what turns a score into a spending decision.

This post is about one return: not funding the wrong improvement. It sits under the EDGEBIC results guide and is the diagnostic companion to the capacity you already have but cannot see.

The Problem With Reporting One Number

OEE gets reported monthly in a great many plants and acted on in very few. The reason is structural, not cultural.

The figure is a product of three factors. That means a station reading 70 percent could be a machine that sat idle for a third of its shift, or one that ran the whole shift at two-thirds of its expected rate, or one that ran flat out and remade a third of what it produced. Those are three different plants' worth of problems sharing one number.

Multiplication also flatters and punishes in ways that surprise people the first time they see it. Three factors that each look respectable produce a product that looks poor:

Those are arithmetic illustrations, not results from anyone's shop. Look at the last two rows. They land within three points of each other and have almost nothing in common: one is a quality problem, one is a time problem, and any effort aimed at the wrong one is wasted. A manager reading only the right-hand column cannot tell them apart.

The Mechanism: Three Factors, Three Sources

The OEE report computes each factor from a different data source, which is precisely why the decomposition is diagnostic.

Availability is actual hours divided by shift-available hours. It comes from logged actuals against the shift calendar. It answers: did this station get the time it was supposed to get?

Performance is actual pieces divided by the theoretical pieces implied by the routing's cycle time. It answers: while it was running, did it run at the rate the routing claims?

Quality is good pieces divided by good plus scrap plus rework, from kiosk punches. It answers: of what it made, how much was usable first time?

OEE is the product of all three.

One design decision is worth calling out because it affects how much you can trust the number. If kiosk punches are not being captured, quality cannot be computed, and because OEE is a product, neither can OEE. The report shows n/a rather than quietly assuming 100 percent quality. Availability and performance still display. That blank is honest reporting of an unmeasured factor, and it is far more useful than a flattering number nobody should act on.

From a Factor to an Intervention

Here is the mapping that makes the decomposition worth doing, because each factor points at a different department, a different cause, and a different order of cost.

The row that saves the most money is the performance row, and not for the reason people expect. A performance figure well below 100 percent is very often a routing problem rather than a machine problem: the per-unit hours in the routing were estimated once, optimistically, and never revisited. Buying a faster machine to fix an over-optimistic routing is the most expensive way to correct a spreadsheet. Check the routing against variance data before pricing equipment.

The Causal Chain to Actual Money

Decomposition only pays if the chain reaches throughput. It does, but only under a condition.

1. You identify the weakest factor at a station. The report gives you this directly.
2. You fund the matching intervention. Not a general improvement push, one specific action against one factor.
3. That factor rises, and the product rises with it arithmetically. Recovered hours or recovered good pieces appear at that station.
4. If the station is your constraint, plant throughput rises. More good parts leave the plant.
5. If it is not the constraint, nothing happens to throughput. You created idle time at a station that already had spare capacity.

Step five is where most OEE programs quietly fail. This is why protecting the constraint has to come before any OEE ranking: the report will happily show you your worst station, and your worst station is frequently the one where improvement is worth the least. Rank by importance first, then by OEE.

The one honest exception is quality. Scrap made at a non-constraint station still consumes constraint hours downstream when the part is remade, so a quality fix anywhere upstream of the constraint does protect throughput. Time losses on a non-constraint do not.

Measuring It in Your Own Plant

This is straightforward and needs one month plus a habit.

1. Get quality punching in place at the stations you care about. Without good, scrap, and rework counts the third factor stays blank, and the kiosk captures both hours and pieces at once, so there is no separate mode to configure.
2. Run OEE per work center over a full month. A week is too noisy for a spending decision.
3. Record all three factors per station, not the product. A simple grid: station down the side, three factors across.
4. Mark the minimum factor per station. That is your candidate intervention for that station.
5. Sort stations by importance, not by OEE. Constraint first, then stations feeding it, then everything else.
6. Price the top one or two interventions, and estimate the hours or good pieces each would release at the constraint. That estimate against that cost is your business case, and it is yours rather than borrowed.
7. Track the factor you funded, monthly. Not the product. The product will move when the factor does, and watching the factor tells you whether the money worked.

For documented outcomes rather than typical ones, the User Solutions and RMDB lineage includes GE Railcar moving from 30 percent to 90 percent on-time delivery, and the USS Nimitz refit coordinating more than 26,000 tasks. Those belong to their engagements and are quoted as heritage, not as forecasts for a specific machine.

Where This Does Not Apply

Five limits, each of which has misled somebody.

Low availability is not always an equipment problem. Availability is actual hours over shift-available hours, so a station that sat idle because nothing was scheduled for it reads low without anything being wrong with the machine. Before calling maintenance, check whether the station simply had no queue. That is a demand or planning observation, not a reliability one.

Performance depends on the routing being right. The theoretical piece count comes from the routing's cycle time. An optimistic routing makes a perfectly healthy machine look slow, and a padded one makes a struggling machine look fine. The factor measures the machine against a claim, and the claim is data you control.

Quality is unknowable without punches. Not approximate, unknowable, which is why the report says n/a. Any OEE quoted while quality is blank is being quoted from two of three factors.

OEE says nothing about whether the work was worth doing. A station can post an excellent OEE building inventory nobody ordered. Effectiveness at the wrong thing is not a return, and the schedule, not the OEE report, is what tells you whether the work belongs there.

Improving a non-constraint buys idle time. Repeating it because it is the costliest mistake in the category: outside the quality exception above, an hour recovered off the constraint does not become an hour of output. It becomes an hour of waiting.

The takeaway

OEE reported as one number invites a general improvement effort, and general effort spread across three factors is too thin to move any of them. Reported as three factors per station, it names one intervention, one owner, and one price, and it lets you check that the station is even worth improving before you spend. Decompose it, sort your stations by importance rather than by score, fund the weakest factor at the station that actually sets your output, then track that factor instead of the product. To see the three factors against your own stations, book a demo of EDGEBIC, and if you are running the older platform, the move from RMDB to EDGEBIC brings the reporting with it. Read this next to how EDGEBIC raises bottleneck utilization and how kiosk actuals close the planning loop.