Blog » Industry
OEE Driver Tree: Count Minutes, Not Percentage Points
August 29, 2026 · 12 min read
OEE driver tree: the three factors multiply exactly, yet their percentage points do not add. Build the loss ladder in minutes, with worked arithmetic.
The Week Availability Improved and OEE Fell
A packaging line ran 4,400 minutes of planned production time in two consecutive weeks. In the second week the maintenance crew cut unplanned stops from 440 minutes to 352, and availability rose from 90.00 percent to 92.00 percent. Overall equipment effectiveness fell from 76.95 percent to 70.32 percent.
An OEE driver tree explains that result mechanically. It decomposes overall equipment effectiveness into the losses that produced it, in a structure where every parent node equals a defined function of its children, so one explanation reconciles instead of competing with a second explanation built by someone else.
The trap is that most OEE trees are built on the three percentages. Availability moved plus 2.00 points. Performance moved minus 5.07 points. Quality moved minus 5.00 points. Those three add to minus 8.07 points. OEE moved minus 6.63 points. The 1.44 point gap is not rounding, and no amount of decimal precision closes it.
The gap closes when the tree is built in minutes.
What Is an OEE Driver Tree?
An OEE driver tree is a hierarchical decomposition of overall equipment effectiveness into the equipment losses that produced it. The root is OEE, or the productive time behind it. The branches are availability loss, speed loss and quality loss, measured in minutes of the same shift, so every child sums into its parent exactly rather than approximately.
Overall equipment effectiveness is availability multiplied by performance multiplied by quality. ISO 22400-2:2014 defines the three components and the wider set of manufacturing operations indicators they sit inside, so two plants following the standard reach the same number from the same facts.¹ Semiconductor fabs use a parallel specification, SEMI E79, for equipment productivity, and SEMI E10 for the underlying equipment state definitions that availability is computed from.²³ Both retrieved 29 August 2026.
A driver tree adds one thing the definitions do not. A definition says what the number is. A tree says which child moved it, and by how much, when the number changes between two periods.
That distinction matters because the standards deliberately stop at the definition. They give the formula and the elements. They do not give an attribution method, and attribution is the whole reason anyone opens the report on Monday morning.
The Time Ladder Beneath the Three Factors
Overall equipment effectiveness is not three independent measurements. It is three cuts of one descending time ladder, where each rung equals the rung above it minus one category of loss.
Week one on the packaging line, at an ideal cycle time of 0.45 minutes per unit:
Planned production time is 4,400 minutes. Subtract 440 minutes of unplanned stops and run time is 3,960 minutes. Availability is 3,960 divided by 4,400, or 90.00 percent.
The line produced 7,920 units inside that run time. At the ideal cycle time those units should have taken 3,564 minutes, so 396 minutes went to speed loss. Performance is 3,564 divided by 3,960, or 90.00 percent.
Of the 7,920 units, 396 failed inspection. The 7,524 good units carry 3,385.8 minutes of ideal time, so 178.2 minutes went to quality loss. Quality is 7,524 divided by 7,920, or 95.00 percent.
The bottom rung, 3,385.8 minutes, is fully productive time. Divided by the 4,400 minutes at the top it gives 76.95 percent, which is the same number as 0.90 times 0.90 times 0.95.
Why Do the OEE Factors Multiply Exactly When DuPont Factors Do Not?
Because each OEE factor is a ratio of two consecutive rungs on the same time ladder. Run time is the numerator of availability and the denominator of performance, so it cancels. The product telescopes down to fully productive time over planned production time. DuPont factors share no such rungs, so their separate effects leave a residual.
Write the product out and the cancellation is visible. Availability is run time over planned production time. Performance is net run time over run time. Quality is fully productive time over net run time. Multiply the three and run time cancels against run time, net run time cancels against net run time, and what survives is fully productive time over planned production time.
That is why 0.90 times 0.90 times 0.95 lands on 76.95 percent with no residual. The identity is true by construction, not by approximation.
This is the opposite result from a ratio chain whose factors are drawn from different statements. In a DuPont ROE decomposition the three factors multiply but their individual effects do not sum to the change in ROE, and the unallocated cross-term is real. OEE has no cross-term problem in its levels.
It has a different problem, one period later.
The Minutes Bridge That Reconciles
Week two, same line, same 4,400 minutes of planned production time.
Unplanned stops fell to 352 minutes, so run time rose to 4,048 minutes and availability rose to 92.00 percent. The line produced 7,640 units, worth 3,438 minutes of ideal time, so speed loss rose to 610 minutes and performance fell to 84.93 percent. Scrap rose to 764 units, so quality loss rose to 343.8 minutes and quality fell to 90.00 percent. Fully productive time is 6,876 good units times 0.45, or 3,094.2 minutes, and OEE is 70.32 percent.
Bridge the two weeks in minutes of fully productive time. The starting value is 3,385.8 and the ending value is 3,094.2, a fall of 291.6 minutes.
Downtime loss fell 88 minutes, which adds 88 minutes. Speed loss rose 214 minutes, which subtracts 214. Quality loss rose 165.6 minutes, which subtracts 165.6. Planned production time did not move, so it contributes nothing.
Plus 88, minus 214, minus 165.6 is minus 291.6. The bridge reconciles to the tenth of a minute, with nothing unallocated.
Why Do OEE Percentage Points Not Add Up?
Because a percentage point is a different quantity of minutes at each rung of the ladder. In week two one availability point is 44 minutes of planned production time, one performance point is 40.48 minutes of run time, and one quality point is 34.38 minutes of net run time. Adding them treats three different units as one.
Each factor is measured against its own denominator, and those denominators shrink as you descend. Availability divides by 4,400 minutes. Performance divides by 4,048. Quality divides by 3,438.
So one availability point buys 44 minutes, one performance point buys 40.48 minutes, and one quality point buys 34.38 minutes. An availability point is worth 1.28 quality points in shift time.
The practical consequence is a ranking error. Two losses inside 28 percent of each other on percentage points can reverse order once converted to minutes, and the loss that gets the maintenance budget is the wrong one. On the packaging line the order happens to survive the conversion, but the magnitudes do not: performance and quality look 0.07 points apart and are 48.4 minutes apart.
The rule that follows is simple. Store minutes and units. Derive every percentage by dividing two stored columns. Never store a ratio as a leaf and never rank leaves by ratio movement.
The Six Big Losses as the Leaf Layer
Seiichi Nakajima introduced the six big losses at the Japan Institute of Plant Maintenance and published them in Introduction to TPM in 1988. They map onto the three OEE factors two apiece, which makes them the natural leaf layer of the tree.⁵ Retrieved 29 August 2026.
Under availability sit equipment failure and setup and adjustment. Under performance sit idling and minor stoppages, and reduced speed. Under quality sit process defects and rework, and startup and yield losses.
That mapping is what makes the tree operational rather than descriptive. The week two speed loss of 610 minutes is a number nobody owns. Split into 418 minutes of minor stoppages and 192 minutes of reduced speed, it becomes two different conversations with two different owners, and only one of them is a maintenance problem.
Six leaves is usually enough. Plants that add a seventh and eighth category almost always break the sum, because the new category overlaps one already there, and overlapping siblings stop the tree reconciling.
Where the Planned Production Time Boundary Moves the Number
The most common way an OEE tree produces a wrong answer has nothing to do with the arithmetic. It is the boundary at the top rung.
Planned production time is scheduled time minus planned stops. What counts as a planned stop is a local convention. Breaks, changeovers, planned maintenance, and unsold capacity can each sit inside or outside the boundary, and each choice changes OEE without changing a single minute of what the machine actually did.
Move the 400 minutes of planned downtime on the packaging line back inside the top rung and week two runs against 4,800 minutes instead of 4,400. Fully productive time is unchanged at 3,094.2 minutes. OEE becomes 64.46 percent instead of 70.32 percent, a difference of 5.86 points on identical production.
That sensitivity is why ISO 22400-2 exists, and why the semiconductor industry wrote SEMI E79 rather than borrowing a general definition.¹² Comparability is the whole point of a standard, and a tree that does not state its boundary has no comparability to defend.
Two plants reporting 70 percent OEE are not necessarily equal. Ask what sits inside planned production time before comparing anything.
Three Ways to Root an OEE Tree
The root choice decides what the tree can answer. All three roots below are defensible, and the wrong one for a given question is not fixable further down.
| Tree root | Parent-child math | Answers well | Where it breaks |
|---|---|---|---|
| OEE percentage | Three ratios multiplied down the ladder | Tracking one asset against its own history | Contributions do not sum, and a point means a different number of minutes at each rung |
| Fully productive time in minutes | Planned time minus three additive loss columns | Which loss cost the most shift time, and by exactly how much | Needs an ideal cycle time, and mixed-SKU lines must weight it by product |
| Good units produced | Total units minus scrap, driven by run time and cycle time | Linking equipment loss to the order book and to shipped revenue | Hides time loss whenever the ideal cycle time is set loosely |
Four Correctness Tests for an OEE Tree
Test one, the ladder test. Every rung equals the rung above it minus exactly one loss column. Sum planned production time against downtime loss, speed loss, quality loss and fully productive time. If the four do not add to the top rung, a loss is double counted or missing.
Test two, the telescoping test. Multiply the three ratios and compare with fully productive time divided by planned production time. The two must agree to at least four decimals. Disagreement means one factor was computed against the wrong denominator.
Test three, the unit test. No leaf stores a ratio. Every stored column is minutes or units, and every percentage in the tree is derived by dividing two stored columns. This is the same discipline that keeps an additive price, volume and mix bridge from drifting.
Test four, the boundary test. Recompute with planned downtime moved across the top boundary. If the two answers differ by more than a point or two, the boundary is carrying more of the result than the improvement work is.
Where an OEE Tree Is the Wrong Instrument
OEE measures one asset in isolation, and that is exactly its limit. Raising OEE on a machine that is not the constraint produces inventory rather than throughput, and overproduction is the more expensive of the two outcomes.⁶ Retrieved 29 August 2026. Run the tree on the bottleneck, or accept that the number is describing a local optimum.
In a make-to-order plant where demand sets output, low OEE is a scheduling fact, not a maintenance failure. The tree will faithfully attribute the shortfall to availability loss and mislead everyone who reads it.
Mixed-SKU lines break performance outright when a single ideal cycle time is applied across products with different rates. Weight the ideal time by production mix or drop performance from the tree.
The 85 percent world-class figure is worth the same caution. It comes from Nakajima's observation that plants winning the JIPM Distinguished Plant Prize exceeded 85 percent, not from a study of what any given line should reach.⁴ Retrieved 29 August 2026.
And a target attached to a performance review stops measuring reality and starts measuring reporting behaviour.
Building the Ladder in kpitree.io
kpitree.io is a self-service KPI tree builder for finance, business and product analysts. It takes a CSV upload and turns the columns into an interactive tree, which is the shape an OEE ladder already has.
The columns are planned production time, downtime minutes, speed loss minutes, quality loss minutes, total units and good units, one row per period per asset. Every one of those is summable, so every parent node in the tree is a sum or a difference of its children, and the ladder reconciles by construction.
Availability, performance and quality are then computed inside the tree by dividing two summable columns. No row stores a ratio, which is what keeps the week-to-week bridge additive instead of leaving a residual.
The same pattern applies wherever a rate hides a partition, including the four-cell OTIF order partition on the delivery side of the same plant.
The smallest useful next step is one CSV: one line, two periods, six columns. Upload it and decompose the OEE number your Monday review already argues about.
Frequently Asked Questions
Is OEE availability the same as equipment availability under SEMI E10? No. SEMI E10 defines equipment states and computes availability against total time, while OEE availability runs against planned production time.²³ The two numbers answer different questions and should not be compared.
What OEE score is good? The widely quoted 85 percent comes from Nakajima's observation of JIPM prize-winning plants, not from a benchmark study of your process.⁴ A rising trend against a stated boundary is more informative than the level.
Should planned maintenance sit inside planned production time? Either convention works if it is stated and held constant. Moving 400 minutes across that line changed the packaging example by 5.86 points on identical output.
Can I build an OEE tree without an ideal cycle time? You can build the availability branch alone in minutes. Performance and quality both need an ideal cycle time, because both are expressed as time equivalents of unit counts.
Does line OEE equal the average of machine OEEs? No. A line is limited by its constraint, so averaging machine scores overstates line performance in almost every case.
Closing: Count the Minutes
Overall equipment effectiveness is one of the few operating metrics whose factors multiply exactly, and that exactness makes the percentages look safer than they are. The product telescopes. The point changes do not add.
Driver-based practice is still uncommon enough that most teams have no default here. The 2026 AFP FP&A Benchmarking Survey on integrated planning drew on 332 finance professionals across 54 countries, and driver-based modelling remains a stated ambition more often than a built system.⁷ Retrieved 29 August 2026.
Build the ladder in minutes. Put the six big losses at the leaves. State the top boundary out loud. Then the bridge from last week to this week reconciles to the tenth of a minute, and the argument moves from whose number is right to which loss to attack first.
kpitree.io builds the tree from your own uploaded data, deriving every rate by dividing two summable columns.
Sources
- ISO 22400-2:2014, Key performance indicators for manufacturing operations management, Part 2: Definitions and descriptions
- SEMI E79, Specification for Definition and Measurement of Equipment Productivity
- SEMI E10, Specification for Definition and Measurement of Equipment Reliability, Availability and Maintainability
- World-Class OEE: Set Targets To Drive Improvement
- The Six Big Losses (Nakajima 1988): Definition, Mapping to OEE
- What the OEE is Good for and What it's Not
- 2026 AFP FP&A Benchmarking Survey Report: Integrated Planning