OEE Across Multiple Plants: Compare Performance Without Misleading Rankings
Two plants can run equally well and still report different overall equipment effectiveness (OEE). If one excludes changeovers from planned production time while the other includes them, the regional ranking partly reflects a reporting decision. McKinsey found that 74% of more than 100 manufacturing COOs surveyed reported having a global production system, while 29% reported full implementation across all sites. For APAC manufacturers, that gap makes it worth checking how each site applies the measurement rules before comparing scores. [1][2]
Compare plants on OEE only after aligning the measurement boundary, planned time, reference speeds, and good-count rules. Use matched production groups for site comparisons and planned-time weighting for an eligible network rollup. Then examine availability, performance, and quality separately to decide where to investigate. [2]
1. Agree on what each plant measures
Start with the physical boundary. Decide whether the score represents a machine, a work cell, or a complete line. For a line, agree on the output counting point and the signals that establish when the line is producing. Keep individual machine scores available for diagnosis. A line-level measure needs its own elapsed time and output, since summing the operating time of machines in series would count overlapping minutes. [2]
Next, publish a common rule for planned production time. Define the treatment of breaks, scheduled maintenance, changeovers, and periods without demand. Decide where short stops enter the calculation so each loss appears once. Keep the original schedule and any approved corrections. The OPC Foundation’s machine-tool KPI guidance identifies planned busy time and planned run time per item as inputs typically supplied by a manufacturing execution system (MES) or enterprise resource planning (ERP) system. [2]
2. Calculate the three OEE factors
The OPC Foundation’s guidance, drawing on ISO 22400-2, defines OEE as availability × effectiveness × quality rate. This guide uses performance for the effectiveness factor. For a run with one approved reference cycle time, calculate each factor as follows. [2]
| Factor | Calculation | Question for the plant |
|---|---|---|
| Availability | Actual production time ÷ planned production time | How much of the scheduled window did the asset spend producing? |
| Performance | Reference cycle time × total units ÷ actual production time | How did output during production compare with the approved rate? |
| Quality | Good units ÷ total units | What share of output met the agreed quality rule? |
Record counts at the same boundary and specify whether reworked units qualify as good. If the group adopts first-pass quality, apply that rule consistently. Store the reference cycle time with its product, equipment configuration, and effective date. Investigate performance above 100% as a possible issue with the standard, count, or time mapping before publishing the result. [2][3]
3. Work through a transparent example
Suppose two plants run the same product on comparable lines, use a reference cycle of one minute per unit, and count first-pass good units. These illustrative figures use the formulas above. [2]
| Input or result | Plant A | Plant B |
|---|---|---|
| Planned production time | 480 min | 240 min |
| Actual production time | 420 min | 216 min |
| Total units | 400 | 210 |
| Good units | 380 | 204 |
| Availability | 87.5% | 90.0% |
| Performance | 95.2% | 97.2% |
| Quality | 95.0% | 97.1% |
| OEE | 79.2% | 85.0% |
At Plant A, the product of the three unrounded factors simplifies to 380 reference minutes of good output divided by 480 planned minutes. For these eligible observations, the network score is 584 ÷ 720 = 81.1%. A simple average gives 82.1%, which answers a different question by giving each plant equal weight even though Plant A accounts for twice the planned time. [2]
4. Normalize the context before comparison
Start with comparable cohorts, such as the same product family on equivalent lines under a shared measurement policy. For mixed products, calculate the reference time represented by each product’s good output, sum those minutes, and divide by the corresponding planned time if the scope supports pooling. Keep product-level components available. A simple average of product quality rates can obscure the effect of different cycle times and production volumes. [2][3]
Show each cohort’s product mix, planned hours, changeover policy, and data coverage beside its score. If one site runs longer campaigns and another handles frequent small orders, compare equivalent runs or report the groups separately. Agree on how to handle missing machine records, counter resets, and revised quality dispositions. Publish any exclusions and mark incomplete periods as provisional. [1][2][3]
Technical deployment pipeline
- Define the metric. Approve the measurement boundary, loss rules, and calculation version. [2]
- Map the inputs. Link machine states and counts to production orders, schedules, and quality records. [2][3]
- Validate reference rates. Confirm product and equipment settings with process engineering. [2][3]
- Reconcile a sample period. Compare calculated times and counts with reviewed shift records and investigate discrepancies. [2][4]
- Test regional reporting. Verify weighting, product cohorts, and drill-down from site totals to source records. [2][4]
- Assign ownership. Give plant teams responsibility for exceptions and regional operations responsibility for definition changes. [1][4]
5. Use differences to investigate losses
Return to Plant A. The example shows 60 minutes lost to availability, 20 reference minutes lost to performance, and 20 reference minutes lost to quality. Start by asking what drove the availability gap, then check whether the same loss recurs on comparable runs. Rank actions by recoverable time at the constraint and validate the result in a subsequent comparable period. [2]
Lessons Learned from the Field
TMA reports that a client using its Machine Monitoring System achieved a 17% improvement in OEE and a 10% productivity gain across shifts, which it attributes to real-time visibility and downtime alerts. The public account does not give the baseline, measurement period, or calculation policy. Treat the figures as a reported client outcome and set pilot targets from your own reconciled records. [4]
A related TMA case shows why the response workflow matters. At a Singapore nutritional ingredient manufacturer, critical spare shortages could prolong shutdowns. TMA combined a hierarchical asset register and maintenance workflows with low-stock alerts. In procurement, ask a low-cost offshore development centre scoped mainly to reporting to demonstrate how it would trace a lost minute to a stop, a part shortage, and a completed work order. That exercise tests who can own the interfaces and the operational response. [4][5]
TMA reports 29 years of engineering experience, 4,000 engineers serving clients in 30 countries, and mission-critical programmes involving hundreds of engineers. Its telecom R&D centres provide a precedent for moving from implementation to a longer-term partnership. TMA also lists ISO/IEC 27001 and reports GDPR checks in software quality assurance. Verify certification scope and agree on access, retention, and regional sharing where records identify personnel. [6][7][8]
FAQ
Can we average OEE across plants?
Use planned-time weighting when plants share a comparable scope and calculation policy. For unlike processes, report separate cohorts. A simple average gives each plant equal influence regardless of planned hours. [2]
Should planned changeovers count against OEE?
Apply the agreed planned-time policy consistently. If changeovers occur inside the included production window, account for them under the approved loss rules and show the treatment beside site comparisons. [2][3]
How should we calculate OEE for mixed products?
Apply the approved reference cycle time for each product and equipment configuration. For a comparable rollup, divide the sum of reference minutes for good output by the corresponding planned minutes. [2][3]
What should we do when performance exceeds 100%?
Check the reference rate, count units, duplicate records, and time mapping. Preserve the raw result for diagnosis and resolve the discrepancy before using it in a ranking. [2][3]
How should buyers evaluate an OEE vendor?
Ask the vendor to reproduce one shift’s score from controller, schedule, and quality records. Include a product change and a late correction. Confirm who owns source integration, metric governance, and support. [2][3][4]
Start with a comparison the plants can reproduce
Bring two comparable lines, their shift schedules, cycle standards, and quality counts to a TMA Innovation OEE assessment. Reproduce each score from its source records, agree on the rollup rules, and select the first recurring loss to investigate. [2][4]
References
[1] McKinsey — Transforming factories: The power of continuous, connected insights, May 2026.
[2] OPC Foundation — OPC UA for Machine Tools, Annex C: KPI Calculation.
[3] OPC Foundation — OPC UA for Machinery: Job Order Parameters and Response Data.
[4] TMA Solutions — Smart Manufacturing Solutions: An Integrated Approach for Scalable Industry 4.0 Deployment.
[5] TMA Innovation — Smart CMMS and Singapore Production Maintenance Case Study.
[6] TMA Solutions — Company Overview and Engineering Capabilities.
[7] TMA Solutions — Telecommunications Engineering and R&D Centres.
[8] TMA Solutions — CSR Report 2024, Security and Compliance.


