Data Envelopment Analysis
Compare the relative efficiency of comparable operating units, identify peer benchmarks, and translate the efficiency frontier into practical input or output targets.
Why use this tool?
Use DEA to compare branches, depots, suppliers, plants, service centres, or other units that use similar resources to produce similar outcomes.
It identifies the observed efficiency frontier, relative scores, benchmark peers, proportional improvement potential, and remaining input or output slacks.
Load the sample to compare distribution centres using staff and operating cost as inputs, and completed and on-time orders as outputs.
A score below 100% indicates a unit is below the observed frontier. Its peers and targets show what comparable performance could look like.
- 1Prepare DataDefine measures and comparable units
- 2Choose ModelSet returns to scale and orientation
- 3AnalyzeBuild the observed efficiency frontier
- 4InterpretReview peers, targets, and slacks
1. Prepare Comparable Data
Use units performing the same broad function, measured over the same period with consistent definitions.
Select an illustrative operating scenario, then review and adapt its measures, units, assumptions, and data before analysis.
Define measures and add comparable decision-making units manually, or configure the measures before importing a CSV.
Prepare your DEA file
Upload a CSV with one row per decision-making unit (DMU). The first column identifies the unit; remaining columns are the inputs and outputs being compared.
Example row: Depot North | 20 | 900000 | 12000 | 10800
Important: DEA is sensitive to measure selection. Avoid using the same concept as both an input and output, and do not mix units with fundamentally different responsibilities.
Input measures
Output measures
Define the measures above, add rows manually, load a sample template, or import a CSV.
Decision-making units
At least two units are required. More observations generally create a more informative comparison.
Add comparable units to review the sample size against the selected number of measures.
Data Quality Check
Review basic governance signals before choosing a DEA model.
2. Choose the DEA Model
Confirm the assumptions after reviewing the units, measures, scale differences, and decision objective.
Model Selection Guidance
Use BCC when units differ materially in scale. Use CCR when constant proportional scaling is a reasonable assumption. Choose input orientation when managers mainly control resources; choose output orientation when the decision focus is increasing outcomes with current resources.
3. Run DEA
Validate the prepared data and build the observed efficiency frontier using the selected model.
About Data Envelopment Analysis
Data Envelopment Analysis is a non-parametric operations-research method for comparing the relative efficiency of decision-making units that use multiple inputs to produce multiple outputs. Efficient units form an observed frontier; other units are assessed relative to feasible combinations of those peers.
DEA does not require a predefined production function or externally assigned input/output weights. That flexibility is useful, but it can also allow a unit to receive favourable implicit weights. Results should therefore be reviewed with operational knowledge and sensitivity checks.
Guidance
Compare units with similar objectives, services, operating environments, and access to resources. Segment materially different units before analysis.
Too many inputs and outputs relative to the number of DMUs can make most units appear efficient. Use decision-relevant measures and gather more units where possible.
Use the same time period, currency basis, accounting rules, and output definitions. Investigate missing values, outliers, and exceptional operating conditions.
Targets are mathematical projections onto the observed frontier. Confirm that peer practices and projected changes are operationally feasible before acting.
Methodology, Assumptions, and Limitations
Input-oriented envelopment
Minimise θ, subject to Xλ ≤ θxo, Yλ ≥ yo, and λ ≥ 0.
Output-oriented envelopment
Maximise φ, subject to Xλ ≤ xo, Yλ ≥ φyo, and λ ≥ 0. Reported efficiency is 1 / φ.
BCC / VRS condition
Σλj = 1. The CCR / CRS model omits this convexity restriction.
Scale efficiency
SE = CCR efficiency / BCC efficiency. Values range from 0 to 1; a value near 1 indicates that scale contributes little to the observed inefficiency.
Slacks
After radial adjustment, remaining input excesses and output shortfalls are calculated against the peer-weighted frontier target.
Units are comparable, inputs and outputs are measured consistently, values are non-negative, and the observed sample provides a meaningful production possibility set.
DEA is deterministic and sensitive to measurement error, outliers, omitted environmental factors, sample size, and measure selection. It does not establish statistical causality.
FAQ
What is a decision-making unit?
A DMU is the comparable entity being assessed, such as a branch, depot, supplier, plant, hospital, school, team, or service centre.
Should I choose CCR or BCC?
Use CCR when proportional scaling is a reasonable assumption. Use BCC when units operate at different sizes and you want to separate managerial efficiency from scale effects.
What does scale efficiency mean?
Scale efficiency compares the CCR score with the BCC score for the same data and orientation: Scale efficiency = CCR efficiency / BCC efficiency. A value close to 100% suggests the unit is operating near an efficient scale relative to the observed sample. A lower value suggests operating size contributes to its overall efficiency gap. This ratio does not identify whether the unit should become larger or smaller; determining increasing or decreasing returns to scale requires an additional returns-to-scale analysis.
What does an efficiency score of 82% mean?
In an input-oriented model, it indicates the unit could proportionally reduce controllable inputs by about 18% before considering slacks, relative to the observed frontier. In an output-oriented model, inspect the reported target increase and peers.
What do input slack and output slack mean?
A DEA score first applies a proportional, or radial, adjustment to all inputs or outputs. Slack is any additional non-proportional gap that remains before the unit reaches its peer-based frontier target. Input slack indicates excess use of a particular input after the proportional reduction. Output slack indicates a shortfall in a particular output after the proportional expansion. A zero slack means no additional adjustment is identified for that measure. Non-zero slacks are diagnostic signals, not automatic targets; confirm that the peer comparison and proposed change are operationally feasible.
Can financial and non-financial measures be combined?
Yes. DEA can combine different units of measurement because each measure is compared consistently across DMUs. The measures must still represent coherent resources and outcomes.
Does DEA prove that an efficient unit is well managed?
No. Efficiency is relative to the selected sample and measures. A 100% score means no observed comparator dominates the unit under the chosen model; it is not proof of absolute excellence.
Is my data uploaded?
No. CSV parsing, optimisation, results, and exports run locally in the current browser.