Industrial Productivity Benchmarking
Compare productivity across processes while accounting for products, resources, scope and contextual differences.
Updated:
Benchmarking compares processes and identifies practices worth testing. Comparing percentages without their definitions can create misleading rankings. The aim is to understand relevant differences while maintaining quality, safety and service, rather than maximise direct-work share alone.
Before comparing
Document each study's:
- Population, workstations, period, shifts and represented hours.
- Categories and rules: do setup, inspection and coordination count as direct or support?
- Unit: person-instant, machine-instant or team state.
- Selection plan, missing data, weights and dependence.
- Volume, mix, quality, automation and outsourcing.
- Proportions, effective information and uncertainty.
If one plant outsources cleaning while another includes cleaners in the observed workforce, the proportions answer different questions. Do not interpret the difference as superiority without aligning scope.
Start with internal references
An internal baseline uses known definitions. Compare shifts or campaigns only with compatible scope; a shift doing more setup may handle a more complex mix.
A professional directory can help locate specialists but is not itself a sector benchmark database. Before using an external reference, request its original source, date, sectors, definitions, sample and method. Without these, do not present it as a verified target. No universal wrench-time or direct-work range diagnoses efficiency.
Teaching example: adjusted comparison
Plant A reports 60% direct work in 400 records; B reports 52% in another 400. Under simple independent samples, the difference is eight points and its approximate normal 95% interval is 1.1–14.9 percentage points. Statistical difference establishes neither superiority nor cause.
Category review finds A includes setup as direct work: ten points. B separates setup. For a common operation-excluding-setup definition, A becomes 50% and B remains 52%. The initial ranking reverses. Recalculate uncertainty from compatible data rather than merely changing the label.
If A spends 80% of hours on a simple product family and B only 30%, stratify by family or standardise to an explicit common mix. Keep actual plant results alongside adjusted ones and do not extrapolate to families without data.
Statistical design and monitoring
For one proportion under independent Bernoulli records, n=ceiling[z²p₀(1−p₀)/e²]. At z=1.96 and p₀=0.5, five points require 385 records and three require 1,068. Comparing differences with target power needs a specific design, not merely sizing each proportion separately.
Randomisation reduces convenience selection but ensures neither independence nor elimination of observer reactivity. Document routes and missed readings and assess within-round correlation.
Narrower intervals do not establish improvement, and non-significance does not establish equality. Aggregate results may conceal opposing shift conditions.
Turn differences into learning
Investigate methods, supply and scheduling with the people doing the work. Test a transferable practice while monitoring good throughput, quality, risk and cost. Time outside direct work is neither available cash nor surplus staffing.
OEE requires compatible times, ideal cycle time and total/good units; active-person proportions do not reconstruct it. WorkSamp supports observation planning and recording, without promised savings or automatic external-system integration.