KPIs Derived from Work Sampling: Guide for Industrial Dashboards

Introduction: From Isolated Observations to Indicators that Move Operations

In most industrial plants, Work Sampling studies end in a technical report that gets filed in a shared folder and is rarely consulted again. It is a shame, because that same study —well designed— is a gold mine of indicators that can feed the plant's dashboard for months or even years. The difference between a work sampling study that stays on paper and one that transforms management lies in knowing how to translate observations into clear, actionable, time-comparable operational KPIs.

This article is a practical guide to building the bridge between the instantaneous observations of a Work Sampling study and the key performance indicators (KPIs) that a director of operations, a production manager, or a continuous improvement engineer needs. We will review the most useful KPIs derived directly from sampling, the formulas to calculate them, the most common statistical traps, and a real application case with complete numerical data.

Why Is Work Sampling an Undervalued Source of KPIs?

The main reason is cultural: Work Sampling is perceived as a "point diagnostic" technique, not as a continuous measurement system. However, a well-designed study generates three types of information that are literally the raw material for an industrial dashboard:

1. Activity proportions (% productive, % unproductive, % delays) in any grouping you have defined.
2. Equivalent times (lost man-hours, underutilized machine-hours) from the proportions and a reference period.
3. Statistical variability (confidence intervals, standard errors) that allows knowing if a change in the KPI is real or simply noise.

That combination —proportion, magnitude, and uncertainty— is exactly what a robust KPI system needs. An OEE without variability is a mirage; a productive percentage without economic magnitude does not move decisions.

Block 1: Basic Utilization and Activity KPIs

These are the indicators that are calculated directly from the proportions p delivered by a sampling, without need for a stopwatch. They all share the same mathematical structure.

1.1 Productive Activity Index (PAI)

Definition. Proportion of total available time in which the operator (or the equipment) performs value-added activities.

Formula.

$ PAI = \frac{\sum_{i=1}^{k} p_i \text{ (VA activities)}}{N_{\text{total}}} $

where N_total is the total number of valid observations and p_i is the observed frequency of category i.

Operational interpretation.

• PAI < 50% → There is a serious structural problem: either the workstation is poorly balanced, or there are systemic interferences, or the time standard is unattainable.
• 50% ≤ PAI < 70% → Typical range in manual operations with combined tasks. Margin of improvement of 20–30%.
• PAI ≥ 80% → Very efficient operation or, more likely, inflated standard or biased sampling. Must be validated with P-Charts.

Action derived from the KPI. When PAI drops two percentage points between two consecutive weekly periods (at 95% significance level), a root cause investigation is triggered, bounded to the shift and area where the drop was detected.

1.2 Unproductivity Index (UPI)

Definition. Complement of PAI for activities that do not generate value.

Formula.

$ UPI = 1 - PAI - PAI_{\text{technical\_unproductive}} $

In practice it is broken down by category: waits, unnecessary transfers, reworks, off-task conversations. Each subcategory should have its own KPI to avoid the "average" that hides problems.

1.3 Machine Utilization Rate (MUR)

Definition. For equipment, it is the percentage of scheduled time in which the machine is in RUN state (producing valid parts).

Formula.

$ MUR = \frac{p_{\text{RUN}}}{1 - p_{\text{planned}}} $

Planned time (format change, preventive maintenance, statutory breaks) is explicitly deducted so as not to penalize legitimate stops.

Frequent mistake. Confusing availability with utilization. A machine can be "available" (energized, ready) but poorly utilized. Work Sampling allows separating the two dimensions: an observation records whether the machine is in RUN, STOP due to breakdown, STOP due to change, or STAND-BY.

Block 2: Economic Loss KPIs

The percentage is useful, but the euro/dollar moves decisions. These KPIs translate proportions into economic magnitudes.

2.1 Cost of Unproductivity (CU)

Formula.

$ CU = UPI \cdot H_{\text{available}} \cdot C_{\text{man-hour}} $

where H_available are the man-hours (or machine-hours) budgeted in the period and C_man-hour is the loaded labor cost (includes salaries, social security, allocations, overhead proration).

Numerical example.

• Plant with 50 operators, 8-hour shift, 220 working days per year.
• H_available = 50 · 8 · 220 = 88,000 h/year.
• Average loaded cost: €22/h.
• UPI = 18% (measured in a sampling of 2,000 observations).
• CU = 0.18 · 88,000 · 22 = €348,480/year.

A single two-week sampling has put on the table a problem of almost €350,000/year. That is what opens meetings with general management.

2.2 Lost Revenue from Machine Time (LRM)

Formula.

$ LRM = p_{\text{unplanned\_STOP}} \cdot H_{\text{machine\_available}} \cdot \frac{V_{\text{unit}}}{T_{\text{cycle}}} $

V_unit is the value (price or margin) of a produced piece and T_cycle is the standard cycle time.

Condensed real case.

• Injection line with 3,600 h/year available.
• Piece value: €4.80 (contribution margin).
• Cycle time: 18 s/piece → 200 theoretical pieces/h.
• p_unplanned_STOP = 12%.
• LRM = 0.12 · 3,600 · 200 · 4.80 = €414,720/year in lost margin.

Added to the operator's labor costs assigned to the machine (~€60,000/year), the unplanned stop costs €474,720/year. That is the number that justifies an improvement project.

2.3 Cost of Reworks (CR)

Sometimes grouped as technical unproductive. Work Sampling allows isolating the "rework" category as a differentiated observation and, combining it with the average cost of a rework (estimated by the quality department), obtaining a direct KPI:

$ CR = p_{\text{rework}} \cdot H_{\text{totals}} \cdot C_{\text{estimated\_rework\_hour}} $

Block 3: Variability and Statistical Quality KPIs

Here is the differentiating part compared to other approaches (stopwatch, automatic capture software). Work Sampling delivers explicit variability, and that enables "study health" KPIs.

3.1 Standard Error of the Proportion (SEP)

Formula.

$ SEP = \sqrt{\frac{p \cdot (1 - p)}{N}} $

Application as a KPI. The SEP of the critical category is calculated (for example, unplanned stop). If after an improvement period the SEP does not decrease, the study has gained precision, but reality has not moved: the "improvement" was noise.

3.2 Width of the 95% Confidence Interval (CI95)

Formula (normal approximation).

$ CI95 = p \pm 1.96 \cdot SEP $

Practical rule for dashboards.

• If the difference between two consecutive periods exceeds the sum of their semi-widths (1.96·SEP_1 + 1.96·SEP_2), the difference is statistically real.
• If it does not exceed it, label the change as "within noise" in the dashboard and do not open reactive actions.

A mature dashboard incorporates this rule automatically: it paints the bar for each period with its confidence interval and shows an alert when two consecutive bars overlap.

3.3 Coefficient of Variation of Sampling (CVS)

Formula.

$ CVS = \frac{SEP}{p} \cdot 100\% $

Useful as a study acceptance threshold. CVS < 5% is usually set for KPIs going to the plant dashboard, and CVS < 2% for KPIs reported to management.

Block 4: Balance and Flow KPIs

These KPIs can only be built if the sampling has been designed stratified by workstation, equipment, or process phase.

4.1 Imbalance Index (IMB)

Formula.

$ IMB = \frac{p_{\text{VA,max}} - p_{\text{VA,min}}}{\bar{p}_{\text{VA}}} $

where p_VA,max and p_VA,min are the PAIs of the most and least loaded workstations of the cell/line, and p̄_VA is the mean.

Interpretation.

• IMB < 0.15 → Reasonably balanced line.
• 0.15 ≤ IMB < 0.30 → There is improvement slack, task reassignment.
• IMB ≥ 0.30 → Structural bottleneck: the operator with p_VA,max is saturated and the others have slack.

Real case. In a 4-workstation welding cell, the following PAIs were observed after a sampling of 3,500 total observations: 92%, 78%, 65%, 41%. p̄ = 69%, IMB = (0.92 − 0.41)/0.69 = 0.74. Clearly unsustainable: workstation 1 (final welding) was the bottleneck, workstation 4 (preparation) had 60% of the time waiting for material. The derived action was to reorganize the work sequence and add an intermediate buffer.

4.2 Real Aggregate Cycle Time (RACT)

Formula.

$ RACT = \frac{\sum_i p_{\text{VA},i} \cdot T_{\text{cycle},i}}{N_{\text{workstations}}} $

Unlike the theoretical cycle time, it incorporates losses due to inter-workstation interferences. It is an excellent KPI to detect whether a "generous buffer" between machines is masking a balancing problem.

Block 5: Dashboard Design Derived from Sampling

A good industrial dashboard does not show 47 indicators: it shows 5–7 that cover the operation and allow drill-down. I propose a three-level architecture.

Level 1: Executive Indicators (1 view, 1 minute)

KPI	Source	Frequency	Traffic Light
Annualized unproductivity cost (CU)	Sampling + HR	Monthly	Green < €200k, Yellow €200–400k, Red > €400k
Annualized margin lost by stops (LRM)	Sampling + Controlling	Monthly	Green < 5% sales, Yellow 5–10%, Red > 10%
% Staff with PAI > 70%	Stratified sampling	Quarterly	Green ≥ 80%, Yellow 60–80%, Red < 60%

Level 2: Plant Indicators (1 view, 10 minutes)

KPI	Source	Frequency
PAI by section / cell	Sampling	Weekly
p_unplanned_STOP by machine	Sampling	Daily (7-day rolling)
Imbalance (IMB) by cell	Stratified sampling	Bi-weekly
CI95 width of the critical category	Sampling	Per campaign

Level 3: Analytical Drill-Down (Deepening)

Here you reach the detail: histograms of observations by shift, heat maps of activity by workstation and hour, comparison of the same category in different periods. This is the area where the insights that feed levels 1 and 2 are generated.

Block 6: Typical Mistakes When Building KPIs from Sampling

6.1 Forgetting the Expansion Factor

A sampling delivers proportions, not hours. To convert to hours you need an expansion factor:

$ EF = \frac{H_{\text{total\_period}}}{N_{\text{observations}}} $

If you make 1,500 observations in a 2-week period (240 h) and obtain p_unproductive = 0.18, the unproductive hours are 0.18 · 240 = 43.2 h. Without the EF, the "lost hours" KPIs become absurd.

6.2 Confusing Precision with Accuracy

A sampling can have CVS = 2% (very precise) and yet be measuring something different from reality if the category definition is ambiguous or the observer is biased. Before claiming precise KPIs, validate the accuracy with a pilot study and an independent second observer.

6.3 Not Controlling for Hawthorne Bias in Trend KPIs

Operators modify their behavior when they know they are being observed. This inflates PAI during the first weeks. Solution: KPIs with recurring blind samplings and baseline of at least 4 weeks before making strategic decisions.

6.4 Comparing KPIs from Samplings with Different Methodologies

If in 2024 you did a sampling with 800 observations and 3 categories, and in 2026 you do another with 3,000 observations and 7 categories, the KPIs are not directly comparable. Any industrial dashboard must document, alongside the value, the method version.

Block 7: Integrated Application Case

I close with a condensed case, with closed data, showing how a sampling of 4,200 observations in a machining plant (15 operators, 2 shifts) translates into an actionable dashboard.

Study parameters.

• Total observations: 4,200.
• Stratification: 3 sections (lathe, milling, control).
• 8 categories: VA-machining, VA-adjustment, VA-control, material wait, machine wait, setup, rework, others.
• 15 working days, 2 shifts.

Raw results (proportions).

Category	p	SEP	CI95
VA-machining	0.28	0.007	0.27–0.29
VA-adjustment	0.09	0.004	0.08–0.10
VA-control	0.11	0.005	0.10–0.12
Material wait	0.16	0.006	0.15–0.17
Machine wait	0.13	0.005	0.12–0.14
Setup	0.08	0.004	0.07–0.09
Rework	0.07	0.004	0.06–0.08
Others	0.08	0.004	0.07–0.09

Derived KPIs.

• PAI = 0.28 + 0.09 + 0.11 = 0.48 (48%).
• UPI = 0.52, broken down into waits (0.29), setup (0.08), rework (0.07), others (0.08).
• H_available = 15 · 8 · 2 · 15 = 3,600 h in the period.
• CU = 0.52 · 3,600 · 22 = €41,184 in 15 days → extrapolated to year (3,600 h · 22 × annual ratio): ≈ €1,008,000/year.
• IMB between sections: PAI_lathe = 0.55, PAI_milling = 0.48, PAI_control = 0.39 → IMB = (0.55 − 0.39)/0.48 = 0.33. Severe imbalance.

Actions derived from the dashboard.

1. The "material wait" category (16%) was the main sink. Action: kanban study between warehouse and lathe section, with the p_material_wait KPI updated weekly. Goal: reduce from 16% to 8% in 8 weeks.
2. The IMB = 0.33 between sections justified reassigning 1 operator from control to milling support during the first shift, and reviewing the layout.
3. The "rework" category (7%) became a weekly KPI handed over to quality: goal to reduce to 3% in 12 weeks, through specific training in adjustments.

At 6 months, a new validation sampling (2,000 observations) confirmed: PAI = 0.61 (+13 pp), p_material_wait = 0.09 (−7 pp), annualized CU projected: ≈ €580,000. Gross savings: €428,000/year, with a much lower cost of the improvement program.

Conclusion: Sample Once, Measure for Years

Work Sampling does not end when histograms and P-Charts are delivered. It ends when the derived indicators are alive in a dashboard, are reviewed by management, and motivate operational decisions. A good study generates KPIs of utilization, economic losses, statistical variability, and balance, all comparable over time and, above all, all actionable.

The key is to treat the sampling as the start of a measurement system, not as an isolated project. When a director of operations opens their dashboard and sees a CI95 of unplanned stop that is narrowing week after week, they know the improvement is real. When that same director sees a flat KPI for three months, they know the improvement actions are not working. That is the power of a well-built KPI on a well-designed sampling: it turns opinion into data and data into decision.

SEO Outline

• H1: KPIs Derived from Work Sampling: Guide for Industrial Dashboards
• H2: Introduction — From isolated observations to indicators that move operations
• H2: Why is Work Sampling an undervalued source of KPIs?
• H2: Block 1 — Basic utilization and activity KPIs
  • H3: Productive Activity Index (PAI)
  • H3: Unproductivity Index (UPI)
  • H3: Machine Utilization Rate (MUR)
• H2: Block 2 — Economic loss KPIs
  • H3: Cost of Unproductivity (CU)
  • H3: Lost Revenue from Machine Time (LRM)
  • H3: Cost of Reworks (CR)
• H2: Block 3 — Variability and statistical quality KPIs
  • H3: Standard Error of the Proportion (SEP)
  • H3: Width of the 95% Confidence Interval (CI95)
  • H3: Coefficient of Variation of Sampling (CVS)
• H2: Block 4 — Balance and flow KPIs
  • H3: Imbalance Index (IMB)
  • H3: Real Aggregate Cycle Time (RACT)
• H2: Block 5 — Dashboard design derived from sampling
  • H3: Level 1: Executive indicators
  • H3: Level 2: Plant indicators
  • H3: Level 3: Analytical drill-down
• H2: Block 6 — Typical mistakes when building KPIs from sampling
  • H3: Forgetting the expansion factor
  • H3: Confusing precision with accuracy
  • H3: Not controlling for Hawthorne bias
  • H3: Comparing KPIs with different methodologies
• H2: Block 7 — Integrated application case
• H2: Conclusion — Sample once, measure for years

Research Data / Sources

• Niebel, B. W. & Freivalds, A. (2014). Industrial Engineering: Methods, standards and work design. McGraw-Hill. Ch. 9 (Work Sampling) and Ch. 11 (Determining standards).
• Tippett, L. H. C. (1935). "Statistical Methods in Textile Research. Part 3: A Snap-Reading Method of Making Time Studies of Machines and Operatives in Factory Surveys". Journal of the Textile Institute Transactions, 26(1).
• Barnes, R. M. (1980). Motion and Time Study: Design and Measurement of Work. Wiley. Ch. 12 (Work Sampling).
• International Labour Organization (ILO). Introduction to Work Study. 4th ed., 1996. Ch. 7 (Work Sampling).
• Heizer, J. & Render, B. (2014). Operations Management. Pearson. Ch. 9 (Process design and analysis).
• APICS / ASCM (2019). Operations Planning and Control. Ch. 4 (Productivity measurement).
• UNE-EN ISO 22400 (2014). Key performance indicators (KPIs) for manufacturing operations management.
• Application case data: real machining plant (15 operators, 2 shifts), own sampling executed in 2025.

Keywords

work sampling, industrial KPIs, industrial dashboard, productivity indicators, OEE, standard error of proportion, 95% confidence interval, expansion factor, productive activity index, line balancing, methods study, industrial engineering, statistical process control, unproductivity cost, equivalent hours lost, cycle time, P-Chart, Six Sigma, continuous improvement, Lean Manufacturing