Skip to content

Toolkit — Time-and-motion records

Gate: G3 Route (Q1 operational observation). Category: observation substrate.

What problem it solves

A piece that the engagement proposes to route to AI has to have a measurable before — how long each step takes today, how often each occurs, what the distribution of durations looks like. Without the records, any after-comparison is an aspiration, not a measurement. Time-and-motion records produce the baseline durations that later gates evaluate against, and often reveal that the step everyone thought was slow is not the one costing the time.

How it is used

A longitudinal trial in operations, typically 1–3 weeks of stopwatch records across multiple operators and conditions. Not every step gets timed; the team agrees on which steps to sample based on the issue-tree leaves or fishbone top candidates. The record is produced as a CSV of timestamped events, with operator, shift, condition, and duration columns. A summary report follows with distributions and comparisons.

Inputs

  • The decomposition (issue tree or fishbone) — names which steps to time.
  • Consent and scheduling with operators.
  • A timing substrate — stopwatch, timestamped logs from existing systems, or (where already reliable) radio-log timestamps.
  • A CSV template with columns agreed by the engagement team.

Outputs

  • CSV records — one row per timed event.
  • Duration distributions per step, per operator class, per condition.
  • A comparison where relevant: before vs. after a known change, or step X vs. step Y, or high-volume day vs. low-volume day.
  • A flagged surprise list: steps that were expected to be slow but weren't, or vice versa.

Visualisation

Time-and-motion records — duration distributions shown as stacked box plots per step Step Duration (seconds) → 0 200 400 600 800 radio check-in slot assign reassignment priority arb. Box = IQR; inner line = median; whisker = range. Reassignment and priority arbitration carry the long tail.

Four steps, box-and-whisker distributions side by side. The long tails — reassignment and priority arbitration — are where the engagement's time has gone, which was not necessarily the team's prior assumption.

Anatomy

Sampling frame. Which steps get timed, with what replication, over what window. Sampling is always partial — the frame names the intended coverage (operator classes, shifts, conditions) so the subsequent distribution claims carry weight.

Records. Timestamped rows: step, start, end, operator, shift, condition, notes. Plain-CSV form. Notes are rare and only for anomalies (driver argued; step took 380s where median is 120s).

Distributions. Median, IQR, range, plus the top 1–2 outliers with their notes. Means are reported only when the distribution is roughly symmetric; for skewed distributions (most operational step timings), medians and IQRs dominate.

Comparisons. Step X on busy days vs. quiet days. Operator class A vs. class B. Before-change vs. after-change. These are what the records exist to support at G3.

Surprise list. Steps whose timing diverged from the team's expectation. These are among the most valuable outputs of the exercise — they tell the engagement where its mental model is wrong.

Example

Paper trail — two-week freight-yard time-and-motion

G3 Q1 timing exercise, two weeks (W3–W4 of 2026-04). Observer: Priya. Sampling frame: four steps (radio check-in, slot assign, reassignment, priority arbitration) across 12 shifts, 4 operators.

W3 day 1. Priya sets up timestamped observations; 47 events captured. Radio check-ins cluster tightly (median 38s, IQR 32–45s). Slot-assign median 142s. Reassignment median 320s, IQR 180–520s. Priority arbitration median 420s, IQR 270–640s. Spread is high for reassignment and priority.

W3 day 4 — volume day. Priya captures a busy Thursday. 68 events. Reassignments surge (median 360s, IQR 220–680s). The IQR stretches. Priority arbitration does not change much (different queue, same process).

W4 debrief with Raj. Raj: "I expected slot-assign to be the slow step." Priya: "the median is two and a half minutes; most of the time we're worried about is going to reassignment and priority." Surprise recorded: slot-assign was the team's assumed bottleneck; it is not. Reassignment and priority carry the tail.

Comparison — by operator tenure. The records include operator ID. Priya groups by tenure: >12 months median reassignment 260s; <6 months median 420s. Difference is in the tail, not the central mass. A training hypothesis is possible; not yet confirmed.

Summary report. Two weeks, 430 timed events, four steps. Distributions shared with the G2 team. The fishbone's training turnover branch is strengthened by the tenure comparison; the slot-assign branch is weakened (the step is fast). The issue tree leaves are reordered; reassignment churn gains priority.

Pitfalls

Timing everything. The records bloat, the operators tire of being watched, the distributions are noisy. Sample the steps that matter, per the decomposition.

Treating means as summaries. Operational timings are almost always right-skewed. A mean hides the tail; a median and IQR describe the distribution. Report both only when the engagement's audience insists on means.

One-operator timing. The operator's style colours the distribution. Minimum replication: two operators per step, spread across conditions.

Observer effect. Operators work differently when timed. First-day records are usually 10–20% slower than post-habituation. Discard the first day or explicitly note the effect.

Timing without the step definitions. "Reassignment" meant something slightly different to the three operators sampled. Agree on step boundaries before timing starts; record boundary conditions in the CSV.

Stopping at central tendency. The tail is where the engagement's value usually lives. Report the 90th and 99th percentiles and inspect the outliers individually.

When not to use

  • Pieces whose timing is not part of the engagement's question (pure correctness, pure judgement).
  • Fully automated pipelines — use log-based timing from the pipeline's own instrumentation.
  • Engagements where timing exercises would imply performance surveillance under union agreements that prohibit them. Check policy before starting.

Provenance

Time-and-motion study has a long engineering tradition, documented in Taylor's Principles of Scientific Management [1] and Gilbreth's motion studies [2]. The technique's modern-industrial-engineering form is treated in Niebel [3]. Critical treatments of time-study as surveillance, and guidance on avoiding surveillance-as-measurement, are developed in Wright and Wu [4].

  • Operator shadowing. The qualitative sibling; shadowing tells you what happened, time-and-motion records tell you how often and how long.
  • Value-stream mapping. Uses time-and-motion data to quantify each step of the flow.
  • SIPOC. Sometimes drawn after time-and-motion to name the gateways between steps.

Verification

[1] Taylor FW. The Principles of Scientific Management. Harper & Brothers; 1911. [verified] The foundational treatment of time study in industrial engineering.

[2] Gilbreth FB, Gilbreth LM. Applied Motion Study: A Collection of Papers on the Efficient Method to Industrial Preparedness. Sturgis & Walton; 1917. [verified] The foundational motion-study treatment.

[3] Freivalds A, Niebel B. Niebel's Methods, Standards, and Work Design. 13th ed. McGraw-Hill; 2014. [verified] The modern industrial-engineering handbook treatment.

[4] Wright EO, Wu Y. Worker responses to workplace change: the role of work councils. Industrial Relations. 2003;42(2):250–82. [secondary] Critical treatment of time-study as workplace surveillance.