The completion log shows a slower interval after lunch, while the machine display still shows its usual cycle time. Both records may describe valid measurements with different boundaries. The Lean Enterprise Institute defines cycle time through actual measurement of producing a part or completing a process. For a production review, specify the repeating event you observed before dividing elapsed time by output.
An observed completion interval measures elapsed time between consecutive equivalent completions. A machine cycle can instead measure the machine’s operations on one piece, while an operator cycle follows the repeating work sequence. Retain that scope in the label. Otherwise, a gap caused by loading or waiting can disappear when a machine-only duration becomes the reported line performance.
Choose a repeatable event and the decision it supports
For a final assembly output check, you might time consecutive accepted completions at the end of the line. To study manual work, observe the operator returning to the same point in the sequence. To investigate equipment behavior, capture the machine’s declared cycle start and finish. Each approach needs the resource, product, and event definition alongside the duration.
Write down what the boundary includes before timing. Loading, unloading, inspection, waiting for material, and changeover can enter different records under different definitions. Keep excluded elements in separate observations when they affect the decision. A faster automatic cycle does not resolve an output gap if the machine waits for loading between runs.
Use timestamps from a common time basis. If two devices use different clocks or time zones, reconcile that difference before subtracting events. Preserve the original timestamps and the adjustment rule so a negative duration remains an exception to investigate. Substituting an absolute value would turn a clock error into a plausible measurement.
Count elapsed intervals rather than timestamp rows
Assume a hypothetical single-piece process with no parallel resources. Equivalent completion events occur at 09:00:00, 09:01:20, 09:02:50, 09:04:10, and 09:05:40. The five timestamp records define four intervals. The measurement window starts with the first completion, so it does not include the time spent producing that first piece.
| From completion | To completion | Observed interval |
|---|---|---|
| 09:00:00 | 09:01:20 | 80 seconds |
| 09:01:20 | 09:02:50 | 90 seconds |
| 09:02:50 | 09:04:10 | 80 seconds |
| 09:04:10 | 09:05:40 | 90 seconds |
The measured span is 340 seconds and the mean interval is 340 / 4 = 85 seconds per completion. Dividing by five gives 68 seconds, which assigns the first completion an interval that was never observed. Retain the start event, end event, and interval count with the result so someone can reproduce the denominator.
A different method can use a window defined before the first counted output and after the last counted output. Then elapsed production time divided by the completions within that window gives a window-average time per completion. Declare its boundary and inclusion rules; it will not always equal the average of selected timestamp gaps. Pieces already in process at the window’s start can affect interpretation.
Show the effect of a delay without deleting it
Extend the first hypothetical event log with a sixth completion at 09:10:40. The next interval is 300 seconds, making the full span 640 seconds across five intervals. Its mean is 128 seconds. The original four intervals still average 85 seconds; the extra delay explains the change in observed output pace.
If you need a running-cycle study, separate the 300-second interval under a declared rule and retain it as a loss event. Show the inclusive output interval alongside the selected running measurement. Removing the delay from the only reported result would leave the production review unable to explain why output fell.
Do not infer the cause from the long interval alone. A completion gap could include a breakdown, inspection hold, unavailable material, or a missed event. Match the interval with operator notes and resource status evidence before assigning a reason. If no evidence identifies the cause, preserve an unclassified delay rather than manufacturing a diagnosis.
Inspect a production measurement with source evidence
Start with an accessible completion export and explore how event boundaries and accepted definitions can support a repeatable report.
Keep parallel output and batch duration distinct
Suppose two independent machines each complete one piece every 120 seconds, offset so the combined output arrives every 60 seconds. These are hypothetical idealized conditions with no stops or defects. The combined stream’s completion interval is 60 seconds, while each machine’s repeating cycle remains 120 seconds. Dividing the combined interval among machines would misstate their measured work.
For a hypothetical oven completing 12 pieces in a 24-minute batch, the equivalent average is two minutes per piece. Each piece still occupies the oven for the batch duration, and completions emerge as a group. Retain batch quantity and elapsed batch time (including the stated loading boundary) when comparing alternative plans. An equivalent per-piece value alone cannot show when individual units will become available.
State how partial batches enter the summary. If a batch produces fewer pieces, calculate its observed output using its actual count. Do not reuse the full-batch multiplier without confirming the quantity. Keep rejected pieces distinguishable from accepted output so a lower good-output rate does not get mistaken for a longer mechanical cycle.
Summarize comparable work before comparing shifts
Group observations by the product, routing revision, resource, and work method when those conditions change the work. A shift making heavier assemblies can show a higher average while each product’s process remains unchanged. Keep product-level measurements available before claiming that a staffing or improvement action changed performance.
If you combine groups, calculate from compatible elapsed time and output counts. A simple average of shift means gives equal influence to a short run and a long run, even when their observed interval counts differ. For an interval-weighted mean, preserve each group’s sum of measured intervals and interval count; do not blend overlapping windows or incompatible boundaries.
Report the spread as well as the mean when the decision depends on repeatability. The example’s 80- and 90-second observations differ from a log with short bursts and one long delay even if an average happens to match. A short observation window establishes what happened in that window. It does not establish a stable production standard without a review of representative operating conditions.
Compare cycle time with demand pace and elapsed flow
The takt time calculation sets a required pace from demand and available time. Cycle observations supply evidence about the work. A target faster than the observed interval calls for a feasibility review using matching boundaries and output units; replacing the observed value with the target would hide the gap.
Lead time spans a declared start and finish across a flow, including waits within that boundary. A piece can move through a station on a short cycle and still wait before or after it. The published lead time example uses purchasing events, where order release, receipt, and usable stock form different elapsed-time measures. Name the boundary whenever those terms appear together.
An ideal cycle standard also has a separate role. The overall equipment effectiveness formula uses an accepted ideal cycle input to evaluate performance. Substituting an observed average changes that benchmark and can conceal speed losses. Retain the standard’s source and effective period alongside the measured result.
Permute: traceable completion records across reports
Permute sells the context layer between business data and the AI tools people already use. We connect approved production exports and related records from sources such as Excel and NetSuite, reconcile resource and product references, and preserve the evidence behind a calculation. Your chosen event boundaries and exception rules determine what the output means.
We implement accepted definitions so a supervisor can reuse the same measurement in a report or application. A question about a slower shift can return the relevant records and their scope, subject to current permissions. Claude, ChatGPT, or Copilot can work with that prepared context while the source evidence remains available for review.
Keep observation design and process approval with the plant
We do not replace machine controls, certify equipment capability, or determine a safe work method. This analytical workflow reads connected records and runs declared calculations; it does not change machine settings or release production instructions. Incomplete events and unknown delay causes remain visible until the responsible team resolves them.
Cycle time supports a production decision when it names the measured resource, repeating event, and included conditions. The completion example averages 85 seconds across four observed intervals, then rises to 128 seconds when the delayed interval enters the same boundary. Parallel machines and batch output need their own labels. Preserve those distinctions before using a shorter average as evidence of improved work.
Review the evidence behind a cycle-time report
Bring completion events, resource references, and the rules used to classify delays. We can review what a repeatable measurement needs from the connected data.
Questions about cycle time
How should I handle a missing completion scan?
Flag the affected interval and seek independent evidence of the missing event. Keep the original log available if a correction is accepted. Without that evidence, the interval cannot establish whether production slowed or a timestamp was lost.
How many observations are enough?
Choose a measurement plan that covers the operating conditions relevant to the decision, including product mix and restarts where they apply. A small sample can document those cycles but cannot prove the whole shift behaves the same way. Keep the observation window and conditions visible when reviewing the result.
Should a repaired piece become another new completion?
Preserve its identity and distinguish the repair event from its first pass. A throughput view may need to record the additional work, while a first-pass quality view must retain the original failure. Define the count for each report before combining their totals.