Operations & Systems

Throughput vs Busy Time: Count Finished Work Before Calling a Team Productive

A weekly report shows more recorded hours. The open request list still needs an explanation.

Hours describe one part of the week. A finished-work count describes another. Put both on the report with their definitions intact.

The May 2025 Kanban Guide defines throughput as an exact count of work items finished per unit of time. The count does not assign equivalent effort or business value to those items.

The worksheet below proposes a small-team reporting convention. Every request and hour is fictional. Use the example to inspect a definition, not to set a staffing target or judge an employee.

Define the Work Item Before Counting It

For this example, one work item is one accepted service request with a stable request identifier. A request can contain several internal tickets. Those tickets remain supporting records under the same request.

The request starts when the team accepts it into delivery. It finishes when the agreed delivery conditions are met and the reviewer records acceptance. “Draft sent” and “work performed” do not meet this finish rule.

Keep the acceptance event reference beside the request ID. The report needs evidence that the finish occurs inside its period. A current status of Finished cannot establish the date of that event by itself.

Exclude test requests, duplicates confirmed by source identity, and cancellations. Record cancellations separately. Stopping unwanted work can be a responsible decision, but this proposed count measures accepted delivery.

Record the definition before viewing the result. Otherwise a convenient status or ticket count can quietly become the work unit. Use the business metric dictionary to keep the approved meaning attached to the displayed number.

The request boundary belongs to this example. A different workflow can count another unit. Make that unit visible whenever you compare results.

Fix the Period and Reopened-Item Rule

Use two equal periods in UTC. Week one starts at 2026.10.12 00:00 and ends before 2026.10.19 00:00. Week two starts at that ending boundary and ends before 2026.10.26 00:00.

Include a finish event at the starting boundary. Exclude one at the ending boundary. That convention assigns a boundary event to exactly one period.

For this proposed report, count each distinct request's first qualifying finish once. If that request later reopens for a correction, retain its identifier. Do not count another finish for it in this series.

Show reopened requests and their later finishes in a separate correction note. A repeat-finish count can answer another question. Label it separately so it cannot inflate the distinct-request series.

The rule keeps the first-finish history visible. It does not establish that the original result remains acceptable. If acceptance evidence is invalidated as an error, issue a labeled correction to the affected report.

Retain both the original result and the corrected version when that happens. State the changed evidence and affected request. Reopening for more work and correcting a false finish are different events under this convention.

Compare Two Fictional Weeks

The fictional team handles page corrections, review packets, and import specifications. All three types use the request-level finish boundary above. Their size, complexity, and value can differ.

Activity hours mean hours entered in the team's fictional activity log during the period. They include work on both finished and unfinished requests. They exclude unrecorded time by definition. No claim about log completeness is made.

Observation Week one Week two
Recorded activity hours 32 40
Page corrections first finished 4 2
Review packets first finished 2 2
Import specifications first finished 0 1
Total distinct requests first finished 6 5
Previously finished requests receiving another finish after reopening 0 1

Week one's six request identifiers are W101 through W106. Four are page corrections and two are review packets. Each has one qualifying first-finish event inside the week.

Week two adds five identifiers, W201 through W205. Two are page corrections, two are review packets, and one is an import specification. Each meets the same finish condition inside week two.

W102 reopens during week two and receives another finish. It remains among week one's six first finishes. Its week-two correction appears in the separate note, contributing zero to week two's first-finish count.

The arithmetic is four plus two plus zero equals six. Two plus two plus one equals five. Week two records eight more activity hours and one fewer first-finished request.

Those differences do not establish a cause. They do not tell you whether week two delivers less value. The work mix changes, and the table contains no common value measure.

Read Busy Time Within Its Boundary

The activity log says forty hours are recorded in week two. It does not say every available hour is occupied. Availability requires another definition and another source.

A utilization measure would need an explicit numerator, denominator, and eligibility rule. For example, choosing which activities belong in the numerator is a business decision. This article supplies no preferred utilization target or staffing formula.

Keep the label “recorded activity hours” in the example report. Calling it utilization would imply a denominator the example never supplies. Calling it effort required by finished work would also misstate the population.

Some week-two activity can belong to requests finishing later. Some week-one finishes can depend on activity recorded before week one. Dividing the displayed hours by the displayed count does not reconstruct request-level effort.

If the team needs effort by request, inspect activity attribution and its coverage separately. Carry missing records as unknown. Do not distribute hours evenly across completed requests because a chart needs a value.

An operator dashboard can hold several measures beside one another. Each measure still needs its own population and interpretation. Sharing a row does not make hours, requests, and outcomes interchangeable.

Watch What Ticket Splitting Does to the Count

Consider W205, the fictional import specification. Its agreed delivery is one accepted specification covering fields, sample records, and review conditions.

In one arrangement, the team tracks that request in a single ticket. In another, it creates separate tickets for the field map, sample, and review checklist. All three support the same accepted request.

Under the proposed request-level definition, both arrangements contribute one. The accepted delivery boundary and request identity remain unchanged.

If a report counts internal tickets at their first qualifying finish instead, the second arrangement can contribute three. Assume all seven tickets first finish during week two. W102's repeat finish remains excluded. That alternative count becomes seven: four other requests represented by one ticket each, plus three tickets for W205.

Five accepted requests and seven first-finished tickets describe different units. The increase from five to seven does not prove additional delivered value. It can follow entirely from how the team divides its records.

Keep the parent request ID on each internal ticket. Preserve the ticket count if it helps coordinate work. Name it “internal tickets first finished” and keep it outside the request-throughput series.

If the actual service boundary changes into three independently accepted deliveries, document a new counting rule. Treat that as a definition change. Do not present the new series as directly comparable without examining the old population.

Copy This Throughput Definition Card

Use one card for each reported series. The entries below define the fictional example, rather than a rule every service team must adopt.

Field Proposed entry
Population Accepted service requests in the team's delivery workflow
Work item One stable request ID; internal tickets remain children
Completion event Delivery conditions met and reviewer acceptance recorded
Period Seven days in UTC; inclusive start and exclusive end
Counting rule Distinct request IDs with a first qualifying finish in the period
Exclusions Test records, confirmed duplicates, cancellations, and child tickets
Work mix Page corrections, review packets, and import specifications shown separately
Reopened rule Retain ID; later finishes appear in a correction note without another first-finish contribution
Source evidence Request ID, request type, acceptance reference, event timestamp, and event history
Missing evidence Unknown coverage or missing finish time produces an incomplete result state
Definition owner Named person approving boundaries and changes
Interpretation limits Counts do not establish equal effort, equal value, quality, causality, or individual performance

Add the definition version and actual extraction timestamp above the card. Keep the report period separate from the time you obtain its records.

If your source cannot distinguish first finishes from later finishes, mark that limitation. A current-state export alone may leave the reopened rule unimplementable. Resolve the history question before labeling a result comparable.

Keep a Count Separate From a Score

A team may want to give an import specification more weight than a page correction. That proposal creates another measure with assumptions about weights.

Keep it outside the throughput label used here. The Kanban definition cited above calls for an exact work-item count. A weighted result needs a name, a version, and an explanation of what the weights represent.

For instance, assigning two points to each specification would change week two's score while leaving its five-request count intact. The chosen weight does not demonstrate twice the effort or twice the value.

Do not silently replace the count on a historical chart with that score. Preserve the old unit. Start a clearly labeled series when a changed measure serves a different question.

Likewise, the team total cannot isolate one person's performance. This fictional count contains shared review, prior-period work, and several request types. It has no evidence that attributes delivered outcomes to an individual.

Use the result to choose a question about the workflow. “Which acceptance events need inspection?” is answerable from the records. “Who works hard enough?” is not established by this table.

Review the Evidence Before Explaining the Change

Open one ordinary request from each week. Trace its stable ID through acceptance and the source timestamp. Confirm that it contributes once under the card's rule.

Then inspect W102's reopening history. Confirm that the second finish stays outside week two's first-finish count. Read the correction note so the reader knows the original delivery required more work.

Review work mix next. Week two contains a specification absent from week one. Keep that visible when discussing the totals. The table provides a comparison worth investigating, not an explanation already proved.

Finally inspect unfinished work separately. Five finished requests do not establish the size, age, or condition of what remains. The dispatch system holds the queue; this report counts qualifying exits during a fixed period.

When source coverage fails, publish an explicit incomplete or unavailable state. A missing export cannot establish zero finishes. When the definition changes, preserve the boundary before comparing another week.

Count the finished work. Keep its meaning attached.

Sources

  • The Kanban Guide, May 2025. Throughput as an exact count of finished work items per unit of time and workflow-defined boundaries. Checked 2026.10.28. Reporting rules, the definition card, and every record are editorial proposals.

Topic map: Business reporting and statistics

Articles in this cluster, then the 3 clusters that share the most topics with it.

This cluster: Business reporting and statistics

Neighbor: Spreadsheets, CSV files, and data handoffs

Neighbor: Records, access, and continuity

Neighbor: APIs, webhooks, and automation