Define the clock before you start it

Abstract

Decision latency is the elapsed time between a request reaching the executive and a recorded decision, yet most offices never measure it because the obvious methods feel invasive. A support desk can use timestamps that already exist in email, calendar, and task systems to produce an aggregate view without monitoring individual behavior. This review defines the clock, separates queue time from review time, and sets guardrails for privacy and fairness. [1][2][3] The design is deliberately narrow so the number can be repeated and audited.

Methodology

The review uses internal-control and privacy guidance as the basis for a measurement design, not as a source of benchmark numbers. It defines a decision as a recorded choice with an owner and a date, and it uses a sample of requests rather than continuous monitoring. Timestamps are drawn from systems the office already uses, and results are reported at the aggregate level. Individual scores are excluded by design. [1][2][4]

Define decision latency before measuring it

A decision has a start and an end, and both need a definition before anyone measures the gap. The start could be the moment a complete request reaches the executive, while the end could be the moment a choice is recorded in the decision log. If the start is the first email in a thread, the measurement includes time spent clarifying the request. That may be useful, but it is a different number and should be labeled accordingly.

The clock should also exclude periods when the decision could not have been made. A request that arrives at the start of a two-week board recess is not late in the same way as one that arrives during a normal week. Record the date the request became complete and the date the decision was due. The difference between the actual decision date and the due date is often more useful than raw elapsed time. [1][3]

Why surveillance fails

A desk can measure response time by watching every message, but that approach changes the behavior it is trying to understand. People optimize for the metric, rush decisions, or avoid recording difficult ones. The measurement also creates a privacy problem because it captures message content and individual work patterns. A support function depends on trust, and a surveillance system spends that trust quickly.

Surveillance also produces misleading data. A fast response to a simple request looks better than a slower response to a complex one, even when the second decision is more valuable. An individual score ignores the queue, the dependencies, and the quality of the input. The measurement should describe the system, not rank the people inside it. [2][4]

Use timestamps that already exist

Most offices already record when a request arrived, when it was assigned, and when a decision was posted. Use those timestamps rather than asking people to start a stopwatch. A task system may record status changes, and a calendar may show when a decision meeting occurred. The goal is a lightweight extract that can be repeated without adding administrative work.

If the existing records are incomplete, improve the record before expanding the measurement. Add a required decision field to the task template or a decision date to the meeting note. Do not create a parallel tracking sheet that only the support desk maintains. The measurement should ride on the operating record, not become a second job. [1][4]

Separate queue time from review time

A decision can wait in a queue before anyone looks at it, then wait again while the executive considers it. Those two periods have different causes and different fixes. Queue time may reflect an unclear request, a missing attachment, or a backlog at the desk. Review time may reflect competing priorities, missing analysis, or the need for a meeting. Combining them hides the bottleneck.

Record the handoff points so the two periods can be separated. A request that sat for three days because it lacked a budget figure is a different problem from one that sat for three days because the executive was traveling. The first calls for a better intake form, while the second calls for a coverage plan. A single average would obscure both. [2][3]

Sample decisions rather than monitor people

Choose a sample of decisions from a defined period, such as one month or one quarter. Include decisions of different types and sizes so the sample is not dominated by routine approvals. Review the sample with the desk lead and the executive, and agree on the definitions before extracting the data. A sample is less invasive and often more instructive than a full monitoring system.

Record the sample size and the selection method in the report. If the sample is small, say so and avoid strong conclusions. If a category is missing, note the gap rather than filling it with an assumption. The purpose is to understand where decisions stall, not to produce a comprehensive audit of every choice. [1][4]

Report distributions, not individual scores

Report the median, the range, and the share of decisions that missed their due date. A median alone can hide a long tail of decisions that waited far too long. Show the distribution by decision type and by quarter so the reader can see whether the pattern is stable. Avoid a single composite score that mixes unrelated decisions.

Aggregate the data so that no individual can be identified from a small cell. If a category has only one or two decisions, combine it with a broader group or omit it. Report the limitations alongside the result, including missing timestamps and excluded categories. A transparent method is more credible than a precise number with no context. [2][4]

Test one change at a time

Use the baseline to choose one intervention, such as a clearer intake form, a daily decision window, or a standing review for low-risk approvals. Run the change for a defined period and compare the same decision types. Do not change the intake, the meeting cadence, and the approval threshold at once, because the result will be impossible to interpret. A written baseline also makes it harder to attribute an improvement to an unrelated change.

Set the success measure before the test begins. A shorter queue time is useful only if the decision quality holds. Add a check for rework, reversals, and missing information. If the change shifts work to another team without improving the decision, the test has moved the problem rather than solved it. [1][3]

Guardrails for privacy and fairness

Limit the data to timestamps, decision type, and due date. Exclude message content, recipient identity, and after-hours activity unless the office has a separate, lawful reason to collect them. Restrict access to the aggregate dataset and keep the detailed extract only as long as needed. Document the purpose and the retention period before the first extract.

Give the executive and the desk a chance to correct the record before it is used in a planning discussion. A timestamp may be wrong because a decision was made verbally and recorded later. A fair process allows that correction and notes it in the report. The measurement should support a conversation about the system, not a judgment about a person. [2][4]

Limits and responsible use

Decision latency cannot capture the quality of a decision, the value of deliberation, or the cost of a wrong choice. A fast decision can be worse than a slow one. The measure is a diagnostic for flow, not a performance rating. It should be used alongside quality checks, rework counts, and the judgment of the people doing the work.

The method also depends on reliable records. An office with inconsistent task tracking will produce an incomplete picture. Start with a small sample and improve the record before expanding. For related reading, see our guide to <a href="/blog/build-an-executive-decision-log-that-survives-staff-turnover">building a decision log that survives turnover</a> and our <a href="/services">executive support services</a>. [1][3]

Key findings

  • Define the start and end of the decision clock, and label the result so it is not confused with request-response time. [1][3]
  • Separate queue time from review time because the two periods have different causes and fixes. [2][3]
  • Use timestamps that already exist and sample decisions instead of monitoring every message. [1][4]
  • Report distributions and aggregate categories, excluding small cells that could identify an individual. [2][4]
  • Test one change at a time and pair latency with quality and rework measures. [1][3]

Sources

  1. Standards for Internal Control in the Federal Government, U.S. Government Accountability Office
  2. Security and Privacy Controls for Information Systems and Organizations, NIST
  3. NIST Privacy Framework, NIST
  4. O*NET OnLine Summary Report 43-6011.00, O*NET OnLine

Return to the Research Content library