Question and scope

This protocol measures the elapsed time between a cross-team request entering a coordination queue and a decision being recorded in the system of record. A decision is recorded when an authorized person writes approve, reject, defer, or change scope with a date and a reason. The measure is not the time spent doing the work, and it is not the time until someone replies in chat. It isolates the waiting period that the requesting team cannot see or control.

Scope is one request type handled by one coordination team during a defined window, such as eight consecutive weeks. Include requests that need a budget, access, priority, or scope decision from another department. Exclude requests that only need a routine acknowledgment or a status update. State the request type and the window in the report so another reviewer can reproduce the boundary.

Start and stop clocks

Start the clock when the request first appears in the intake queue with enough information to be triaged. If the request arrives incomplete, record the first timestamp and a second timestamp for the resubmission that made it triageable. Report the two intervals separately, because a slow requester and a slow reviewer create different problems. Do not start the clock at the first chat message unless that message is also the formal intake record.

Stop the clock when the decision is written in the system of record, not when a meeting ends or a verbal agreement is reached. If the decision is made verbally and recorded two days later, use the recording time and note the gap. If a request is withdrawn, stop the clock at withdrawal and mark the outcome rather than treating it as a decision. The stop rule must be identical for every request in the sample.

Unit of analysis and exclusions

The unit is one request tied to one requester, one decision owner, and one requested outcome. Split a bundled request when each part needs a separate decision or a separate approver. Keep a bundled request as one unit only when the parts cannot be approved independently. Record the original wording so the unit can be checked later without relying on a summary.

Exclude duplicates, test records, and requests that were opened only to reserve a placeholder. Exclude requests that were cancelled by the requester before triage, but report the count and the reason. Do not exclude a slow request simply because it was eventually rejected, because rejection is still a recorded decision. Every exclusion needs a written reason and a reference to the record.

Definitions and measures

A cross-team request is a request for a decision or resource that the requester's own team cannot supply. A recorded decision is a dated entry by a named person with one of the agreed outcomes and a short rationale. A decision owner is the person authorized to choose, not the coordinator who schedules the conversation. These definitions should be short enough to apply consistently without a glossary open on the desk.

Report the median and the 90th percentile wait in business hours, not calendar days. Business hours should follow the coordination team's published schedule and exclude weekends and recognized holidays. Show the count of requests behind each figure and the number still open at the end of the window. A median without a count is easy to misread, especially when the sample is small.

Evidence and data dictionary

Each row needs a request reference, the intake timestamp, the triage timestamp, the decision timestamp, the decision outcome, and the decision owner. Add the requested date and the date the requester says the outcome is needed. Include the queue or system where each timestamp was captured so the source can be audited. Store the export beside the coding sheet rather than inside a chat thread.

The data dictionary should define triageable, recorded, deferred, withdrawn, and still open. It should also define how to treat a decision that changed after it was first recorded. If a decision was reversed, keep both entries and mark the later one as the current outcome. A one-page dictionary with two examples per category is enough for most coordination teams. Keep it in the same file as the coding sheet.

Sampling and test design

Use a census when the queue produces fewer than roughly one hundred requests in the window. Above that volume, take a random sample and document how it was drawn from the queue export. Stratify the sample by requesting team or request type so one high-volume group does not dominate the result. Freeze the sample before looking at the timestamps to reduce selection bias.

Have a second reviewer recode a subset of at least ten requests and compare the results. Record disagreements about start time, stop time, or outcome before resolving them. If the disagreement rate is high, rewrite the definition and recode the full sample. A measurement that two trained reviewers cannot apply the same way is not ready for a management report.

Decision rules before the test

Decide before coding how to handle requests that were reopened after a decision. One option is to keep the first decision as the stop point and treat the reopen as a new request. Another is to restart the clock at the reopen and report both intervals. The choice changes the result, so write it down before the first row is coded.

Agree on the action that follows a long median wait. A practical response is to set a decision service level by request type and route anything that breaches it to a named reviewer. Another is to publish a weekly list of requests waiting more than five business days. Avoid a general demand for faster decisions without a named owner, because the delay usually sits in an unassigned handoff.

Analysis and reporting

Report the median wait, the 90th percentile, the share decided within the agreed service level, and the count still open. Break the result down by decision owner and requesting team, but suppress any group with fewer than five requests. Pair the timing with the outcome mix, because a fast rejection and a fast approval have different operational consequences. The report should help a manager find the handoff where requests stall.

Show the age of the oldest open request and the date it entered the queue. A single old request can reveal a decision owner who is unavailable or a request that was never actually triaged. Do not remove it as an outlier unless the exclusion rule was written in advance. The oldest item is often the most useful evidence in the report.

Reporting template

Use a one-page template with five blocks: scope, median and 90th percentile wait, share within service level, oldest open request, and one action for the next cycle. Put the raw counts beside every percentage and name the source of each timestamp. Add a short note for every exclusion and every change to the clock rules. The template should be readable in a five-minute review.

Store the completed template with the queue export, the coding sheet, and the sample list. A later reviewer should be able to rebuild the headline figure without asking for context. If the request type or the system changes, start a new series instead of merging the old and new data. A clean break is more honest than a trend line built from two different processes.

Limits and sources of error

The cited public sources discuss records management, internal control, and operational practice in general terms; they do not provide a benchmark for a specific queue. A recorded decision can still be a poor decision, and a short wait can reflect a rubber stamp rather than good coordination. The measure also depends on whether the system captures the true first contact. A request that arrived by phone and was entered later will look faster than it was.

Backfilled timestamps and edited records are common sources of error in retrospective measurement. Check a sample of entries against the original messages and mark any row that was reconstructed. Exclude a row when the start or stop time cannot be verified, but report how many rows were excluded. A precise number built on guesses is worse than a range with a clear caveat.

Review cadence

Review the result at a fixed point, then repeat it after two full cycles before setting a permanent target. A service level that nobody can meet will be ignored, and a target that is too easy will hide the handoff that still fails. Keep the clock rules and the sample method in the same document as the result. Compare like periods and mark any process change on the timeline.

Retire the measure if it stops informing a decision, because an unused metric will be maintained badly. If the team keeps it, publish the median and the oldest open request together so speed and neglect are visible at the same time. The purpose is a decision that someone can act on, not a perfect record. Stop when the team can show that outcome consistently.

Sources