Developer Offshore logoDeveloper OffshorePowered by StealthAgentsScope a developer
Table of contents

Developer Offshore research

Change lead time decomposition in outsourced delivery

An analysis of change lead time in outsourced delivery: the start and end, the four segments, wait versus work, handoff effects, measurement pitfalls, and a practical protocol.

Use this report with the Research library and related developer guides to turn evidence into a bounded work brief.

Change lead time decomposition in outsourced delivery

Key stats

  • 4 segments: queue, active work, review, and release
  • 2 boundary rules: a stable start and an end that reflects real use
  • 3 wait reasons: capacity, dependency, and decision
  • 1 rule: report the median and the tail, not only the average

Key takeaways

  • Fix the start and end of the clock before comparing teams or periods.
  • Decompose the total into queue, active work, review, and release so the largest wait is visible.
  • Record why a change waited, because capacity, dependency, and decision delays need different fixes.
  • Report the median and the share above a threshold, because an average hides the slow tail.
  • Change one segment at a time and pair the result with change failure rate so speed does not replace control.

Question and scope

This report examines change lead time as a delivery signal in outsourced software work. The unit is one change from the moment it is ready to be worked on to the moment it is running in production. The scope covers the waiting and working time inside that path, including review, testing, approval, and release. It excludes discovery work that precedes a defined change.

The question is whether a buyer can use lead time to find coordination cost without turning it into a speed contest. A short lead time is not automatically good if it skips review, and a long one is not automatically bad if the change is risky. The report decomposes the time so the team can see where it waits and why. The decomposition is a diagnostic, not a target that every change must meet.

Define the start and end

Choose a start that the team controls. Ready for development is more useful than the date a customer asked, because it separates discovery from delivery. Choose an end that reflects real use, such as deployed and serving traffic. A merge is not the end if the change waits for a release train.

Record the definition and keep it stable. A team that moves the start earlier can appear faster without changing anything real. A team that moves the end to merge can hide the release queue. Write both boundaries in the metric definition and review them when the process changes.

The four segments

Split the time into queue, active work, review, and release. Queue is the wait before someone starts. Active work is the time spent building and testing. Review is the wait for a reviewer and the response. Release is the wait for approval, a window, or a deploy.

Report each segment separately and as a share of the total. A total of ten days may be two days of work and eight days of waiting. The share tells the team where to look. The absolute number tells them how much room there is to improve.

The segments should add up to the total, with a small allowance for clock gaps. If they do not, the timestamps are inconsistent or a step is missing. Reconcile the difference before reporting a trend. A decomposition that does not sum is a sign that the process has a step nobody has named.

Wait time versus work time

Wait time is often the largest segment and the least visible. A ticket can sit in a queue because the pod is at capacity, because a dependency is missing, or because nobody has decided the priority. Those causes need different responses. A segment that is small in one team may be the largest cost in another, so compare the shape rather than the total.

Distinguish blocked time from idle time. A change waiting for access is blocked, while a change waiting for a free engineer is queued. Record the reason in the ticket when the wait begins. Without the reason, the lead time chart shows a delay but not a decision.

Handoff and review effects

Every handoff adds a wait. A change that moves from the pod to a buyer reviewer and back can lose a day even when each side responds quickly. The number of handoffs is often a better predictor of lead time than the size of the change. The reason field is more useful than a longer list of statuses because it names the constraint.

Review capacity matters as much as review speed. A reviewer with a full queue will delay every change behind the current one. Track the queue depth and the time to first response, not only the time to approval. Those two signals separate a slow reviewer from an overloaded one. A reason field with no owner is still useful, because it names the constraint for the next review.

A wait reason should be easy to choose from a short list. If the list has twenty options, people will pick the first one or leave it blank. Keep five or six reasons and a note field for the detail. Review the distribution monthly and rename a reason if it is used for everything. The categories should help the team act, not decorate the chart.

Measurement pitfalls

Averaging lead time hides the shape of the distribution. A few very long changes can raise the mean while most changes are fast. Report the median and the share that exceed a threshold. The tail is where the coordination cost usually lives.

Exclude or separate changes that were deliberately parked. A change waiting for a customer decision is not the same as a change waiting for a free engineer. Tag the reason so the metric reflects the process rather than the calendar. A reviewer who is fast but unavailable for half the day may still be the bottleneck.

Handoff cost can be reduced without removing the review. A clear request with the diff, the risk, and the question lets a reviewer start immediately. A queue with a visible owner and a response target reduces the search time. Those changes do not weaken the control, and they often improve the review because the context arrives with the change.

A practical protocol

Instrument the four timestamps: ready, started, review requested, and deployed. Add a reason when a change waits more than one day. Collect for a month before comparing teams or periods. Segment by change size and work class.

Pick the largest wait segment and change one thing. Add a reviewer, clarify a decision right, or reduce the release batch. Measure the same segments for a month. The goal is a shorter path with the same control, not a faster path that skips review.

A dashboard can make a segment look better by changing when a status is set. For example, marking a change as started when it enters a queue removes the queue from the metric. Review the definitions whenever the workflow changes. The number should describe the work, not the label that happens to be available.

A worked example

A pod reports a median lead time of nine days. The decomposition shows one day of active work, one day of review, and seven days of queue. The queue is not caused by capacity but by a weekly prioritization meeting that assigns work only on Mondays. Use a threshold that fits the work, such as five days for routine changes and a separate view for larger ones.

The team moves small changes to a daily pull queue and keeps the weekly meeting for larger items. The median falls to four days without adding staff. The review segment stays the same, which shows the change did not weaken the control. Keep the reason list short so people actually use it.

Data to collect first

Record the ready date, the start date, the review request date, the approval date, and the deploy date. Add the change size, the work class, and the reason for any wait longer than a day. Use one time source and one time zone. The queue was visible only after the team separated waiting from working.

Collect at least a month before reporting a trend. Check a sample against the ticket history to confirm the timestamps are real. If a timestamp is missing, mark it rather than estimating. A metric built on guesses will not support a decision.

Counterevidence and limits

Lead time can reward small, low-risk changes and punish necessary large work. Segment by size so a complex migration is not compared with a copy change. Do not use one target for every class of work. A single timestamp source prevents a disagreement about which clock is correct.

A fast lead time can also mean the team is under-reviewing or releasing without evidence. Pair the metric with change failure rate and review depth. The useful question is whether the path is proportionate to the risk, not whether it is the fastest. Mark estimated values clearly and keep them out of the main trend.

Evidence table

Signals, inspection points, and accountable owners
SignalWhat to inspectOwner
OutcomeAcceptance evidence for the bounded taskTask reviewer
ControlAccess, test, and approval boundaryInternal owner
HandoffOpen risks and next decisionNext owner
Good distributed work is observable at the handoff: the result, evidence, limitations, and next owner are explicit.

Frequently asked questions

Should lead time start when the customer asks?

Usually not. Use a start the delivery team controls, such as ready for development, and measure discovery separately.

Is a lower lead time always better?

No. Pair it with change failure rate and review depth. A fast path that skips evidence is not an improvement.

How should we handle a deliberately parked change?

Tag the reason and report it separately. Mixing a customer decision delay with a capacity delay distorts the process signal.

Sources

  1. DORA, Lead time for changes
  2. NIST SP 800-218, Secure Software Development Framework
  3. GAO Cost Estimating and Assessment Guide

Related research