Key stats
- 1 finish line per estimate
- 4 weekly signals: accepted, review wait, decision wait, rework
- 0 activity counts used as a performance score
- 1 named constraint to act on each week
Key takeaways
- Count accepted work at the finish line, not commits or messages.
- Report review and decision waits separately from coding time.
- Treat reviewer capacity as the constraint it usually is.
- Keep the unit and exclusion rule stable across the window.
- Use the estimate for planning, never as an individual performance score.
Question and scope
This analysis asks how a small offshore delivery pod can estimate throughput without using activity counts as a proxy for useful work. The unit is one finished, accepted change tied to a ticket and a review record. The window is a fixed number of working weeks with the same team and a comparable ticket mix. Estimates about future delivery are bounded by that window and should be revised when the mix changes.
The purpose is planning, not ranking. A throughput estimate helps a buyer decide how much work to start, when to add review capacity, and which commitments are unrealistic. It does not measure effort, loyalty, or the value of an individual. The record should make that distinction explicit so a manager does not reuse it as a performance score.
Why activity counts mislead
Commits, messages, and online hours rise when work is fragmented and fall when people take time to think. A developer debugging a difficult production issue may produce little visible activity for a day and still prevent a larger failure. A developer splitting one change into ten commits can look busier without finishing more work. The metric reflects tool behavior more than delivery.
Activity counts also reward the wrong response to a bottleneck. If review is slow, a team can start more work and generate more commits while finished output stays flat. The resulting dashboard looks productive while the queue of unfinished changes grows. Throughput measures the end of the process, where the buyer actually receives a result and can decide what to do next.
Choose a unit of finished work
Use a unit the team can count consistently, such as a ticket that passed its acceptance checks and was accepted by the named owner. A merged pull request is not enough if the change still waits for a release check or a customer decision. A released change is not enough if it was rolled back for a reason the review missed. Define the finish line once and keep it stable.
Some work does not fit the unit neatly, including incidents, dependency upgrades, and exploratory tasks. Record those in a separate category rather than forcing them into the ticket count. The aim is a comparable measure for the planned work, not a total that hides everything else. State what was excluded and why so the estimate is not read as a complete account.
Measure waiting, not just working
For each finished item, record the time it spent waiting for a reviewer, a decision, or an environment. The sum of those waits often explains more about delivery than the time spent coding. A pod can look slow because one approval arrives two days late every week. Waiting time gives the buyer a concrete lever that does not require pushing developers harder.
Keep waiting categories separate and report the largest one first. If review is the main wait, adding development capacity will make the queue worse. If a buyer decision is the main wait, the fix belongs with the buyer owner. The estimate should point to the constraint rather than imply that every hour is interchangeable or that every delay has the same cause.
Review capacity as a constraint
A small pod can only finish what its reviewers can check with care. Estimate review capacity from the number of changes a reviewer can handle in a normal week, then compare it with the arrival rate of new work. A team that starts work faster than it can review creates a hidden inventory of unfinished changes. The inventory consumes attention even when nobody is coding.
Protect review time in the calendar and treat it as delivery work rather than a favor. Track how often reviews are delayed because the reviewer is in meetings or on support duty. If the same person reviews every shared-contract change, that person is a single point of failure. The throughput estimate should show the constraint plainly enough for the buyer to act on it.
Evidence from a bounded window
Collect the eligible ticket set, the accepted count, the review wait, and the rework count for the window. Use the same source systems and the same definitions throughout. Report raw counts beside any rate so a small sample cannot look more certain than it is. A window with three finished items supports a planning conversation, not a forecast with decimal precision.
Include one adverse case, such as a ticket that was blocked for most of the window or a change that returned twice for review. Explain how the case affected the count rather than removing it as an outlier. Outliers often reveal the constraint the team most needs to fix. Keep the exclusion rule fixed before the window closes.
What the numbers cannot say
A throughput estimate cannot prove that a developer is productive, honest, or a good fit for the team. It cannot compare two people fairly when they worked on different systems with different review support. It also cannot predict a week with a major incident or a new product direction. Those changes require a new estimate, not a reinterpretation of the old one.
The estimate is also vulnerable to changes in ticket size. A team can raise its finished count by splitting work into smaller pieces, which may be useful or may hide a larger change that never lands. Pair the count with the acceptance evidence and the scope of the released work. A number without its unit and its boundary is easy to misread, especially when it travels into a planning meeting without the table behind it.
Building a simple evidence table
A table with one row per accepted item is enough for most small teams. Include the ticket, the finish date, the review wait, the decision wait, and whether the change returned for rework. Add a short note for any item that does not fit the normal path, including the reason it was excluded. The table should be small enough to inspect and specific enough to answer a question about one week without opening five other systems.
Keep the evidence table separate from the performance review system. A planning record should not follow a developer into an annual assessment without context. Restrict access to the people who need it and remove temporary copies when the window closes. Clear boundaries make it easier for the team to record problems honestly, including the weeks when the plan was wrong.
Reviewing the estimate with the buyer
Bring the estimate to the buyer with the constraint and one decision to make. For example, show that review wait is rising and ask whether to pause new work or add a reviewer for a specific area. Do not present a finished number without the assumptions behind it. The buyer needs to see what would change the result before committing to a plan.
Record the decision and the next review date in the same note as the estimate. If the buyer chooses to keep starting work, record the expected queue growth and the risk to the commitment. If the buyer chooses to pause, name the work that will wait. A planning conversation is useful when it ends with a tradeoff, not a request for more effort.
When the estimate becomes a commitment
A commitment needs more than a throughput estimate because it also carries a date, a scope, and a consequence for the buyer. Use the estimate to describe a range and the conditions that would move the work inside it. State which constraint must hold for the date to remain possible. If the constraint changes, revisit the date instead of treating the original estimate as a promise.
A practical reporting routine
Publish a short weekly view with four lines: accepted items, items waiting for review, items waiting for a buyer decision, and items returned for rework. Add one sentence about the constraint the team observed. Keep the definitions and the source links beside the numbers. The view should take minutes to read and should lead to one decision or one follow-up.
Review the estimate at the end of the window and write down what changed. If the constraint moved from review to access, adjust the plan and the owner. If the numbers do not match the lived experience, inspect the unit and the finish line before doubting the team. A planning measure is useful when it helps people make a better commitment, not when it settles an argument about effort.
Evidence table
| Signal | What to inspect | Owner |
|---|---|---|
| Outcome | Acceptance evidence for the bounded task | Task reviewer |
| Control | Access, test, and approval boundary | Internal owner |
| Handoff | Open risks and next decision | Next owner |
Good distributed work is observable at the handoff: the result, evidence, limitations, and next owner are explicit.
Frequently asked questions
Can throughput estimates compare two offshore developers?
Not fairly when they work on different systems with different review support. Use the estimate for capacity planning and keep individual evaluation tied to the work and its context.
What counts as accepted work?
A ticket that passed its agreed acceptance checks and was accepted by the named owner. Merge alone may not be enough if release checks or a customer decision still remain.
How often should the estimate be updated?
Review it at the end of each fixed window and whenever the team, ticket mix, or review capacity changes. Do not adjust it weekly to make a commitment look better.
Sources
Prepare a throughput review
Define the finish line, collect accepted work and waiting causes for a fixed window, and name the constraint your team will act on. Keep release approval with your own owner.
