Skip to content
Anil Thapa
Leading data teams

The review queue has one person’s name on it

The person everyone sends their work to becomes the team’s bottleneck, and for a long time on my team that person was me. How to share the judgment, make room for the work, and recognize the contribution without keeping the dependency.

6 min read

For a long time, the name on my team’s review queue was mine. I had the most domain knowledge and the technical skills to go with it, so model changes, analyses and recommendation decks came through me before they reached a stakeholder. It made me the bottleneck.

What fixed it was not reviewing faster. I added automated checks for everything a machine could catch, and trained the team to review work beyond their own specialty, with coaching and regular knowledge-sharing sessions. In time, a data engineer who built pipelines could review an analyst’s model change, or a change to business logic, and validate the output. Things got simpler, and nobody was a single point of failure anymore. It mattered more once analysts owned their models, as I wrote in Giving the models away: more people could write a change, and somebody still had to make it safe.

It happens to senior analysts too, often without the authority over the plan that a team lead has. Picture the one everyone brings their work to. Model changes, cohort comparisons and recommendation decks all pass through them before anything reaches a stakeholder. They know the awkward history behind the definitions, they spot a change in grain, and they ask the question that makes a recommendation hold together. Their own investigation keeps slipping, and at planning both the investigation and the reviews are still expected to fit, because only the investigation has an estimate beside it.

That is not a personal time-management problem. The team has handed someone responsibility for other people’s delivery without reducing what it expects them to deliver alone. Asking them to review faster leaves both decisions unmade: who carries the reviews, and what comes off their plan to make room.

Find out what the signature is supposed to mean

A request for “a quick check” can hide several different jobs:

Work arriving for review Question that needs an answer Responsibility to make explicit
A model change Does the implementation preserve the intended grain, definitions and dependencies? A qualified reviewer for the affected model and its consequences
An analysis Does the evidence support the claim, with the relevant comparisons and uncertainty shown? A reviewer able to assess the method and the business context
A proposed action Is this a decision the organization should make? The person authorized to make that decision, informed by the analysis

One person can contribute to all three, but approving one does not cover the others. A sound cohort comparison can support several reasonable actions, and the analyst who reviewed it may have no authority to choose among them.

Look at a recent sample of reviews and note what the senior analyst actually supplied. Some caught defects. Others rebuilt missing context, rewrote the argument, or settled a definition nobody had agreed on. Those need different fixes: training or access to the stakeholder when work arrives unfinished, and a change to the approval rules when several people could review it but only one person’s approval is accepted.

Put review on the plan before assigning the next investigation

Calling review part of everyone’s job does not make anyone available to do it. Estimate the recurring load from real work, a short sample is enough, and then make a choice at planning: reduce the analyst’s delivery commitments, move some reviews elsewhere, or reduce the work entering the team. Urgent reviews need a route too, including which planned work moves when someone interrupts.

The review work already has a cost. Leaving it off the plan only hides which commitment will pay for it.

An analyst who does not control the plan can still bring the evidence: a small record of what arrived, what judgment it needed, and which commitment it displaced. Then the manager has to decide what the role contains. Leaving the analyst to negotiate every interruption alone turns a staffing decision into a test of their goodwill.

Give the next reviewer a question they can answer

A rotation helps only when the people on it can do the review and have the time. Start with a bounded area: routine changes to a model another analyst already maintains, or a familiar kind of analysis. Route by consequence, not by the number of lines changed, because a small change to a shared denominator can matter more than a large one to a local chart. The author brings the question, what changed, what they checked, and when the result is needed, so the reviewer does not have to discover the assignment from a nearly finished deck.

Make the end of the learning period explicit. Once the new reviewer can explain the material risks and recognize a limit, routine work in that area stops needing the original person’s signature. Keeping that signature forever keeps the queue and adds a step in front of it.

Leave a review another person can learn from

An unexplained approval gives the next reviewer nothing to reuse, and a rewritten query fixes today’s result while leaving the author unable to say what was wrong. For a material issue, record the claim being challenged, the evidence, and what would resolve it. If a join duplicates accounts, show a case and ask the author to verify the grain. If the analysis treats association as cause, name the alternative explanation the recommendation has to survive.

Separate a reason to block release from an optional improvement, or the most exacting reviewer quietly sets the capacity of the whole team. Google’s published code-review guidance draws the same line, between changes needed for code health and polish or teaching comments marked as optional. Analytical review needs that line plus one question code review does not ask: whether the evidence supports the decision.

Repeated comments are a signal. A stable rule about an identifier belongs in an automated check. An unsettled definition needs a named owner. A recurring mistake in cohort construction needs a worked example and an hour of teaching.

Stop routing every disagreement back to the original reviewer

If every disagreement is settled by asking the original reviewer what they would have done, the new authority lasts only until it is tested. When two approaches are acceptable, let the author and the assigned reviewer choose. When the disagreement is about business meaning or the risk of acting, route it to the owner of that decision. Seniority alone does not settle it. And practice the coverage during ordinary weeks: someone’s leave should not be the first time the team finds out which approvals nobody else may give.

Recognize the judgment that no longer needs a signature

Counting comments or approvals rewards a growing dependency. What I would look for instead are inspectable contributions: a recommendation that became better supported because a reviewer exposed an assumption, a recurring failure that became a check, another analyst who can now review a class of work and explain the hard cases. Read those next to waiting time, rework, defects found after release, and the state of the reviewer’s own commitments. Fewer reviews by the senior analyst is progress if qualified colleagues are carrying them. It can also mean people stopped asking.

Agree how review, teaching and standards count in the role before the performance conversation, not during it. And leave the analyst work that develops their own judgment, because a permanent assignment to repair everyone else’s nearly finished analysis uses up the opportunities they need for their next step.

This takes capacity before it returns any. Early reviews slow down, another reviewer will make choices the specialist would not, and the original reviewer gives up being the person every important piece passes through. Those costs need naming and support, not a lecture about delegation.

A senior analyst should be able to grow by making judgment available to others, without remaining the signature every result needs.