How do you measure the health of a team as an engineering manager?

In this article

1. Before you start
2. Why interviewers ask this
3. Reading team health from first principles
4. Worked example
5. Trade-offs and edge cases
6. How to answer it in an interview
7. Common mistakes
8. Follow-up questions
9. Sources

Before you start

Why interviewers ask this

A team can look fine in a dashboard while people are quietly leaving, so interviewers use this question to see whether you read more than one signal. They also want to hear you separate the team from the individual: a team signal moves slowly and needs a pattern, while a person signal shows up in a conversation. A weak answer names one metric, such as velocity or ticket count. A strong answer names three or four signals, says how each can mislead and says what you would do when one moves. Prepare the answer alongside the people management round, where it is one family of questions.

Reading team health from first principles

Team health

Team health is the combined condition of how reliably a team delivers, how long its people stay and how safe members feel to speak up, read through several measures instead of one.

Psychological safety

Psychological safety is a shared belief, in Amy Edmondson's words "held by members of a team that the team is safe for interpersonal risk taking" (Google re:Work, originally published 2015), that lets people admit a mistake, ask a question or offer an idea without being punished.

Burnout

Burnout is a syndrome that the World Health Organization describes as the result of chronic workplace stress that has not been successfully managed, marked by exhaustion, mental distance from the job and reduced professional efficacy.

In plain words

Reading team health is like reading a car dashboard. Speed alone does not tell you the engine is fine. A warning light alone does not tell you how fast you are going. You read several dials, and you know what to do when one turns red. The analogy breaks in one place. A car has no feelings, so some dials here are what people tell you, and a manager has to earn that honesty. Said precisely, choose a few measures across delivery, people and climate. Read them as prompts for conversation. Say what you do when each one moves. That last sentence is what interviewers remember.

Step 1: Separate the team signal from the person signal

A team signal is a pattern across several people, and a person signal is one individual's state; they need different tools. The input is everything you can see. The output is two short lists. Team signals include delivery numbers, how many people leave and how comfortable people are disagreeing in meetings. Person signals show up in 1:1s: energy, interest in the work, withdrawal. Say that you never diagnose a team from one person's mood, and never excuse one person's decline as "the team", because each mistake sends your attention to the wrong place. The one-on-ones page covers the person side. What you now know: which tool fits which signal.

Step 2: Pick a small, balanced set of measures

A balanced set mixes delivery, people and climate so that one number cannot hide a problem. The DevOps Research and Assessment (DORA) group defines five delivery metrics: change lead time, deployment frequency, failed deployment recovery time, change fail rate (the share of deployments that need immediate intervention) and deployment rework rate (DORA, updated January 2026). DORA also lists "one metric to rule them all" as an anti-pattern and warns that broad targets invite teams to game the numbers. The Satisfaction, Performance, Activity, Communication, Efficiency (SPACE) paper says developer productivity cannot be measured by a single metric or dimension (Microsoft Research, February 2021). Add attrition for people. What you now know: the set, and why one number never decides.

Step 3: Ask the team

Asking the team means putting a few plain statements to everyone and reading the answers. Google's study of 180 teams found five dynamics of effective teams: psychological safety, dependability, structure and clarity, meaning and impact, and the page gives a statement for each, such as "If I make a mistake on our team, it is not held against me" (Google re:Work, originally published 2015). The input is those five statements; the output is the share of people who agree with each. Use an anonymous survey and your 1:1s. In the worked example below, structure and clarity was the weak spot. What you now know: where the team itself says the problem is.

Step 4: Watch for strain and burnout

Watching for strain means looking for the early signs of exhaustion before they become attrition. The World Health Organization (WHO) describes burn-out as an occupational phenomenon with three dimensions: feelings of exhaustion, mental distance or cynicism about the job, and reduced professional efficacy (WHO, May 2019). Translate those into what a manager sees: tired, flat or sarcastic tone, slipping quality, people no longer volunteering. Check workload, on-call load and unclear priorities first, because the cause is often the system. Burnout is not a medical diagnosis in that classification. What you now know: three signs to watch and where to look first.

Step 5: Say what you do when a number moves

A measure only helps if you can name the action it triggers and the way it can mislead. The input is a change in a signal. The output is a conversation or a change in the work. If change fail rate rises, ask what changed in testing or review. If one survey dimension drops, discuss it with the team and agree one change. If attrition rises, hold career conversations with each person. Say how each measure can be gamed, since DORA warns of gaming when targets are broad (DORA, January 2026). What you now know: a number prompts a conversation, not a verdict.

1 · Separate2 · Measures3 · Ask4 · Strain5 · ActReplay

Step-by-step description

1. Split team signals from person signals.
2. Pick a small, balanced set.
3. Put plain statements to the team.
4. Watch for exhaustion and distance.
5. Name the action each number triggers.

• A team signal is a pattern across several people, and a person signal is one individual's state.
• A measure only helps if you can name the action it triggers and the way it can mislead.

Worked example

An invented team of eight engineers shows the arithmetic. Every input is illustrative, made up for this page and not a benchmark; the figures come from a short script.¹ In the year, one person left, which is an attrition of 12.5%. Of 50 deployments, 4 needed immediate intervention, so the change fail rate is 8.0%. Both look healthy. Then eight people rate the five re:Work statements, and the share who agree (a score of four or five on a five-point scale) is 62.5% for psychological safety, 87.5% for dependability, 25.0% for structure and clarity, 87.5% for meaning and 50.0% for impact.¹

The lesson is that the team looks fine on delivery while one dimension says otherwise, so the manager's next move is to ask what is unclear, such as ownership or how decisions are made, then agree one change.¹

Text description of the diagram

• The share who agree is 62.5% for psychological safety, 87.5% for dependability, 25.0% for structure and clarity, 87.5% for meaning and 50.0% for impact.
• Structure and clarity was the weak spot in the example.

Trade-offs and edge cases

Measures have costs and limits, and a strong answer says so. First, small teams give noisy numbers: with eight people, one answer moves a share by 12.5 points, so read trends and talk to people. Second, anonymity helps honesty, yet on a small team people can be identified, so group results and say how you protect responses. Third, any number can be gamed once it becomes a target; DORA lists broad targets as a risk. Fourth, compare a team with its own past, not with other teams, as DORA advises. Fifth, a poor score is not a verdict on a person; the low performer answer covers when the cause is individual.

Text description of the diagram

• If change fail rate rises, ask what changed in testing or review.
• If one survey dimension drops, discuss it with the team and agree one change.

How to answer it in an interview

The model answer below names the separation, a balanced set and the action.

Spoken structure: separate team from person, name the balanced set, say what you do when a number moves.

Common mistakes

1. One metric. (wrong) "We track velocity." (right) "I read delivery, retention and a team survey together."
2. Treating a number as the verdict. (wrong) "Our failure rate is low, so the team is healthy." (right) "It is one input; I also ask the team."
3. Ignoring the person. (wrong) Only dashboards. (right) Name your 1:1s and what you look for.
4. No action. (wrong) "I monitor it." (right) "If it drops, I ask the team and agree one change."
5. Comparing teams. (wrong) Ranking teams by one metric. (right) Compare a team with its own history.

Follow-up questions

How do you measure the health of a team versus a person?

For a team, read a balanced set of delivery, retention and survey signals as patterns. For a person, use 1:1s and watch energy, interest and workload. The two answers differ because one is statistical and the other is a conversation.

What signs indicate team morale needs boosting?

Look for fewer questions in meetings, work that nobody volunteers for, more sarcasm, slipping quality and rising absence. Google's team research ties psychological safety to people feeling safe to ask questions and admit mistakes (Google re:Work, originally published 2015).

How do you keep engineers motivated during challenging times?

Restore meaning and clarity: say what matters, why and what is out of scope, remove avoidable work and recognise real progress. Meaning is personal and varies, so ask each person what keeps them engaged. See growing engineers.

How do you spot burnout?

Watch for the three signs the World Health Organization lists: exhaustion, cynicism or distance, and reduced effectiveness. Check workload and priorities first, then talk in private and adjust the work. Involve your human resources team if the person needs support.

What if two engineers are the source of the tension?

Treat it as a specific conflict and act early; the page on two engineers who can barely stand each other gives the steps.

Sources

1. Understand team effectiveness (Project Aristotle), Google re:Work — Google re:Work (originally published 2015; month and day not stated on the page), 1 Jan 2015
2. DORA software delivery performance metrics — DORA (Google Cloud), 5 Jan 2026
3. Burn-out an "occupational phenomenon": International Classification of Diseases — World Health Organization, 28 May 2019
4. The SPACE of Developer Productivity: There's more to it than you think — Microsoft Research (Forsgren et al., ACM Queue 19(1), February 2021), 1 Feb 2021