What makes a good manager, in data

A management-software company, 210 people, six product teams (a mid-size Brazilian company, figures in Brazilian reais). The internal survey flagged Team B as the least satisfied in the house. The immediate read was that the manager was the problem, so they replaced him. Seven months later, Team B was still the least satisfied — now with a different, equally exhausted manager.

What nobody had looked at: Team B was the only one serving three sales areas at once, with no priority rule. No leadership behavior fixes a queue with three owners. But that doesn't mean the manager doesn't matter — and there's reasonable evidence about what, within what he controls, does make a difference.

01The idea: the hypothesis the data killed #

Project Oxygen started at Google around 2008 with a hypothesis nearly opposite to its finding: that managers had little effect on engineering team performance. The analysis crossed team evaluations, performance indicators and internal surveys, and it did not hold up. Teams with better-rated managers showed better results in satisfaction, retention and performance.

The next step was turning that into a description of behavior. Based on interviews and text analysis of the evaluations, the company built a list of behaviors associated with the best-rated managers. The case was documented by David Garvin and colleagues at Harvard Business School in 2013, and Laszlo Bock, then head of people operations, told it in Work Rules! (2015).

The behaviors, in Google's own public formulation, include: being a good coach; empowering the team without micromanaging; expressing interest in team members' success and well-being; being productive and results-oriented; being a good communicator, listening and sharing information; supporting career development; having a clear vision and strategy for the team; having key technical skills to help advise the team.

02The honest caveat: what doesn't travel #

That list circulates as if it were universal, and it isn't. Three limits matter to anyone managing a 200-person company in Brazil.

The first is the sample's context: highly qualified engineers, with strong market leverage and a lot of technical autonomy. The behavior of 'empowering without micromanaging' assumes someone who already knows how to do the work and understands the goal. In an operation with high turnover and short learning curves, the same behavior can turn into abandonment.

The second is the apparatus. Google measured this with a dedicated people-analytics team, a semiannual survey, structured upward evaluation and a sample of thousands of teams. A 200-person company has thirty managers and no statistical power — copying the instrument without the base produces a fragile number treated as truth.

The third is the most important: the evidence is correlational. It shows good results and certain behaviors travel together, not that one causes the other. It's plausible that teams with a better context produce better-rated managers, not the other way around — as with Team B in the opening.

03The cost of replacing the manager without changing the system #

At the scale of the 210-person company, a 9-person team, stated assumptions:

Want to see how this looks inside a real operation? Explore the platform.

The math: replacing two technical people at roughly R$ 80,000 each, plus the new manager's ramp-up estimated at four months at 50% — R$ 50,000. What doesn't make the spreadsheet is the second manager leaving with the impression of having failed at a problem that was never about people management.

04Where this breaks down in practice #

1. The list turns into an evaluation form. Eight behaviors become eight scores from 1 to 5, and the conversation about the work disappears.
2. The manager doesn't have the data. You can't 'communicate priority' when it changes in another room and nobody logs the change.
3. Upward feedback without real anonymity. In a nine-person team, the answer is identifiable, and the instrument ends up measuring caution.
4. Nobody teaches how to manage. The best technical performer gets promoted and is expected to show new behavior with no preparation at all.
5. The structural problem is read as personal. A queue with three owners, a contradictory goal, and understaffing all show up as a leadership deficiency.

05What management with data answers #

What travels from Project Oxygen to a mid-size company isn't the list: it's the method. Describe observable behavior instead of an abstract quality, and check it against results instead of impressions. That requires a minimum of work being logged — without it, evaluating a manager becomes a popularity contest.

Much of the cited behavior depends on information being available at the right moment, which connects to the manager's fragmented day and to managing yourself. And the condition for a team to say what's stuck is the one discussed in teams that report more errors.

On the platform, that's the people, tasks and intelligence modules, which give the manager workload, priority and progress before the conversation. Solutions by area show the breakdown by operation, and a conversation helps separate what's a management problem from what's a design problem.

Before replacing a manager, it's worth answering: will the next one have any information this one didn't? If the answer is no, the change is expensive and predictable.