The manager your dashboard can't see

What does a bad manager look like on a dashboard? Often, good. His own metrics were strong. His team hit or beat its quarterly OKRs in three of four quarters. The two reports who left last year gave polite exit reasons: better opportunity, new challenge. On every standard review measure, he was a high performer.

The pattern showed up only when the CEO pulled cohort data across his 18-month tenure. Of the eleven reports who passed through the team, seven had left the company or transferred out. Of those seven, four told an exit interview or a later conversation that his style was a real factor. Of the four who stayed, two slid after he left, which suggests they had been carried by their own effort, not his management. The two who stayed strong had worked around him, not through him.

The cohort outcomes were the real data. None of it lived in a review system. It had to be rebuilt by hand, against a question the company had never asked.

This is the most expensive kind of bad management to catch. It posts good short-term metrics and bad long-term human outcomes. The manager's own numbers are strong. The team hits plan. Leadership has no reason to look closer. The cost shows up slowly, in the people who cycle through. They leave. They stall after leaving. The talent who should have grown never does.

The bill compounds at the top. A company's senior pipeline depends on managers who grow reports into the next layer of leaders. A manager who burns through reports without developing them starves that pipeline in silence. By year three or four, the company has a leadership gap no single event explains. It's the predictable result of keeping one corrosive manager a year too long.

The behavior is rarely malicious. The manager usually thinks they're managing well. The habits that do the damage are micromanagement, public criticism, taking credit for the team's work, withholding context. They're often invisible to the manager. They're obvious to the reports, who quietly adjust how much effort, initiative, and loyalty they're willing to spend.

The tool that surfaces these managers is cohort tracking. For anyone past six months of tenure, track a few things. How many reports have they had? What share left within a year? How did the leavers do in their next role? What share got promoted from under them? How many of the team's high ratings held up once the manager was gone? None of these are standard review questions. All of them sit in HR data you already keep. The same blind spot that lets two words distort a performance review hides these managers: the system measures the wrong thing well.

Run the numbers and you can finally tell two managers apart. One team's good metrics come from the team, despite the manager. The other's come from the manager's actual skill. On the quarterly dashboard they look identical. Over two years they're nothing alike.

The harder part is acting once the data is in. By the existing system, this manager has been performing. Removing them feels unfair, because the record shows no underperformance. The cohort data is the record the company never kept. Acting on it means overriding the system you've trusted and believing the numbers you rebuilt. Most companies don't. The data is uncomfortable and the review system is defensible, so the corrosive manager stays. The pipeline gap grows. The best reports keep leaving at a rate no one traces back to the manager who produced it. It's the same misread that lets a wrong hire look fine on paper until the pattern repeats.

Before your next leadership offsite, ask:

• For each manager past a year of tenure, what happened to the reports who worked under them?
• Which managers run above-average turnover among reports?
• Which managers produced reports promoted at higher rates than their peers?
• Is any of this tracked in a system leadership reviews?

The answers point at the managers you're failing to judge. The standard review grades a manager alone. The cohort data grades them in context. The two rankings differ, and the second is closer to true.