The scenarioIn 2008, Google's People Analytics team launched Project Oxygen to answer a question many engineers considered already settled: do managers matter? The data said yes — teams with effective managers had lower turnover, higher satisfaction, and better performance. But the follow-up question was harder: what specifically do effective managers do? Google's initial manager feedback surveys were full of vague assessments — "good communicator," "empowering," "has a clear vision." These told managers they needed to improve but gave them nothing actionable to change. A manager rated poorly on "communication" had no idea which communication, in which context, was the problem.
How the tool appliedGoogle's People Operations team restructured their upward feedback instruments around behavioural specificity that mirrors SBI's core logic. Instead of asking "Is your manager a good communicator?" they asked questions anchored to observable situations and behaviours: "When you share a new idea with your manager, does your manager ask follow-up questions to understand it before responding?" and "In your last 1:1, did your manager provide specific examples when giving you feedback on your work?" The feedback recipients — managers — received data tied to specific interaction patterns rather than abstract trait ratings. The impact component was captured through outcome correlations: teams whose managers scored high on these behavioural questions showed measurably higher psychological safety scores and innovation output.
What it surfacedThe behavioural specificity revealed that Google's best managers weren't distinguished by charisma, technical depth, or strategic vision — the traits most people associate with leadership. They were distinguished by a small set of observable, coachable behaviours: asking questions before giving answers, providing feedback with specific examples, and expressing interest in direct reports' career development through concrete actions (not just stated intentions). The worst-performing managers weren't "bad leaders" in some holistic sense — they simply didn't do these specific things. That distinction mattered enormously for intervention design. You can't coach someone to "be a better leader." You can coach someone to ask two follow-up questions before responding to a proposal.
