What the research found
Amy Edmondson’s study of 51 manufacturing work teams, published in Administrative Science Quarterly, examined the relationship between psychological safety, learning behaviour and team performance directly. Her finding was specific: team psychological safety predicted learning behaviour even after controlling for team efficacy, and learning behaviour — the visible willingness to raise problems, test ideas and seek feedback — was what actually mediated the link to performance. In her own words, “team psychological safety is associated with learning behavior, but team efficacy is not, when controlling for team psychological safety.” This is a feedback-loop finding at the team level: the loop only functions if people believe it is safe to put information into it.
Nelson Repenning and John Sterman, at MIT Sloan School of Management, approached a related question from a different angle entirely — system dynamics modelling combined with more than a dozen case studies, published in California Management Review. Their focus was why process-improvement gains fail to sustain themselves. Their answer centred on what they termed the “capability trap”: organisations under pressure default to working harder rather than working smarter, and the people whose work prevents problems receive no visible credit for problems that never happened, because an avoided problem is invisible by definition. This creates a specific asymmetry in the feedback loop — visible effort is rewarded, invisible prevention is not — which erodes the very behaviour that would sustain a gain over time.
Michael Beer and Russell Eisenstat’s research for Sloan Management Review, now MIT Sloan Management Review, examined this from a leadership-level angle. Their study of strategy implementation identified a set of mutually reinforcing “silent killers,” including poor vertical communication and a top-down or laissez-faire senior management style, that block organisational learning largely invisibly to the leaders sitting above the block. Their conclusion was that this particular barrier has to be addressed directly before any of the others can be, because it is the barrier that determines whether signal from below the leadership team can travel upward at all — a precondition for any feedback loop that spans more than a single team.
V. F. Ridgway’s earlier and foundational research, published in Administrative Science Quarterly through Cornell’s Johnson Graduate School of Management, contributes a fourth angle: what happens when a feedback loop is built entirely around a single quantitative measure. His documented cases — interviewers optimising for volume over placement quality, investigators selecting easier cases near a quota deadline — describe a loop that technically functions, in the sense that it produces a consistent signal and a consistent response, while the signal itself has become disconnected from the reality it was meant to represent.
Where it converges and where it diverges
Across these four bodies of research, a consistent theme emerges: a feedback loop is not simply a mechanism for collecting information. It is a mechanism that only functions if three conditions hold simultaneously — the information collected has to still reflect reality (Ridgway’s territory), the people who hold that information have to feel safe surfacing it (Edmondson’s territory), and the organisation’s incentive structure has to reward the loop’s actual function rather than a proxy for it (Repenning and Sterman’s territory, and Beer and Eisenstat’s, from the leadership level). Where any one of these three conditions fails, the loop can continue operating in form while failing in substance — producing a steady stream of data or reports that look like a functioning feedback mechanism without actually catching the drift it exists to catch.
Where the research diverges is more instructive than where it agrees. Edmondson’s and Beer and Eisenstat’s work locate the primary failure point in psychological and relational conditions — whether people feel safe contributing honest signal upward. Ridgway’s and, differently, Repenning and Sterman’s work locate the failure point in structural and incentive conditions — whether the measurement or reward system itself is pointed at the right target. Neither camp claims the other’s factor is irrelevant; the divergence is closer to a difference in emphasis, driven partly by the different methods each study used — Edmondson’s field study of team interactions naturally foregrounds relational conditions, while Repenning and Sterman’s system dynamics modelling naturally foregrounds structural and incentive conditions. Read together rather than separately, the four studies suggest that a feedback loop’s durability depends on both sets of conditions holding at once, and that fixing only one — building psychological safety without also fixing the incentive structure, or fixing the incentive structure without addressing whether people feel safe contributing to it — leaves the loop vulnerable to the failure mode the untouched condition still permits.
A further point of convergence, cutting across all four studies, is timing. Repenning and Sterman’s capability trap describes a gain that fades once attention — a proxy for active, ongoing feedback — moves elsewhere. Ridgway’s distorted measures typically drift gradually rather than failing all at once. This suggests feedback loops are not simply present or absent; they degrade over time in specific, researchable ways, most commonly once the initial conditions that made them function — heightened attention, a leader’s personal involvement, a recent memory of why the measure mattered — fade into the background of ordinary operations.
It is worth noting what the four studies, taken together, do not claim. None of them argues that a feedback loop, once properly designed, can be left to run indefinitely without further attention — the opposite is closer to their shared implication, since each identifies a different way that a loop’s supporting conditions can erode even while the loop’s formal structure remains outwardly unchanged. This has a specific consequence for how “sustained performance” should be understood: not as a state achieved once a loop is correctly built, but as an ongoing property that depends on the loop’s supporting conditions being periodically checked, in much the same way the loop itself is meant to periodically check the performance it monitors.
What remains unresolved
How much of a feedback loop’s durability depends on a specific individual’s ongoing attention, and how much can genuinely be built into a system that persists once that individual moves on — the research points toward this distinction mattering greatly, without offering a settled answer for how to tell the two apart in advance.
Whether psychological safety and structural incentive alignment are separable in practice, or whether attempting to build one without the other systematically undermines it — the four bodies of research examined here approach this question from different angles without directly resolving it.
How organisations can detect the gradual, still-functioning-in-form failure of a feedback loop before it produces a visible consequence, given that Ridgway’s and related findings suggest this kind of drift is, by its nature, hard to see from inside the very system that has drifted.
The same pattern appears across many kinds of organisation; see Other Sectors for how it is read elsewhere.
Questions worth considering
- What is the organisation’s mechanism for noticing that something has drifted — and when did anyone last check that the mechanism itself still reflects reality?
- Do the people who hold the relevant information feel able to share unwelcome news?
- If the person who currently keeps a feedback loop working moved on, would the loop continue to function?
More on this topic: Performance.
Your team is facing similar challenges?
Subscribe to The Riddle Ledger
Perspectives are sent as they are published, with no promotional email, and you can unsubscribe at any time.