The feedback signal completes the flow of value, it governs the corrections that let fast-loops function as double-loops, and it’s the mutual sensing that lets two systems entrain instead of operating independently near each other, despite their interdependence.
Where feedback was doing the work
Flow of value is understood largely in one direction: knowledge moving from customer to developers, resources from Systems to value streams, or decisions through a network and outward through the organization. Feedback is the return path that closes the loop. The information about how that value was experienced, flowing back upstream. A team can hit every metric of delivery, and never learn how the thing delivered was received; not unless something structural is put in place to carry that information back.
Entrainment depends on feedback even more directly. Kuramoto described oscillators continuously sensing each other’s phase, and adjusting in response. Remove the continuous sensing, and there can be no entrainment. You get two systems that happened to align once in a while but mostly drift apart as conditions change. Feedback is the mechanism of entrainment.
Fast loop and slow loops depend on the correction step: information about the result feeding back into the next attempt. A fast loop without feedback is repetition without learning or growth; a car stuck in the snow, spinning its wheels.
Asking for feedback, manipulates the feedback
The easiest way to get feedback, is to ask. But surveys and interviews cost real time and resources to run well. They’re vulnerable to manipulation, consciously or not, and they depend on a level of honesty and trust that can’t be assumed; especially in the situations where feedback matters most (a struggling team, a strained relationship between vantages, a leader whose own decisions are the thing being evaluated).
In economics they call it hypothetical bias: the documented, repeated finding that what people say they would do (or say happened), diverging from what their behavior shows. Entire methodologies exist specifically to correct what survey answers say, using real behavioral data collected alongside them. Stated answers alone are known to be unreliable. Revealed preference (the term for what people’s choices show), doesn’t have this problem because it’s taking the feedback directly from behaviors, not asking anyone to interpret, or report on it themselves. It’s a clearer signal.
Structural feedback is signals that arrive as a natural byproduct of the work happening, not because someone was asked to produce them (an unintended manipulation, the noise that clouds the signal). A defect rate. A support ticket volume. Whether a delivered feature got used. None of these require accuracy, interpretation, or honesty. Solicited feedback still has real value (some things aren’t visible any other way, and sometimes the interpretation of the experience is precisely what we’re looking for), but it belongs at the edges of a feedback system, not its foundation.
Slow signals, even clear ones, are harder to hear
A fast loop’s feedback is easy to notice. The correction happens in the same short cycle as the action. A slow loop’s feedback is harder to notice, and more challenging to look for. Slow loops often address concerns at a higher altitude of scope. Whether or not an aging structure still deserves to exist (an old policy, a standing routine entrenched in the calendar), won’t flag itself on any dashboard. It shows up, if it shows up at all, as a slow accumulation of warning signs the rigidity trap research describes as rising connectedness. That accumulation is a form of feedback, but it’s feedback most organizations aren’t designed to watch, and are slow to notice. Elliot Jaques’ finding (that different work operates on different time-spans, each with its own real character) underscores that: a slow-loop question needs a feedback window matched to its own scope and pace, not the fast loop’s, or it will look like nothing is happening right up until the moment that something breaks.
Applying this
- Build feedback into the structure of the work itself before adding a survey to check on it. Ask what native signal exists, or could exist, as a byproduct of the work happening, before asking anyone to self-report.
- Treat any solicited feedback as a supplement to a structural signal, not a substitute for it. If a survey result and the structural data disagree, investigate the gap; don’t default to trusting either one automatically.
- Match a feedback mechanism’s time horizon to the loop it’s meant to serve. A slow-loop question needs a feedback window measured in months or years, not a weekly check-in borrowed from the fast loop next to it.
- Watch for the specific signature of missing slow-loop feedback: a structure nobody has questioned in a long time, not because it’s clearly still working, but because nothing was ever built to notice if it stopped delivering value, and continued devouring resources.
- When solicited feedback is necessary, minimize what it costs to give and what it costs to game. A short, specific, low-stakes question produces more honest signal than a long survey asking people to evaluate something they have reason to answer carefully about.
Pitfalls to watch for
Treating the absence of complaints as feedback. Silence is not a structural signal; it’s the absence of one, and mistaking it for a positive result is a common, costly misread.
Building elaborate structural feedback systems that measure what’s easy to measure rather than what matters. A defect count is real signal only if defects are what the work is meant to avoid; chasing measurability over relevance produces confident, useless data.
Assuming structural feedback is immune to gaming simply because it doesn’t require self-reporting. Any measured signal that people know is being watched can still be optimized toward, deliberately or not, once its existence is understood; structural feedback reduces the honesty problem, it doesn’t eliminate the incentive problem underneath it.
References and further reading
- Samuelson, Paul A. “A Note on the Pure Theory of Consumer’s Behaviour,” Economica, 1938, the original source of revealed preference theory.
- Murphy, James J., et al. “A Meta-Analysis of Hypothetical Bias in Stated Preference Valuation,” Environmental and Resource Economics, 2005.
- Kuramoto, Yoshiki. “Self-entrainment of a Population of Coupled Non-Linear Oscillators,” 1975.
- Jaques, Elliot. (1996/1997). Requisite Organization: A Total System for Effective Managerial Organization and Managerial Leadership for the 21st Century. Routledge/Cason Hall.