Single-loop and double-loop learning: fix the work or question the rule?

Single-loop learning changes how work is done while keeping its governing objectives or rules in place. Double-loop learning brings those objectives, rules or underlying values into the inquiry. The practical question is not whether your team is doing enough reflection. It is whether the thing being questioned is the action or the frame that makes the action seem sensible.

Both forms matter. You do not need to reopen the purpose of a service whenever someone corrects a scheduling mistake. But if the same problem returns despite better scheduling, it may be worth asking what the schedule is designed to achieve—and whether that definition still makes sense.

The distinction behind the terminology

Chris Argyris’s 1977 account of double-loop learning distinguishes error correction within existing objectives from inquiry that challenges underlying policies and objectives. The distinction became central to the organizational-learning work associated with Argyris and Donald Schön. It is a way of examining learning, not a two-stage maturity ladder that every organization climbs in order.

An error here need not mean somebody has behaved carelessly. It can be a mismatch between an intended result and what happens. Learning can involve noticing the mismatch, explaining it and changing the conditions that reproduce it. Whether that change is single- or double-loop depends on what is held fixed.

The same visible action can serve either kind of learning. Adding a review meeting might improve execution against an unchanged target. Alternatively, the meeting might give affected people authority to challenge the target itself. Calling a meeting a retrospective does not tell us which of those things occurs.

Follow one repair request through the problem

Consider a fictional repair service whose managers watch how quickly requests are marked closed. Coordinators chase outstanding tickets, technicians record visits and the weekly report looks healthy. Yet customers sometimes call again because a temporary repair did not settle the underlying fault. No real service or measured result is being described here.

A single-loop response could make the existing process more reliable: clarify who updates a ticket, fix a missed reminder, improve the handover between shifts or ensure that the closure field is completed promptly. Those changes might be worthwhile. They address performance within the current definition of completion.

The double-loop question is different: why is a closed ticket being treated as an adequate sign that the customer’s problem is resolved? The team might discover that “closed” means the technician has finished the visit, while customers interpret it as the repair being complete. A target that rewards quick closure could be directing attention away from unresolved work.

That is a hypothesis to investigate, not an accusation of gaming the numbers. Trace real cases before explaining motives. A repeat call might concern a different fault, an unavailable spare part or unclear information. The team needs evidence about what actually happened before deciding that the governing measure is wrong.

Compare the two inquiries without ranking people

Question	Single-loop inquiry	Double-loop inquiry
What is being corrected?	An action or operating method	A governing rule, objective or assumption about success
What stays fixed in the example?	The current meaning of a closed request	The meaning of closure itself is open to examination
What might change?	Reminders, scheduling or handoffs	What counts as resolved and who can confirm it
What evidence is needed?	Whether the revised handling works	Whether the revised definition better serves the intended purpose

The difference is not “small improvement” versus “large transformation.” A costly software replacement can preserve the same assumptions. A short conversation can expose a consequential misunderstanding about whose outcome matters. Size, expense and enthusiasm do not determine the learning loop.

Nicole Gardner’s research on architectural practice uses learning-loop theory to examine digital transformation through practitioners’ accounts. It illustrates why gaining new technical skills is not automatically the same as reconsidering organizational practices. It is a qualitative study in a particular professional setting, not a test showing that double-loop learning is universally superior.

Avoid turning the labels into judgments about colleagues. “You’re only doing single-loop thinking” closes a conversation that should be opening. A more useful question is: “Which rule are we treating as fixed, and is there a reason to examine it?” Sometimes the answer is that the rule is appropriate and execution needs attention.

Start the discussion with an observable mismatch

Choose a bounded episode or recurring pattern. Bring a request history, a handover record or a sequence of decisions, with identifying details removed where appropriate. Ask what was intended, what happened and what is still uncertain. Keep an observation separate from its interpretation.

For our fictional service, “the customer called again after the visit” is an observation. “The technician did not care” is an interpretation about a person. “The closure rule may not capture unresolved faults” is an explanation that can be checked. Keeping those forms separate makes it easier to disagree without pretending that every account is equally supported.

Then ask what seemed reasonable at each point. What information did the coordinator have? What did the technician believe the closure field meant? Which result was the manager expected to report? This does not remove accountability; it makes the operating logic available for examination.

If several explanations remain plausible, use assumption mapping to identify which uncertainty matters most. Do not rush from an uncomfortable story to a new policy. A policy can be just as speculative as the belief it replaces.

Ask about the rule—and who can change it

Useful prompts connect the work to its purpose: what does this target stand for, whose experience is absent, and what would persuade us that our current definition is inadequate? Ask what the team would do differently if a different outcome mattered. The answer may reveal that the reported objective and the rewarded behavior are not quite the same.

In the repair example, the team could distinguish a visit completed from a problem resolved. It could ask who checks whether a temporary repair needs follow-up and how that work remains visible. Those are proposals for examination, not automatic improvements. A customer-confirmation step, for instance, might introduce delays or exclude people who are difficult to reach.

Identify the authority needed for any change. A team can question a reporting measure without being entitled to alter it unilaterally. A useful output may be a documented recommendation to the service owner, including the cases examined, alternative explanations and a proposed bounded trial.

Keep genuine safety or other mandatory constraints in place. Questioning an assumption is not permission to bypass a control. Where a rule serves several purposes, understand those purposes before deciding that it is merely bureaucracy.

Make the conversation safe enough to be informative

People may notice a mismatch and still decide that raising it is too risky. Edmondson and Bransby’s review of psychological-safety research identifies learning behavior as a major theme in the literature. That supports attention to the interpersonal conditions for inquiry; it does not establish that one facilitation technique will make a particular team safe.

In practice, the person with the most authority can begin by naming an assumption of their own that may be incomplete. Ask for concrete counterexamples and respond to them as information. If a concern is dismissed, punished or immediately turned into a demand that the speaker prove a complete solution, an invitation to “be candid” has little substance.

Do not use openness as pressure to disclose personal information or criticize a colleague in public. Allow private routes and distinguish a learning conversation from processes intended to handle individual conduct. Some issues need a different setting, representation or specialist support. A workshop cannot promise protection it does not control.

Turn reflection into a testable change

Suppose the fictional service owner agrees to trial two separate statuses for a limited class of routine repairs: visit completed and follow-up resolved. Before the trial, define what each status means, who owns an unresolved item and how someone can challenge an incorrect record. Prepare people through a change communication plan rather than silently renaming the field.

Observe both the intended benefit and the extra burden. Does outstanding work remain visible? Are responsibilities clearer? Does the new distinction create duplicate updates or encourage a different shortcut? Follow a few cases end to end, rather than assuming a cleaner dashboard represents a better experience.

At the review, compare what was expected with what happened. The AHRQ debrief guidance emphasizes reconstructing events, examining what worked and incorporating lessons into the plan. Its healthcare setting differs from this repair-service example, but that connection between reflection and changed practice is useful. A debrief is not inherently double-loop: it can address either execution or the governing frame.

If the revised definition is helpful, carry it into training, reporting, ownership and knowledge-transfer arrangements. Otherwise the learning may stay with the people in the room while the organization continues to reward the old routine. If the trial reveals new problems, revise or stop it. Learning does not require defending the first replacement rule.

Evidence and boundaries

The learning-loop distinction is an interpretive framework, not a diagnostic score or a guarantee of performance. The original Argyris source is represented here through its publisher’s summary, not a claimed review of the full paid article. The architecture study offers contextual accounts; the psychological-safety review provides a broader research perspective; the debrief guidance contributes a practical review discipline. None tests this invented repair-service intervention.

Do not conclude that a team has achieved double-loop learning merely because its members questioned a target. Look for a consequential change in how the work is understood and governed, and then examine what that change does in practice. Equally, do not dismiss reliable single-loop improvement. Sometimes the right rule needs better execution. Sometimes repeated execution problems are telling you to reconsider the rule.