Why do the same problems keep coming back? Single- and double-loop learning

In organisations, we often address a problem, see a temporary improvement and find ourselves back in the same situation a few weeks or months later. A check is added, a procedure is changed, another meeting is called. The error is corrected, but the mechanism producing it remains untouched.

This is the difference between changing an answer and changing how we arrive at answers; between asking only what we should do better and also asking which beliefs, implicit rules and objectives are shaping what we do.

Chris Argyris and Donald Schön described these two levels through the concepts of single-loop learning and double-loop learning. The first corrects action while leaving underlying criteria unchanged; the second questions those criteria too, making it possible to redesign action from a different understanding.

They are not two alternatives from which to choose once and for all. Single-loop learning is needed to regulate, improve and keep everyday work reliable. Double-loop learning becomes crucial when the same problems return, the context has changed or familiar solutions no longer deliver the expected results.

A theory of organisational learning that remains relevant

Double-loop learning emerged in the 1970s within the theory of action developed by Chris Argyris and Donald Schön. For these scholars, learning is not simply acquiring a new idea or understanding a problem intellectually. Learning takes place when a gap between intentions and results is identified and corrected through action.

This distinction matters. We may take part in a retrospective, recognise a mistake and develop a very clear interpretation, but if we continue to act in the same way at work, learning has not yet changed our actions.

A 2023 systematic review by Mercedes-Victoria Auqui-Caceres and Andrea Furlan, based on 128 studies published between 1974 and 2021, shows that the concept is frequently cited but often used in a limited way. Double-loop learning requires both cognitive change (in assumptions, beliefs or guiding values) and observable changes in behaviour. If only the way we describe the problem changes, or a behaviour changes without examining the premises behind it, genuine double-loop learning has not yet taken place.

Two different ways of correcting an error

In single-loop learning, when a result does not match what we expected, we look for a more effective course of action without changing the underlying objectives and rules. This is how a thermostat works: it detects a difference from the set temperature and switches the heating on or off.

In double-loop learning, a further question comes into play: is the set temperature still right? The system does not simply correct the difference; it can reconsider the criterion used to assess it.

	Single-loop learning	Double-loop learning
Central question	How do we correct the gap?	What are we taking for granted?
What changes	Actions, techniques or procedures	Assumptions, guiding values and objectives, followed by actions
When it is useful	When objectives and rules remain valid	When problems recur or the context has changed
Example	Adding a check to a process	Asking why the process needs that check and what conditions are contributing to the error

Consider a project that is continually delayed. A single-loop response might add checks, reset deadlines or introduce a new planning tool. These actions may help, but they might not be enough. Double-loop learning opens up different questions: Are priorities genuinely shared? Do people have the authority they need to make decisions? Are the objectives of the functions involved aligned, or are they working against one another? What information is not being shared, and why?

The point is not always to search for a deeper cause. It is to understand what level the problem sits at. If the direction remains right, improving the action may be enough. If the assumptions keep producing the same result, changing procedures alone will only bring temporary corrections.

What we say we do and what we actually do

To understand the assumptions guiding action, Argyris and Schön distinguish between espoused theory and theory-in-use. The first consists of the values and principles we say we follow. The second can be inferred by observing how we act in real situations.

An organisation may say it values collaboration while rewarding only individual results. A manager may say they want to delegate while continuing to ask for approval at every step. A team may say that discussion matters, then become defensive whenever someone raises uncomfortable information.

This gap is not necessarily hypocrisy. Often it is unconscious. People sincerely believe they behave according to their stated values, while under pressure they fall back on different rules. That is why theory-in-use cannot be uncovered simply by asking people what they think they do: it has to be reconstructed from observable events, conversations, decisions and consequences.

Double-loop learning begins when the gap between intentions and actions becomes something a group can explore without immediately turning it into an accusation. In that space, the group can recognise implicit assumptions, examine them and design alternatives.

The real obstacle: how we reason under pressure

According to Argyris, when situations feel threatening or embarrassing, people tend to act according to what he calls Model I: they try to maintain control, maximise the chance of being right, minimise the expression of negative feelings and appear rational. They defend their position, make judgements and attribute causes, while leaving the assumptions behind their reasoning implicit and difficult to test.

The result is paradoxical. In trying to protect ourselves, others or the relationship, we create misunderstanding, rigidity and self-reinforcing patterns. People become highly skilled at producing behaviours that undermine the very outcomes they want: a dynamic Argyris called skilled incompetence.

Model II, associated with double-loop learning, does not require people to abandon their position or share every decision. It calls for valid information, a combination of advocacy and inquiry, explicit reasoning that can be tested, and responsibility for choices. The central shift is from unilateral control to making our reasoning open to challenge.

Defensive routines: protecting ourselves at the cost of learning

When Model I behaviours recur, they can become defensive routines: actions, practices or policies that avoid embarrassment and threat while also preventing people from understanding and addressing their causes. They are automatic, learned and often socially accepted. For that reason, they can outlast the people involved: someone leaves the organisation, someone else joins, and the same dynamic continues.

A two-year ethnographic action research study, published in 2026 by Mercedes-Victoria Auqui-Caceres and Andrea Furlan, observed three recurring patterns:

• Blaming external factors. The problem is attributed to the system, another function or another person, leaving less room for the speaker to take responsibility or act.
• Avoiding genuine dialogue. Information, thoughts or feelings are withheld, made vague or manipulated to avoid making things difficult for ourselves or others.
• Avoiding discussion and public testing. Data, experience and opinions are presented as established truths, without creating the conditions for others to explore, test or question them.

These routines do not arise from bad intentions. In the short term, they serve a purpose: protecting people’s image, reducing tension and keeping relationships apparently calm. Over the longer term, however, they shut down conversations, restrict the flow of information and leave problems unresolved. Agreement remains superficial, decisions are based on incomplete information and the status quo is reinforced.

Why awareness alone is not enough

We might think it is enough to make people aware of their defensive routines. The 2026 study points to a more complex dynamic. When people recognise defensive behaviour, they may feel frustrated and justify it: ‘You can’t say everything’, ‘It’s better not to create tension’, ‘That’s how things work here’. The authors call these legitimising responses: apparently reasonable explanations that conceal and reinforce the routine.

When the negative consequences of these behaviours are made explicit too, an even stronger reaction may emerge: ‘We don’t have time for this’, ‘Change is impossible’, ‘Let’s talk about the real problems’. These are closing responses that make the routine itself impossible to question.

Defence, then, is not just a habit to recognise. It is a loop in which behaviour, emotion and reasoning reinforce one another. If discussion feels like a personal judgement, an attempt to interrupt a defensive routine can trigger a new defence.

Creating the conditions for double-loop learning

Double-loop learning is not triggered by one clever question or a general invitation to be more open. It requires a process that makes it possible to observe, reflect, test and redesign action.

Start with observable data. It helps to reconstruct specific events: what was said, what information was available, what happened and what effects followed. Spending too long on labels (‘there is no trust’, ‘they do not collaborate’, ‘they resist change’) encourages interpretations that cannot be tested and blame in both directions.

Make reasoning visible. Tools such as the Left-Hand Column from action science help distinguish what was said from the thoughts, judgements and feelings that remained private. The ladder of inference helps trace the path from selected data to conclusions and action. The aim is not to say everything all the time, but to recognise which unspoken information is influencing the conversation.

Combine advocacy and inquiry. In Model II, people state their own interpretation while also making it open to exploration: ‘These are the data I am using’, ‘This is the assumption I am making’, ‘What am I not taking into account?’, ‘What leads you to a different conclusion?’ The point is not to be less decisive, but to base decisions on reasoning that is more transparent and testable.

Create safe, structured spaces. Psychological safety, as studied by Amy Edmondson, does not mean avoiding disagreement. It means being able to ask questions, acknowledge mistakes, express doubts and share uncomfortable information without fearing disproportionate interpersonal consequences. The structure of discussion matters too: timing, questions, ways of participating and criteria should prevent dialogue from being dominated or turned into a confrontation.

Involve leadership and the wider context. New behaviours will not take root if objectives, incentives, roles, decision-making processes and everyday practices continue to reward the old ones. Double-loop learning becomes organisational when what has been learnt is reflected in meetings, systems, responsibilities and ways of making decisions, rather than remaining an individual insight.

What facilitation can do

Facilitation can create a space in which a group slows down enough to observe how it works while addressing a real situation. It helps distinguish data from interpretations, bring out perspectives that remain at the margins, make assumptions explicit and work through disagreement without losing its informational value.

A retrospective, debrief or workshop does not automatically produce double-loop learning. It may remain at single-loop level if it only gathers what worked and identifies a few corrections. The shift happens when the group can also ask which implicit rules made that way of acting seem reasonable, and whether those rules are still useful.

A facilitator does not tell the group which beliefs to abandon. They design a process that makes those beliefs visible, testable and connected to their consequences. An outside perspective can help because theories-in-use are often easier to recognise in others than in ourselves. But care is needed: exposing a contradiction too quickly can increase the sense of threat and strengthen the defence we are trying to move beyond.

Experiential learning can also support this shift when it uses real cases and conversations, makes behaviour observable, introduces useful models, allows people to redesign their actions and gives them repeated opportunities to try them out. Understanding the model is not enough: people need to practise a different way of reasoning and acting in situations that matter.

A programme, not a one-off session

Research highlights how difficult it is to change deeply ingrained theories-in-use. The 2023 review shows that effective interventions require collective reflection, genuine dialogue, leadership involvement, observation of behaviour and repeated practice over time. In several of the studies reviewed, programmes involved multiple sessions over a period of months.

This is consistent with what we know about developing behavioural skills. When we come under pressure, our habitual repertoire tends to return. To establish an alternative, people need experience, feedback, application at work and the chance to observe the effects of new actions.

That is why a one-off workshop is not enough to transform how an organisation learns. A session can open a door, make a dynamic visible and support an initial choice. Change becomes lasting when the organisation continues to experiment, reflect and adapt, including by changing the conditions that supported the previous behaviours.

Learning to change the question

Single-loop learning asks: ‘What do we need to do differently to get the result?’ Double-loop learning adds: ‘What are we assuming to be true, necessary or beyond question that leads us to act this way?’

This is not an invitation to question every rule or decision all the time. That would be paralysing. It is about recognising when an operational adjustment is enough and when, instead, doing the same thing better simply reproduces the same problem more effectively.

In my work, I start with recurring situations: decisions that keep being reopened, projects that lose momentum in implementation, information that fails to flow, and familiar behaviours that are hard to change. I create the conditions for the group to observe what is happening, test its assumptions and turn understanding into new action.

Double-loop learning, ultimately, means not just finding a new answer, but becoming able to revisit how we arrive at answers. This capability enables people and organisations to interrupt habitual responses, adapt to a changing reality and develop clearer, more generative ways to act.

Sources

• Luigi Mengato, ‘Double-Loop Learning e Routine Difensive Organizzative: verso un apprendimento trasformativo’, 2025.
• Chris Argyris, ‘Double Loop Learning in Organizations’, Harvard Business Review, 1977: https://hbr.org/1977/09/double-loop-learning-in-organizations
• Chris Argyris and Donald A. Schön, ‘Organizational Learning: A Theory of Action Perspective’, Addison-Wesley, 1978.
• Chris Argyris, ‘Overcoming Organizational Defenses: Facilitating Organizational Learning’, Allyn & Bacon, 1990.
• Chris Argyris, ‘Action Science and Organizational Learning’, Journal of Managerial Psychology, 10(6), 1995, pp. 20-26: https://doi.org/10.1108/02683949510093849
• Mercedes-Victoria Auqui-Caceres and Andrea Furlan, ‘Revitalizing Double-Loop Learning in Organizational Contexts: A Systematic Review and Research Agenda’, European Management Review, 20(4), 2023, pp. 741-761: https://doi.org/10.1111/emre.12615
• Mercedes-Victoria Auqui-Caceres and Andrea Furlan, ‘Why Do Defensive Routines Persist in Organizational Contexts? Results from a Two-Year Ethnographic Action Research’, European Management Review, 2026: https://doi.org/10.1111/emre.70055
• Amy C. Edmondson, ‘Psychological Safety and Learning Behavior in Work Teams’, Administrative Science Quarterly, 44(2), 1999, pp. 350-383: https://doi.org/10.2307/2666999