Double-Loop Learning
Double-loop learning is a way of responding to an error. It corrects the action and examines the values and the rules of reasoning that produced it. Single-loop learning corrects the action and keeps the rule. Chris Argyris coined both terms. From them he derives an exercise: the manager writes the planned meeting in two columns, then analyses it with the people involved instead of holding it. The business analyst uses the technique when a deployed change runs into a behaviour that the fixes leave unchanged.
Purpose
Double-loop learning treats an error at two levels: the action that failed and the rule that produced that action. Argyris illustrates it with a thermostat. A thermostat that turns on the heat as soon as the room drops below 68 degrees Fahrenheit (20 °C) learns in a single loop. A thermostat that could ask "Why am I set at 68 degrees?" and look for a more economical temperature would learn in a double loop.
According to Argyris, learning lasts only if managers and employees also look at their own behaviour. They identify how they contribute to the organisation's problems without meaning to, then change how they act. In particular, they learn that the way they define and solve problems can become a problem in its own right.
The technique supports a decision: correct the action or change the rule that produced it. Its deliverable is the two-column case. In it, the manager writes the planned meeting as a script. Beside each line, the manager notes what they would think and leave unsaid. The group then analyses the case and draws up the list of attributions to test. For each attribution, the list gives the fact that would test it and the person who brings that fact. It ends with the contradictory messages in the case and the rule that links them. An attribution is a cause or an intention that a person ascribes to someone else to explain their behaviour: "the clients don't want to learn". This list is our own format.
Usage
When to use it
- Problem that returns after every fix: examine the rule that produces the action.
- Change adopted, then rejected: find which rule the first difficult case triggered.
- Gap between what is said and what is done: use observation to compare what the team says it wants with what it does.
- Project review where the fault always lies elsewhere: test the complaints about clients and superiors against the facts.
- Recurring conflict in a management team: the two-column case written by the person who leads that team opens the discussion.
- Subject that nobody raises in meetings: the case makes it legitimate to talk about it.
- Senior management ready to start with itself: for Argyris, this is the condition for lasting change.
When not to use it
- Error whose cause and remedy are known: correct the action.
- Technical cause still unknown: establish it first with root cause analysis.
- Senior management that refuses to examine its own reasoning: stick to the deployment plan, for example with Lewin's change model.
Description
Two loops
The single loop takes the result back to the action. The double loop takes it back to the governing values that produced the action. Argyris places learning in the way people think rather than in their motivation. Everyone designs and carries out their actions by rules of reasoning. Argyris compares these rules to a "master program" stored in the brain. Defensive reasoning blocks learning even when the person is committed to learning. It acts like the hidden bugs in a computer program, which produce the opposite of what its designers intended.
Espoused theory and theory-in-use
Everyone holds a theory of action: the rules they use to design their own behaviour and to understand the behaviour of others. These rules become so familiar that people no longer know they apply them.
Asked in an interview or through a questionnaire, a person describes their espoused theory. Observation of their behaviour shows their theory-in-use: the rules that would make sense of what they do. The two often diverge. According to Argyris, people consistently act inconsistently without seeing the contradiction.
| Espoused theory | Theory-in-use | |
|---|---|---|
| Content | The rules a person says they follow | The rules their behaviour shows |
| Source | Interview, questionnaire | Observing the behaviour, then stating the rules that make sense of it |
| The consulting team that Argyris studied | It says it believes in continuous improvement. | It acts in ways that make improvement impossible. |
| The person is aware of it | Yes | Rarely |
Governing values
According to Argyris, most theories-in-use rest on the same governing values:
- remain in unilateral control;
- maximise "winning" and minimise "losing";
- suppress negative feelings;
- be as "rational" as possible, that is, set clear objectives and judge one's behaviour by whether one meets them.
These values protect the person from embarrassment, threat and the feeling of being vulnerable or incompetent. The master program they make up is defensive.
Defensive reasoning seals itself off
Defensive reasoning keeps private the premises, inferences and conclusions that guide behaviour. The person does not test them independently. Their attributions therefore remain untested. This reasoning forms a closed loop that opposing views cannot penetrate. When someone points out the defensive reasoning, the person responds with reasoning that is more defensive still. The conclusions are self-sealing: nobody else can test them.
Argyris bases this finding on fifteen years of studies of management consultants. Almost all the consultants he studied hold an MBA from one of the top three or four US business schools. He extends the argument beyond this profession. These professionals are good at single-loop learning: they apply their disciplines to problems in the outside world and almost always succeed. So they have rarely learned how to learn from failure. When their strategy fails, they become defensive, screen out criticism and put the blame on others.
Openness is not enough
Argyris reports a meeting held by the manager of a consulting case team who wanted to review a client assignment. The client was broadly satisfied. The manager judged that the team had not delivered the value it promised. He invited an outside consultant whom the team trusted, had the meeting recorded and announced that no subject was off limits. That included his own behaviour. After some three hours of discussion about his behaviour, he asked the team what errors it might have made.
The team blamed uncooperative and arrogant clients, then unavailable and poorly prepared managers. It claimed that the clients did not want to learn but never presented any evidence. When someone pointed out the lack of evidence, the team repeated its complaints with more force. Both sides were candid. They talked past each other. Argyris grants that the complaints may have been partly true. He judges them useless: by turning attention to others, the team stops learning. An open meeting stays in a single loop until the participants test their attributions against the facts.
A process designed with its users, then rejected
Another of Argyris's cases concerns the deployment of a change. A consulting firm introduces a new performance-evaluation process. The consultants take part in its design and welcome it with enthusiasm: it meets their espoused values of objectivity and fairness. Two years later, the process arouses dissatisfaction. The first unsatisfactory rating sets it off. Management supports six consultants who perform below the minimum standard, then lets them go.
In the two meetings that the CEO holds, some consultants call the process subjective. Others suspect a hidden economic motive. Others still say it works against learning. They demand documented data from management and bring no first-hand example themselves. Argyris states the rules that their behaviour assumes:
- state a criticism in a way that prevents others from judging whether it is valid;
- when asked to illustrate the criticism, give no data that would let others judge the illustration;
- state conclusions in a way that disguises their logical implications. If someone points out those implications, deny them.
Argyris described these rules to the consultants. They found them abhorrent. In defending themselves against them, they almost always confirmed them.
Argyris concludes that new structures or systems are not enough: people remain locked in defensive reasoning even when they want to improve.
Reasoning productively
Argyris asks managers to bring to human behaviour the "tough reasoning" they use in strategic analysis. In his view, reasoning about human behaviour must be as analytical and as grounded in data as any other management discipline.
| Defensive reasoning | Productive reasoning | |
|---|---|---|
| Data | Hardly objective | The most objective available |
| Inferences | Rarely explicit | Explicit and tested constantly |
| Conclusions | Self-serving and impossible for others to test | Open to verification by others |
Starting at the top
For Argyris, the first step falls to senior managers: they examine their own theory-in-use and change it. Until they do, any change is likely to be a fad. A change in reasoning that starts with the employees looks strange or even dangerous to senior managers who still reason defensively. These managers then believe they show care when they bypass and cover up difficult issues. Their subordinates see the same actions as defensive.
Senior managers trained in productive reasoning can influence the whole organisation even when others still reason defensively. The CEO in the evaluation case defused the dissatisfaction by answering with data and by asking for the facts that would contradict them. Argyris ties the training to the company's business problems: a manager keeps up the effort after seeing the effect of productive reasoning on their own performance.
Running the two-column case exercise
Steps 1 to 4 and 9 follow Argyris's text in our numbering. Steps 5 to 8 are our own arrangement of his analysis.
- Choose a real problem
A problem that the manager wants to address or has failed to solve. A real problem shows the manager the effect of productive reasoning on their own work. - Describe the planned meeting
In one paragraph, the manager describes the meeting that will address the problem. Writing the case usually takes less than an hour. - Split the page in two
On the right, the manager writes the meeting like a film or theatre script: what they will say and what the others will probably answer. On the left, beside each line, the manager notes the thoughts and feelings they would not express for fear of derailing the discussion. - Analyse the case instead of holding the meeting
The manager analyses the script with the people involved. In Argyris's case, the CEO of an organisational development consultancy analyses it with his four direct reports. In our view, an outside facilitator whom the group trusts should lead the analysis. Argyris warns that the discussion can turn emotional or even painful. - Name the attributions
The group picks out the attributions in the left-hand column. It puts Argyris's question to the author: "What did they say and do?" - Name the contradictory messages
The group sets the left-hand column against the right-hand column and picks out the messages that contradict each other. - State the rule
The group states the rule that links the two columns and the governing value it serves. For this CEO, the rule was to feign a consensus for the sake of diplomacy. The author names their own part in the problem. When a participant admits a criticism they had kept to themselves, the author responds with the same candour. The author asks: "Was there anything I said or did that gave you the idea that you had better not raise this with me?" - Decide how to test each attribution
For each attribution, the group decides which fact would test it and who brings that fact. It records the result in the list of attributions to test. - Repeat with the other members' cases
Each member writes their own case. According to Argyris, managers need many opportunities to practise.
The analysis taught this CEO that he was feigning a consensus that did not exist. His subordinates were wary instead of reassured. He urged them to "think of the organization as a whole" while ordering budget cuts that put them in competition with each other. Many of his attributions were wrong: he had never voiced them, so he had never checked them. What he thought he was hiding came through anyway. His subordinates read an additional message in it: the boss was hiding things. They learned too: by helping him analyse his case, then by writing their own.
Applying the technique to a deployment
This application is our reading of Argyris. When the adoption of a solution lags, the single-loop response adds a training session, a reminder, a feature or a deadline. The double loop looks for the rule that makes the current behaviour reasonable for those who persist in it. It also looks for the rule that makes the project team respond with yet another training session. The sponsor or the project manager writes the first case before any workshop with the users. The business analyst prepares the analysis by picking out the attributions in each column. The analyst then looks for the facts that would test them by observing the users or by querying the application data.
AI considerations
To prepare the analysis, a language model picks out the attributions ("the users don't want to change") in the transcript of a meeting recorded with the participants' consent. For each one it indicates whether the speaker cites a fact. The group starts from this list to ask "What did they say and do?" The model also compares the espoused theory written in a project's charter or principles with the decisions recorded in minutes and tickets. It flags the gaps as hypotheses to test.
AI does not fill in the left-hand column. That column records what its author thinks and leaves unsaid. Text generated for it becomes fiction about what a person would think. Nor does AI judge whether a person reasons defensively. That verdict is itself an attribution. Only an exchange with the person tests it. An assistant asked to make the left-hand column "more constructive" strips out what the group needs to discuss. Like the CEO who feigned a consensus, it avoids the issue and calls that diplomacy. The left-hand column names colleagues and judges them. That is personal data within the meaning of the Federal Act on Data Protection (FADP). An author who pastes it into an external AI service discloses it outside the group to which they entrusted it.
Examples
The left-hand column shows what the planned meeting leaves out. The following case is our illustration of the form.
| Line | Left-hand column: what Claire thinks and leaves unsaid | Right-hand column: the planned script |
|---|---|---|
| Case | Claire is a project manager at a health insurer. She is preparing a meeting with Marc and Sonia, who lead two teams in the benefits department. Six weeks after the reimbursement tool went live, 35% of claims still go through the old Excel list. | |
| 1 | They've been dragging their feet from the start. Marc never wanted this tool. | Claire: "Thanks for coming. I'd like to understand why the old list is still in use." |
| 2 | His staff didn't attend the training. He didn't release them. | Marc: "The tool is slow. A claim takes twice as long as before." |
| 3 | The vendor measures two seconds per screen. He's exaggerating. | Claire: "I'll ask the vendor to speed up the screens." |
| 4 | Management expects 100% in the tool by the end of March. I can't tell them that. | Sonia: "In the meantime, can we keep the list for urgent cases?" |
| 5 | If I say yes, the list will stay. | Claire: "Of course. Take the time you need to get used to the tool." |
Claire never voiced the attributions in lines 1 and 2: nobody could test them. In line 3, she orders a single-loop fix from the vendor for a speed problem she does not believe in. In lines 4 and 5, she grants more time with the old list. She considers that extension dangerous and keeps quiet about the March deadline. The second loop concerns the rule she follows: show no disagreement in front of the team leaders.
| Line | Claire's attribution | Fact that would test it | Who brings it |
|---|---|---|---|
| 1 | Marc never wanted this tool. | The minutes of the tool selection: what Marc said and did there. | Claire |
| 2 | Marc did not release his staff for the training. | The training attendance list and the team's schedule for those weeks. | Marc |
| 3 | Marc is exaggerating: the vendor measures two seconds per screen. | The same claim timed in the tool and in the old list. | The business analyst |
| 4 and 5 | Contradictory message: "Take the time you need" against the end-of-March deadline. | ||
| Rule | Show no disagreement in front of the team leaders. Value served: suppress negative feelings. | ||
Visualisations
The two-loops diagram places the governing values, the actions and the results with Argyris's thermostat on each loop. The two theories of action, the two kinds of reasoning, the two-column case and the list of attributions to test are tables.
Cost
| Dimension | Level | Rationale |
|---|---|---|
| Preparation | Low | Choose a real problem and write the case: less than an hour per author. |
| Execution | High | The group analyses each case at length. Each member writes their own case and practises repeatedly. Senior management goes first. |
| Documentation | Low | The cases and the list of attributions to test. The cases stay within the group that analyses them. |
Tools
A sheet of paper folded in half is enough to write the case. A word processor with a two-column table does the job when the author wants to rework it. The case stays with its author until the meeting: a shared project space exposes it to readers who take no part in the analysis.
In the meeting, a physical or online whiteboard (Miro, Mural, FigJam) holds the list of attributions to test. The team manager in Argyris's case had the meeting recorded so that he could later check disagreements about what was said. The automatic transcription in Teams or Zoom fills that role with each participant's consent.
Sources
- Argyris, C. (1991), Teaching Smart People How to Learn, Harvard Business Review, May-June 1991, pp. 99-109.

