Single-loop and double-loop learning

Argyris, an organisational psychologist, and Schön, a philosopher, built the idea through the 1970s as part of a wider theory of action, set out in books including Theory in Practice (1974) and Organizational Learning (1978). Argyris took it to managers through the Harvard Business Review, in the late 1970s and again in 1991 with 'Teaching Smart People How to Learn'. The framing owes a debt to Gregory Bateson's earlier work on levels of learning.

The model has aged well for an unflattering reason. The behaviour it describes, teams working furiously on the wrong problem because nobody will question the brief, is still what most weeks in most organisations look like.

Single-loop and double-loop learning compared

Single-loop learning: getting better at the current game

Single-loop learning is error correction inside fixed assumptions. You detect a gap between what you wanted and what happened, and you change what you do. Argyris used a thermostat as the illustration: it senses the room is colder than its setting and turns the heat on. It never asks whether the setting is sensible.

This is not the inferior loop. It is most of the work, and a team that cannot do it reliably has no business questioning strategy. The failure is not using single-loop learning; the failure is using only single-loop learning on a problem that will never yield to it.

When single-loop is the right loop

• The goal is genuinely agreed and the gap is one of execution, skill or capacity.
• The problem is new and you have no pattern yet. Fix it first, then look for the pattern.
• The cost of stopping to re-examine assumptions is higher than the cost of the error.

Double-loop learning: changing the rules the actions serve

Double-loop learning goes back a step. Before deciding what to do differently, you ask what belief, target or norm made the failing action seem correct. The thermostat that asked why it was set where it was, and whether anyone in the building agreed, would be running a double loop.

Single-loop learning asks whether you are climbing efficiently. Double-loop learning asks who chose the wall.

The reason it is rare has nothing to do with intelligence. Running the second loop makes somebody's earlier judgement wrong, usually somebody senior, sometimes yourself. So the organisation develops what Argyris called defensive routines: polite, well-practised behaviours that protect people from embarrassment and reliably prevent the awkward question from being asked. He called the result skilled incompetence, because it takes real skill to avoid a subject that thoroughly.

Signals you have a double-loop problem

• The same issue reappears in the retrospective under a new name every quarter.
• Your fixes keep taking the form of more effort, more meetings or more process.
• People agree in the room and dissent in the corridor. Ask why the room is unsafe, not who is disloyal.
• The metric improves and the customers, or the staff, get worse.

Espoused theory and theory-in-use

The engine underneath the model is a second distinction. Your espoused theory is what you say you believe about how to act. Your theory-in-use is what your actual behaviour implies you believe. Almost everyone has a gap between the two, and almost nobody can see their own.

Argyris and Schön argued that under pressure most people fall back on a common set of governing values, which they labelled Model I, and that a more productive alternative, Model II, has to be learned deliberately.

The uncomfortable finding Argyris kept reporting is that people espouse Model II almost universally and act on Model I almost universally, including the people who teach Model II. Awareness alone does not close the gap. Practice under observation does.

How to run a double loop with a real team

1. Start with the pattern, not the incident. Bring three occurrences of the same failure to the table rather than one. A single incident invites blame; three invite curiosity about the system.
2. Write down the governing value out loud. Finish this sentence together: 'we do it this way because we believe...'. Half the time the belief is a rule nobody has tested since the person who set it left.
3. Make your reasoning visible, then invite attack. Say how you reached your view and what would change your mind. 'Here is my logic, what am I missing?' is the single most useful sentence in the model.
4. Name the undiscussable. Ask directly what people believe cannot be said here. Then respond to the first answer well, because everyone is watching to see whether it is punished.
5. Decide which loop the fix belongs to. Close the meeting by saying explicitly: this one is an execution fix, that one is a change of assumption, and this owner has the authority to make it.

The two-column exercise

Argyris's most portable tool costs nothing. Take a conversation that went badly. On the right of a page, write what was actually said, line by line. On the left, write what you were thinking and feeling and did not say. Then read the left column and ask why it stayed there.

It works because the gap between the columns is your theory-in-use, in your own handwriting, with no researcher required. Most people find they were protecting the other person from a conversation the other person would have welcomed. It pairs naturally with giving constructive feedback, which is where the unsaid column usually does its damage.

Where the model breaks down

1. The two loops are cleaner on paper than in a room. Whether a change counts as adjusting an action or altering a governing value is often a judgement call, and people argue the label instead of the problem.
2. It underplays power and incentives. Argyris treats defensive routines largely as psychological habits. Frequently they are rational: in an organisation that has punished the last three people who questioned the strategy, silence is not a defect of reasoning, it is an accurate reading of the environment. Fix the consequences before you diagnose the mindset.
3. Not every problem deserves a second loop. Teams that discover double-loop learning sometimes relitigate settled decisions forever and ship nothing. Questioning assumptions is a tool, not a personality.
4. Self-diagnosis rarely works. The whole point is that your theory-in-use is invisible to you. Without a colleague, coach or facilitator willing to reflect your behaviour back, most people simply conclude they are already Model II.
5. The evidence base is interpretive. Argyris and Schön worked mainly through intervention case studies rather than controlled trials. The model is a persuasive lens, not a measured law, and it should be judged on whether it changes what your team does.

What about triple-loop learning?

Triple-loop learning is a later addition by other writers, not part of Argyris and Schön's original pair, and definitions vary considerably between authors. It generally means learning about how you learn: examining the way your organisation decides what counts as evidence, who is allowed to raise a question, and how it reviews its own review processes. Useful as a prompt. Treat any confident, precise definition of it with suspicion.

Why this is easier to see than to explain

You can teach the two loops in ten minutes. The gap between espoused theory and theory-in-use is a different matter, because everyone believes they have already closed it. That is why the model belongs in experiential learning rather than a slide deck. Put a team into a short challenge with an ambiguous brief and a faulty assumption buried in it, and watch: they will optimise the given task at speed, and the one person who asks 'should we be doing this at all?' gets talked over inside a minute. Play that back in their own words and the point stays. It is the difference between knowing what a defensive routine is and watching yourself perform one.

Tour De Force runs this as live, experiential training for teams worldwide — online and in person. Talk to us, or play Gamified learning appsThe Weekly Challenge to see the method in ten minutes.