Double-loop learning

In plain terms

Every organization corrects mistakes. The question is how deep the correction goes.

In single-loop learning, you notice that something did not work and you try a different way of doing it. The goal and the assumptions behind it stay the same. In double-loop learning, you also ask whether the goal was right, and what you believed that led you to choose it.

Think of a thermostat. A thermostat that turns the heat on when the room is cold is learning in a single loop. Asking whether the room should be kept at that temperature at all is the second loop.

Where it came from

Chris Argyris, an organizational theorist who taught at Yale for 20 years before moving to Harvard, developed the distinction (Harvard Graduate School of Education, 2013). In a 1976 paper in Administrative Science Quarterly, as quoted by Sadki (2024), Argyris defined double-loop learning as occurring when errors are corrected by changing the governing values and then the actions, and single-loop learning as correcting errors without altering those values (Argyris, 1976).

He brought the idea to managers in a 1977 Harvard Business Review article, "Double Loop Learning in Organizations." It opens with a company whose top management dropped a product after losses of more than 100 million dollars, although at least five people, three plant managers and two marketing officials, knew it was in serious trouble six years before (Argyris, 1977).

With Donald Schön he wrote Organizational Learning: A Theory of Action Perspective (Argyris & Schön, 1978). In a review of his work, Clark (2021) describes how Argyris, drawing on John Dewey and Kurt Lewin, identified defensive routines that block learning at the individual, group and organizational levels, and promoted a shift from single-loop to double-loop learning and from defensive to productive reasoning, which he called Model I and Model II.

Argyris died in November 2013 at 90. Harvard's Graduate School of Education describes him as widely regarded as the father of organizational learning (Harvard Graduate School of Education, 2013; Yale Alumni Magazine, 2013).

How it works

Argyris's framework rests on a few linked ideas, as Clark (2021) summarizes them.

• Learning is error correction. An error is any mismatch between what you intended and what happened (Argyris, 2015).
• Governing values sit behind actions. Argyris held that errors can be corrected by changing the actions alone (single-loop) or by changing the governing values and then the actions (double-loop) (Argyris, 1976, as quoted by Sadki, 2024). Single-loop learning changes the action and leaves those values alone.
• What people say and what they do can differ. Argyris argued that people are often unaware of the extent to which their own actions inhibit learning (Clark, 2021). Double-loop learning requires becoming aware of one's theories-in-use, the theories that actually guide action (Sadki, 2024).
• Defensive routines get in the way. Argyris sought to identify defensive routines that inhibit learning at the individual, group and organizational levels (Clark, 2021).

The ladder of inference is a related tool. It slows down the jump from what someone observed to what they concluded.

What the research says

A 2023 review argues the idea is popular but has had a superficial impact. Auqui-Caceres and Furlan (2023) reviewed 128 studies of double-loop learning published from 1974 to 2021. They argue that although it is arguably one of the most popularized categorizations of organizational learning, it has left a superficial impact on research and practice, because it is complex to define and hard to put into practice. This is the authors' argument from a literature review (abstract read), not a test of whether double-loop learning works. The review is not graded in the Evidence Library, and no result from it is used.

The Evidence Library holds no direct test of double-loop learning. It does hold graded evidence on practices that ask teams to reflect on what happened and why, which is the raw material for both loops.

Debriefs were linked to better performance in a meta-analysis. Tannenbaum and Cerasoli (2013) pooled 46 samples with 2,136 participants. Debriefs were linked to about 25 percent better performance than comparison conditions (d = .67; .54 with outliers removed), with similar results for teams and individuals and somewhat smaller effects in real work settings than in simulations (rated High after a full-text read). Many of the pooled studies were quasi-experimental.

Two features were linked to better after-action reviews. A later meta-analysis found after-action reviews linked to an overall gain (d = 0.79 across 61 studies), and found that two features were linked to larger effects: aligning the review to the individual or the team, and using objective media to review performance (Keiser & Arthur, 2021, rated Medium from its abstract).

One company paired problem solving with learning. In a LEGO department of about 100 staff, compared with a control department of about 50, lean problem solving with leaders acting as learning facilitators went with larger gains in survey-rated learning organization dimensions, efficiency and quality over 18 months (Kristensen et al., 2022, rated Medium after a full-text read). The departments were not randomized and it is one company.

Leaders who reflect in groups. In primary schools, school leaders' participation in a group reflective learning program went with a significant increase in teachers' perceptions of the organizational learning climate (Daniëls et al., 2021, rated Medium from its abstract). The abstract does not describe the comparison group.

Where the evidence is thin

1. No direct test. The studies above test debriefs and reflective programs, which can produce single-loop or double-loop learning. None of them measures whether teams questioned their goals and assumptions.
2. Measurement. Auqui-Caceres and Furlan (2023) point to a proliferation of empirical studies of double-loop learning that ignore its original notion (abstract read).
3. Defensive routines. Argyris's account of defensive routines is theory; the Evidence Library holds no graded test of interventions designed to reduce them.

How it maps to the Unfrozen Loop

1. Week 0
   Before you start
2. 1
   Surface both ways

   Argyris's 1977 article opens with a product that people close to the work knew was in serious trouble six years before the top decided to drop it. Double-loop learning needs bad news to travel up first.

3. 2
   Make sense together, top included
4. 3
   Start it close to the work
5. 4
   Review on a rhythm, then hold the pace

   A regular review is where single-loop and double-loop questions get asked: did we do what we planned, and is the plan still right? The debrief research below is the closest graded evidence for reviewing on a rhythm.

6. 5
   The middle’s own weather
7. Ring
   Safe to speak, and speaking changes something

   Argyris argued that defensive routines inhibit learning. Questioning a goal in front of the people who set it is easier where it is safe to speak.

Highlighted parts are the ones this idea speaks to. The mapping is my reading, not a tested finding. See the whole Loop and the evidence for each part.

Common misreadings

"Double-loop is always better." Many errors need only a better method. Double-loop questions are worth asking when the same kind of problem keeps coming back, which can be a sign that the goal or assumption behind it is the problem.

"It is a training topic." Argyris argued that people are often unaware of the extent to which their own actions inhibit learning (Clark, 2021). Knowing the term does not change the conversation.

"Any review is double-loop." A review that asks only whether the plan was carried out is single-loop. The second loop starts when someone asks whether the plan is still the right one.

Using it well

1. Add one second-loop question to your next review. After "What happened, and what will we do differently?", ask "What did we assume that turned out to be wrong?" Write the answers down and keep them.
2. Use records, not memory. Bring the actual numbers, timeline or customer comments to the review. Keiser and Arthur (2021, abstract read) found that objective performance review media were linked to larger after-action review effects.
3. Ask the people closest to the work first. In Argyris's product example, the plant managers and marketing staff knew years before the top did. Let them speak before the senior person in the room does.