Single-Loop vs Double-Loop Learning
Why the same problem keeps coming back, in slightly different clothes — and what to do about it.
You fix the thing. A few months later, it’s back. Different shape, different person involved, same essential dynamic. You add a process; someone works around it. You hire someone new; the team absorbs them and the pattern continues. You have the conversation; nothing actually changes.
This is one of the most common, and one of the most exhausting, experiences of running a small business. And it is almost always a sign that you’re solving the wrong problem.
Chris Argyris, who spent most of his career studying how organisations learn — or don’t — gave us a simple but precise way to see what’s happening. He called it the difference between single-loop and double-loop learning.
Who was Argyris
Chris Argyris taught at Yale and Harvard and spent decades looking at why intelligent, well-meaning people inside organisations kept producing dysfunctional outcomes. His work, much of it co-authored with Donald Schön, is foundational to what we now call organisational learning. The single-loop and double-loop distinction is one of his clearest gifts to anyone trying to fix a business.
The cruise control example
The simplest way into this framework is the cruise control on a car.
You set the speed at 100. The road curves uphill, the engine slows, the cruise control responds by adding fuel to bring you back to 100. The road levels out, the engine speeds up, the cruise control eases off. The system is doing its job. It’s detecting the gap between the goal and the reality, and correcting.
This is single-loop learning. The goal stays fixed. The behaviour adjusts to meet the goal. Most organisations are very good at single-loop learning. It’s how processes get tightened, KPIs get hit, and the day-to-day actually runs.
But cruise control doesn’t ask whether 100 is the right speed.
If the road becomes icy, or there’s traffic ahead, or you’re suddenly in a school zone, the cruise control still happily maintains 100 — because nobody has asked the second question. That second question — “is the goal itself still right?” — is double-loop learning. It’s the willingness to step back and challenge the assumption underneath the action, not just the action itself.
The reason the same problem keeps coming back in your business is usually that you’re using single-loop learning on a problem that requires double-loop learning. You keep adjusting the speed. You don’t change the destination.
What single-loop looks like in business
These are everyday, often well-intentioned moves that stay inside the loop.
“Let’s add a process”
Something has gone wrong, so you write a checklist or add an approval step. The new process catches the immediate error. It does not ask why the error happened — what assumption, structure, or relationship made it likely. Six months later you write another checklist for the next thing.
“Let’s hire someone new”
You bring in a new person to replace, support, or fix what isn’t working. The new person is competent. Six months in, the same dynamic has reasserted itself, with a different name attached. The dynamic was never about that person. It was about the system. The hire was a single-loop response.
“Let’s have a conversation”
You sit the team down. You raise the issue. People nod. Action items get written. Nothing changes, because the conversation operated inside the existing assumptions about what is and isn’t sayable, and the actual cause was on the other side of that line.
“Let’s set a target”
The number isn’t where you want it, so you set a higher target and apply pressure. The target moves the behaviour briefly. It doesn’t ask what’s structurally producing the lower number, so the underlying cause keeps producing it.
“Let’s try harder”
The most common single-loop response. Effort is doubled. Hours go in. Something improves for a while. The pattern returns when the extra effort can no longer be sustained, because nothing about the system has actually changed.
None of these are wrong. They are appropriate responses to many problems. They become a trap when they’re applied to a problem that requires double-loop work — and you don’t notice that’s the problem you have.
What double-loop learning requires
Double-loop learning starts with a different question. Not “how do we fix this?” but “what assumption is making this keep happening?”
That sounds simple. It is genuinely uncomfortable in practice, because the assumption underneath a recurring problem is usually one that someone in the room — often the owner — has been protecting without realising it. It might be a belief about how the business should be run, a relationship that hasn’t been addressed, a decision made years ago that nobody has revisited, or an unwritten rule about what’s allowed to be said.
In Argyris’s framing, double-loop learning asks the team to do three things.
First, surface the assumption. Make it visible. Write it down. “We don’t say no to that client.” “We hire family.” “The founders don’t get challenged.” “We don’t review the strategy until something breaks.”
Second, test it. Is it still serving the business? Was it ever? What would happen if it weren’t true? What would you do differently if you assumed the opposite?
Third, change it — or consciously keep it. Double-loop learning is not about throwing everything out. The point is that the assumption becomes a choice rather than a default. Sometimes you decide the assumption is sound and keep it. What changes is that you stopped running on autopilot.
This is harder than single-loop work because it usually involves admitting that something the business has been doing for a long time isn’t working — and that admission lands somewhere personal.
Three questions to start with
Pick a problem in your business that has come back more than twice in different forms. Bring these three questions to it.
What rule, assumption, or unspoken agreement is making this keep happening? Not the surface cause. The thing underneath that the surface cause depends on. If you can’t find it, ask the person closest to the problem what they would change if there were no consequences for saying.
What are you doing instead of changing that assumption? This is usually where you’ll find the single-loop pattern — the process, the hire, the conversation, the target. What’s the workaround you’ve been running for years?
What would have to be true for that assumption to change? Sometimes the answer is a decision. Sometimes it’s a conversation you’ve avoided. Sometimes it’s information you don’t have. Naming what would need to be true makes the next move concrete.
Where this connects
Once you start looking for the assumption underneath a recurring problem, you’ll often find it sitting inside how people communicate. The reason the assumption stays invisible is usually that nobody is willing — or able — to say it out loud. That’s the territory of Model I vs Model II Communication, which explains why otherwise-honest people stop telling the truth at work.
If your problem is that the work has gone heavy and the team has gone quiet, the deeper version of that conversation is in Joy at Work — Deming’s framework for the conditions under which good people do good work.
Frequently asked questions
What is single-loop learning?
Single-loop learning adjusts behaviour to meet an existing goal without questioning the goal itself. Argyris and Schön named it after the way a thermostat or cruise control works — detecting a gap between target and reality, then correcting. Most day-to-day operational improvements in a business are single-loop.
What is double-loop learning?
Double-loop learning steps back and questions the goal, the assumption, or the value underneath the behaviour. It asks not “are we doing this right?” but “are we doing the right thing in the first place?” It’s harder, less comfortable, and the only kind of learning that prevents the same problem from coming back.
Why does the same problem keep coming back in my business?
Usually because you’re using single-loop responses — adding processes, setting targets, hiring people, trying harder — on a problem that needs a double-loop response. The assumption producing the problem hasn’t been surfaced or challenged, so the problem keeps regenerating in different forms.
Is this relevant to a small business?
Yes. Argyris’s distinction applies at any scale. In a small business, double-loop learning is often easier in principle (fewer people, less bureaucracy) and harder in practice — the assumption underneath the problem usually involves the owner, and there’s nowhere for that conversation to hide.
How do I tell which kind of learning a problem needs?
A useful test: if you’ve fixed it before and it’s come back in a different shape, single-loop responses are no longer the right tool. The fact that the problem is recurring is the signal that the assumption underneath it hasn’t been touched yet.