Feedback Loops

How it works

There are two types of feedback loops that drive every system you’ll ever encounter.

Reinforcing (positive) loops amplify change in one direction. Good performance → more confidence → better performance → even more confidence. Or: losing customers → less revenue → worse product → losing more customers. These loops create exponential growth or exponential decline. They feel slow at first, then suddenly overwhelming.

Balancing (negative) loops push a system toward a target or equilibrium. Room gets cold → thermostat turns on heater → room warms up → thermostat turns off heater. These loops create stability. They resist change in either direction.

Most real-world situations are shaped by multiple feedback loops operating simultaneously, often in tension with each other. Understanding which loops dominate — and when they might shift — is the core skill of systems thinking.

Case study: How a thermostat failure melted down Three Mile Island

On March 28, 1979, a pressure valve in the Three Mile Island nuclear reactor stuck open, allowing coolant to drain from the reactor core. This was a mechanical failure — but the true disaster came from a feedback loop failure in the control room.

The operators had an indicator light showing the valve’s electrical signal — which said “closed.” But the light showed the signal sent to the valve, not the valve’s actual position. The valve was open. The indicator said closed. The operators reduced emergency cooling because their instruments told them the system had too much coolant, when in fact it had too little.

A negative feedback loop (temperature rises → add coolant → temperature drops) was supposed to keep the reactor stable. But faulty instrumentation turned it into a positive feedback loop (temperature rises → operators reduce coolant because indicators show excess → temperature rises more). Understanding which loops you’re actually in — not which loops you think you’re in — is the difference between stability and catastrophe.

Real-world examples

Social media. Reinforcing loop: more engagement → algorithm shows your content to more people → more engagement. This creates viral content and addictive usage patterns. Balancing loop: too much screen time → feeling bad → using the phone less. When the reinforcing loop is stronger than the balancing loop, you get addiction.

Compound interest. The canonical reinforcing loop: money earns interest → interest gets added to principal → more money earns more interest. Einstein may not have actually called it the eighth wonder of the world, but the structure explains why small differences in savings rates produce enormous differences over decades.

Team morale. Reinforcing loop (virtuous): team ships a win → morale rises → team works better together → team ships another win. Reinforcing loop (vicious): team misses a deadline → blame and frustration → trust erodes → team misses another deadline. Same structure, opposite direction.

When to use it

Look for feedback loops whenever a situation seems to be accelerating (getting better or worse faster over time), whenever you want to understand why a system is stuck (look for the balancing loop holding it in place), whenever you’re designing an intervention (which loop are you trying to strengthen or weaken?), and whenever small causes seem to produce disproportionate effects.

Common mistakes

The biggest mistake is ignoring delays. Feedback loops often have time lags — the effect of today’s action shows up weeks or months later. This makes the loop invisible in the short term and overwhelming in the long term. The second mistake is trying to fight a reinforcing loop with willpower instead of changing the structure. If the loop exists, it will reassert itself. Change the structure that creates the loop.

Try it now

Pick a situation in your life that seems stuck or accelerating. Draw the loop: A leads to B, B leads to C, C leads back to A. Is it reinforcing (amplifying) or balancing (stabilising)? If it’s a vicious reinforcing loop, where could you break or weaken one link in the chain?

Related models

These models complement Feedback Loops — they address similar situations from different angles.