A feedback loop is any arrangement where what comes out goes back in. It sounds abstract and is the most common structure in an operating business, which is why so many business problems resist being fixed by working harder on them.

There are two kinds and they behave completely differently.

Reinforcing loops

More produces more. A satisfied client refers another client, who is also satisfied, who refers another. Or: a rushed job produces a defect, the defect produces rework, the rework consumes the time that would have prevented the next defect.

Reinforcing loops do not care about direction. The same structure that produces growth produces collapse, and it runs at the same speed either way — which is compounding seen as a mechanism rather than as an arithmetic.

Two things follow. They are slow at first and then not, so by the time a reinforcing loop is obvious it is already large. And the intervention point is early, when the loop is small and cheap to break, which is exactly when there is no evidence that it matters.

Balancing loops

More produces less. A queue grows, so people work faster; the queue shrinks. Prices rise, so demand falls; prices come down.

Balancing loops are why things resist change, and why an intervention that should obviously work often does nothing. Cut the workload and people relax into the remaining time — Parkinson's law is a balancing loop with a goal set by the deadline rather than by the work. Add staff to a late project and communication overhead rises to consume the capacity, which is Brooks's law.

The lesson is not that intervention is futile. It is that a balancing loop has a goal implied by its structure, and pushing against a loop without changing the goal produces effort and no movement. Change what the system is holding constant, or accept the level.

Delay is what makes them dangerous

A loop with an immediate response is manageable, because the effect of an action is visible. Put a delay in and the same loop becomes almost impossible to control by feel.

The shower with slow hot water is the standard example and the business version is everywhere: hiring decisions whose results appear in six months, marketing spend that pays back over a year, technical shortcuts whose cost arrives in the third year. The actor cannot see the consequence of their last action before taking the next one, so they overcorrect, and the overcorrection arrives on top of the delayed effect of what they already did.

This is the structural reason second-order effects get missed. Not inattention — the consequence genuinely is not visible yet.

Finding them

Three signs, and they are recognizable without any modeling.

A problem that comes back after being fixed. If the same issue recurs after a real intervention, the intervention was on the symptom and the loop is intact.

Effort producing no movement. A balancing loop absorbing the pressure. The thing to look for is what is being held constant, and by whom.

Something accelerating for no new reason. A reinforcing loop that has been running for a while and has just become visible.

The useful question in all three is not what caused this. In a loop every element is both cause and effect and the search for a first cause has no natural end. The question is which link would be easiest to break, and the answer is often something nobody has been treating as a decision at all — a default, a policy, the order two things happen in.