mindjump
systems

Feedback delays

Published November 9, 2026 · 3 min read

In an unfamiliar shower, the water is too cold, so you turn the tap towards hot. Nothing happens, so you turn it further. Then the water arrives scalding. You turn it back hard, and a few seconds later you are freezing. The tap is not the problem. The pipe is.

The idea

Any system in which you act, wait and then see the result is a feedback loop. When the result shows up immediately, correcting is easy. When it shows up late, you keep reacting to an old situation. Corrections pile up in the pipe, arrive after the problem is already fixed, and push the system past where you wanted it. The longer the delay, and the bigger each correction, the worse the swings.

You turn the tapWater travels down the pipeYou feel the temperatureCompare with 38°Cdelay:a few secondswhat you feel is the past
The loop that every adjustment goes through. The delay sits between your action and what you can observe.
15°30°38°45°60°0 s20 s40 s60 sno delay3 s delay, gentle turns3 s delay, moderate turnstime after you start adjusting
The same shower with three ways of adjusting (simulated). Without a delay, anything works. With a three-second delay, gentle turns settle in about 12 seconds; moderate turns overshoot to 54°C and take about 50.
Case study · Supply chains

The beer game and the bullwhip

In the early 1960s, Jay Forrester’s system dynamics group at MIT turned this problem into a board game, now known as the beer game. Four players run a retailer, a wholesaler, a distributor and a factory. Each one sees only its own orders and stock, and every order takes several weeks to arrive. Customer demand changes only once during the game, by a small step. Yet in game after game, orders swing wildly and stock runs from empty to overflowing, more violently the further a player sits from the customer. Researchers who studied the results found that the same pattern appears whoever plays: people do not account for the orders already on their way.

Real companies see the same thing. In the 1990s, Procter & Gamble noticed that although babies use nappies at a very steady rate, the orders for them swung more and more the further up the supply chain they went. The company called it the bullwhip effect, after the way a small flick of the handle becomes a large crack at the tip. A 1997 study by Hau Lee, V. Padmanabhan and Seungjin Whang traced it to four causes: updating forecasts on every order, ordering in batches, price promotions, and over-ordering when supplies are rationed. Long delays make each of them worse.

Try it yourselfWhy is the shower never right?Change the delay and watch the water overshoot.

How to use it

  1. Find the delay: how long between an action and the first moment you can see its effect?
  2. When the delay is long, make smaller corrections than feel natural.
  3. After a correction, wait at least one delay before judging it.
  4. Keep track of what is already in the pipe — orders placed, changes made but not yet visible.
  5. Where possible, shorten the delay: measure closer to the action, or use an earlier signal.

Where it misleads

Patience has a cost too. A system that is corrected too timidly takes a long time to reach where it should be, and some situations do need a fast, large response. Delays are also rarely fixed: a pipe that takes three seconds today may take five tomorrow, so a rule tuned to one delay can fail with another.

Finally, not every swing is caused by delay. Some systems are genuinely noisy, and reacting to every wobble — delayed or not — adds swings of its own.

In one line: when results arrive late, turn the tap a little and wait.

Next modelSecond-order effects“And then what?” — the question most plans skip.
mindjump · small thinking tools, written down so I actually use them.