Normal Accidents

by Charles Perrow · 5 ideas

  1. Normal Accidents

    In systems that are complex and whose parts are tightly linked, like power plants or financial markets, serious accidents are not rare flukes. They are a built-in result of how the system is made, and nobody has to do anything unusual for them to happen. In such systems, plan for failure as normal: limit how far it can spread and how much it can hurt.

  2. Interactive Complexity

    In complex systems, parts affect each other in hidden, unexpected ways, so small failures combine into events nobody imagined. Operators cannot see or understand what is happening in time. Reduce hidden links between parts wherever you can.

  3. Tight Coupling

    A system is tightly coupled when steps follow each other fast, with no slack and no way to pause. A failure in one part then spreads before anyone can react. Add buffers, slack and pause points so problems can be caught.

  4. Safety Can Add Risk

    Every added alarm or backup system also adds complexity and new ways to fail. Piling on safety features can make a system more dangerous. Check whether a safety fix adds more hidden links than it removes.

  5. Know Your Quadrant

    Rate any system on two scales, how complex it is and how tightly its parts are linked, and you get four kinds of systems. Only the kind that is high on both makes accidents inevitable. Find where your system sits, and if it is high on both, simplify it, loosen it, or question whether to run it at all.

More on systems