People account for interruptions by their duration, treating a two-minute question as a two-minute cost. The larger expense arrives afterward, in the work of getting back.
Complex tasks require assembled context
Difficult work depends on holding several elements simultaneously, including the goal, the current step, recent decisions and the reasons behind them.
That assembly takes time to build and is not stored anywhere outside the moment, so it does not survive being displaced by unrelated material.
The deeper the task, the more there is to assemble, which is why interruption costs fall hardest on exactly the work people most need to protect.
Reconstruction is slower than it feels
Returning involves rereading, retracing steps and re-deriving conclusions that were already reached, none of which produces new progress.
Because reconstruction resembles working, it does not register as recovery time, and people report having worked continuously through a heavily interrupted day.
The mismatch between hours occupied and progress made is the most common description of such a day, and it has a mechanical explanation.
Anticipation degrades the intervals
When interruption is likely, attention remains partly allocated to monitoring for it, whether that means watching a door, a chat status or a phone.
Monitoring consumes capacity even when nothing happens, so a period of possible interruption is less productive than an equally long protected period.
This is why unpredictable availability demands are costly beyond the interruptions they actually produce.
Self-interruption follows the same pattern
A significant portion of switching is initiated by the person working, often triggered by a moment of difficulty in the task itself.
Difficulty is uncomfortable, and switching relieves it immediately, which trains the pattern regardless of any intention to concentrate.
The costs are identical to externally imposed interruptions, but the responsibility is easier to place, which is why self-interruption gets moralized while open-plan noise does not.
Batching works because it protects assembly
Grouping messages, calls and small requests into defined periods leaves other periods with enough continuity for context to be built and used.
The benefit comes from the protected blocks rather than from the batched handling, which is why partial protection yields much less than it appears to promise.
Sustained inability to hold attention on work that once felt manageable, across environments and over time, is worth discussing with a clinician rather than restructuring a calendar again.