Handling interruptions when you cannot lock the door

A colleague asking a question that takes ninety seconds to answer does not cost you ninety seconds. It costs ninety seconds plus however long it takes to rebuild the state you were holding — which figure was on which line, what you had ruled out, what you were about to do next. For simple work the rebuild is trivial. For anything that requires holding several things at once, it can dwarf the interruption itself, which is why a morning with several short interruptions can produce almost nothing while feeling reasonably busy.

You will see specific figures quoted for this recovery time, often around twenty minutes or more per interruption. Treat those numbers with care: they come from particular studies of particular kinds of work, and the honest position is that the cost varies enormously by task and by person. The mechanism is what you can rely on rather than the figure. Work that requires holding a lot in your head is expensive to interrupt; work that does not is cheap. That distinction is enough to act on, and it points at the practical answer, which is to sort your work by interruption cost rather than trying to eliminate interruption everywhere.

The intervention that does the most in a small business is also the least technical: a predictable window when you are available for questions. 'Come to me any time between four and five' converts a stream of individual interruptions into one block, and it works because most questions are not actually urgent — they are just easier to ask the moment they occur.

Two details determine whether it holds. First, the window has to be genuinely reliable, which means being there and being willing when someone shows up. Cancel it twice and people revert to interrupting, correctly, because the alternative you offered turned out not to exist. Second, it has to be paired with a clear rule for what does not wait: a customer standing at the counter, anything involving money going out, anything where a delay makes it worse. Without that rule people will either sit on something genuinely urgent or ignore the window entirely, and both outcomes are worse than no window. The window is not a barrier, it is a promise, and the promise is that a question saved is a question answered properly rather than a question deflected.

In a space where anyone can see you, availability needs to be readable without asking, because 'are you busy?' is itself the interruption. The signal can be almost anything as long as it is unambiguous and physical: headphones on, a small folding sign on the desk, a specific chair you sit in for focused work, a door pushed to rather than closed.

What makes a signal work is not sophistication but consistency and a stated return time. A sign reading 'back at 2pm' is far more effective than one reading 'busy', because it answers the question the interrupter actually has — when can I ask? Someone who knows the answer is two o'clock will wait; someone told only that you are busy has to decide whether their question outranks your busyness, and they will frequently decide it does. The signal also needs to be off sometimes, visibly. If the headphones are permanent, they carry no information, and people will learn to interrupt through them, which leaves you with no signal at all and a slightly worse relationship with your team.

People interrupt because it works, and because it is the fastest route to certainty. If interrupting always produces an immediate answer, no other behaviour has any reason to develop. Changing that is mostly about changing what interrupting gets you, which has to be done without making people feel dismissed.

The most effective response to a non-urgent interruption is to answer it — at the next window — and to say that is what is happening: 'Good question, bring it at four and we'll do it properly.' Over a few weeks this teaches the pattern without anyone being refused. Two things accelerate it. Make the answers to recurring questions available somewhere people can find them, because a large share of interruptions are the same handful of questions and the person asking would happily look it up if looking it up were reliable. And say out loud which decisions people can make without you, because a great many interruptions are not requests for information but requests for permission, from someone who is not sure whether they are allowed to act. That is a clarity problem, not an interruption problem.

The harder skill is ending a conversation that is already underway and has drifted past its useful point. The reason it is hard is that the natural exit lines all sound like rejection, so most people wait for a gap that does not arrive and lose twenty minutes to politeness.

What works is closing on the substance rather than on the clock. Summarise what has been agreed, name the next step, and attach it to a time: 'So you'll send the revised quote today and we'll look at the supplier question on Thursday — good?' That is a complete ending, and it does not communicate that you had run out of patience. Standing up as you say it is remarkably effective and reads as natural rather than rude, because it matches the shape of an ending. If there genuinely is no conclusion to summarise, be direct and specific instead: 'I need to get back to this filing — can we pick this up at four?' A stated reason with a named alternative time is accepted almost universally. Vague signals of impatience are the version that damages the relationship.

Given that some interruption is unavoidable in a small business, the practical approach is to stop treating all your work as equally fragile. Work that requires holding several things at once — a reconciliation, a contract review, a difficult piece of writing — should go in whatever part of the day is genuinely quietest, even if that means unusual hours. Work that is robust to interruption — routine approvals, calls, filing, tidying up admin — belongs in the busy parts of the day, where being interrupted costs almost nothing.

This is a more reliable strategy than trying to make the whole day protected, because it does not depend on other people's behaviour. It also gives you something useful to do when you have been interrupted out of deep work and cannot get back into it: switch to the robust pile rather than attempting a rebuild that will not take. One honest limitation: this only works if you know which of your tasks are which, and that is not something a tool can tell you. It comes from noticing, over a few weeks, which interruptions genuinely set you back and which you barely felt.