The Context-Switching Cost in a Small Shop
Why this matters
The usual advice about interruptions is personal: turn off notifications, batch your email, protect your focus. All fine, and all of it treats the cost as something happening inside one person's head. In a small shop it is not. It is a routing problem, and it concentrates on whoever is the answer to everything, which is almost always the owner.
That distinction changes the fix. If the cost is personal, you manage it with discipline. If it is structural, discipline buys you a few weeks and then the shop routes around it, because your crew is not interrupting you for fun. They are interrupting you because you are the shortest path to an answer they need to keep working.
A sibling card covers the case for batching and what switching costs an individual. This one is about measuring your own number and reducing the count rather than the cost.
Three things that get called the same thing
They behave differently and they have different fixes, so keep them separate.
An interruption is an external grab at your attention. Someone calls, walks up, messages. An interruption is not automatically expensive. One that lands while you are already in Talk mode costs close to nothing.
A switch is a change in the kind of work you are doing. Switches carry a resumption cost - the time between returning to the original work and actually being back in it. An interruption only costs you when it causes a switch.
Fragmentation is the shape of your day: the distribution of your unbroken block lengths. This is the one that determines what work is possible at all. A day with six hours of working time in eleven fragments and a day with six hours in three blocks contain the same hours and cannot do the same work.
Most owners try to reduce interruptions. The higher-yield targets are the second and third.
Why the cost concentrates on one person
In a five-person shop where the owner holds the answers, every person's question terminates at the same place. Add a sixth person and you have added a source of questions without adding a destination. The load on that one person rises with headcount even though nobody decided it should, and it rises fastest in the months right after a hire, when the new person genuinely does not know things.
Two consequences follow that are worth being explicit about.
The first is that hiring, on its own, makes this worse before it makes it better. An owner drowning in questions who hires help will usually see the question load rise for the first month or two. That is not a sign the hire was wrong. It is a sign that the answers still live in one head.
The second is that the shop cannot see the cost. Each person is asking you two or three questions a day, which from their side is entirely reasonable. From your side it is twenty. Nobody is behaving badly and the aggregate is unworkable, and that is exactly the kind of problem that never surfaces in a conversation because no individual is doing anything wrong.
Measuring your own resumption cost without a stopwatch
You do not need a lab. You need ten instances and a pen.
The next ten times you are interrupted while doing Quiet work - an estimate, the numbers, writing something down - note the time the interruption ends and the time you are genuinely back in the work, meaning you have stopped re-reading and are producing again. Two timestamps, ten times.
Take the median, not the average. One instance where you never really got back and drifted for 40 minutes will drag an average somewhere unusable. The median tells you what a typical resumption costs you.
Do the same for one other mode if you want the comparison, and expect a clear ordering rather than any particular number: shallow work resumes almost immediately, a conversation resumes in a couple of minutes, and quiet work resumes in something on the order of ten minutes or worse. The absolute values are yours and will differ from anyone else's, which is why measuring beats reading a figure in an article.
One mode has a resumption cost that is not about attention. When you are interrupted with hands on live work, resuming safely is a physical verification, not a mental one. Before you touch anything again, re-verify the state you left rather than trusting your own memory of it: confirm the circuit you isolated is still off at the disconnect and still locked and tagged, and test it dead again at the point of work; confirm the valve you closed is still closed and the line is still relieved; confirm the gas cock you shut is still shut. The assumption that costs someone a hand is almost always an assumption about your own earlier step, made after a conversation you did not choose to have.
Two levers, and only one of them scales
Cost per switch is the resumption time. You can shave it with a resumption note - one line, before you turn away, naming what you were about to do next. It genuinely helps. It also has a floor, and personal technique cannot take a ten-minute resumption to zero.
Count of switches is the number of times a day the shop pulls you out of one mode into another. This is where the leverage is, because it has no floor and because it is fixed at the shop level rather than the personal level.
An owner who halves the count has done more than an owner who shaves a third off the cost, and the count is reduced by changing where answers live, not by trying harder.
What actually reduces the count
Four structural changes, each of which retires a specific class of question. Name the class before you build the fix, or you will build something that retires nothing.
| Fix | Question class it retires | What it looks like |
|---|---|---|
| A written authority ceiling | "Can I / is it okay if" | Two stated limits, anchored to a multiple of a routine service call, plus time equivalents |
| A single source of location facts | "Where is / who is / which one" | Supplier contacts, priority customers, current pricing, all live in a record everyone can open |
| Default answers for the top recurring customer questions | "What do I tell them" | The five questions your phone gets most, answered in writing, with the boundary of when to escalate |
| A stated question window | Nothing, but it batches the rest | Two fixed windows a day when you are reliably available, and a capture list for everything between |
Note the last row honestly. A question window does not remove questions. It converts a stream of interruptions into a batched conversation, which is worth real money in fragmentation terms while doing nothing about the underlying dependency. Use it, and do not mistake it for a solution. A shop that runs on question windows alone still stops when the owner is out.
The first three are the ones that scale, and they share a property: they move an answer from a head into a place. That is the actual work. Every fix that does not move an answer out of your head is a scheduling accommodation.
A worked measurement
A three-tech shop. The owner counted 213 interruptions across 10 working days, about 21 a day.
On one representative day with 19 interruptions, the owner marked which ones caused a mode switch: 6 landed while already in Talk mode and cost close to nothing, 13 caused a switch, and 4 of those 13 landed during the afternoon Quiet block.
Measured median resumption after a Quiet interruption, over ten instances: 12 minutes.
What that did to a 90-minute block. Four interruptions, roughly 10 minutes of direct handling between them, plus about 48 minutes of resumption at the measured 12-minute median. That leaves roughly 32 of the 90 minutes as productive time, about 36% of the block. The owner's own account of that afternoon was "I got interrupted a few times but I got through it." The measurement says two thirds of the block was overhead.
What changed it. The owner sorted a week's interruptions into classes and found the two largest were "can I" and "where is." A written authority ceiling and a shared record of supplier contacts and priority customers took about two hours to produce between them.
On a comparable day three weeks later, Quiet-breaking interruptions were down to 1. One interruption, about 3 minutes handling it, about 12 minutes resumption, leaving roughly 75 of the same 90 minutes productive, about 83% of the block. Same block length, same owner, roughly 2.3 times the productive minutes inside it.
Worth being precise about what did and did not move. Total daily interruptions fell, but not dramatically - the shop still had questions. What changed was which interruptions reached the Quiet block, because the two classes that were retired happened to be the classes that arrive in the afternoon. Reducing the count in the right window beat reducing the count overall.
The threshold that matters is the run, not the total
The most useful number to track after your resumption median is your longest unbroken run, recorded daily. It is one number, it takes no effort, and it predicts what work you can do.
The working threshold: Quiet work needs an unbroken run of about 45 minutes per instance to be worth starting. Below that, the resumption overhead consumes enough of the run that you will produce a worse estimate or a shallower read of the numbers than if you had waited. If your longest run of the day is under 45 minutes, put Shallow work in the slot deliberately rather than starting Quiet work you will not finish, and move the Quiet work to a day or a placement that can hold it.
The number to tune is yours: 45 minutes assumes a resumption median in the neighborhood of 10 to 12 minutes. If yours measures higher, your minimum viable run is longer, and a shop that cannot produce a run that long has a structural problem no calendar technique will fix.
References
- See related: Batching vs Switching: The Cost of Interruptions
- See related: Managing Interruptions on a Job
- See related: How to Group Your Day by Mode Instead of by Task
- See related: The Owner Bottleneck: When Everything Needs My Approval
- Trade-standard practice for work interruption and resumption in field-service operations