The eighty-fourth article corrected two of our own errors. This one is about a number in wide circulation that we could not find in the paper it is credited to.
Key Takeaway
The 2008 abstract reports: "surprisingly, people completed interrupted tasks in less time with no difference in quality," at the price of "more stress, higher frustration, time pressure and effort."[1] A practitioner page cites that same paper for the claim that "it takes over 23 minutes to refocus."[4] We could not find that figure in it. And on our own arithmetic, the two most-repeated numbers in this area are logically incompatible as usually stated.
The Verdict, Stated First
Five claims, in descending order of confidence.
One. The headline finding runs opposite to its popular version. The abstract reports interrupted tasks completed in less time with no quality difference, and the popular version has people losing enormous amounts of time.
Two. The real cost the paper reports is not time at all. It is stress, frustration, time pressure and effort, which is a serious finding and a different one.
Three. A widely repeated figure is attributed to a paper we could not find it in. One practitioner page cites the 2008 study directly for it; a more careful one attributes it to a different and later work.
Four. The two most-quoted numbers are incompatible. On our own arithmetic, switching every three minutes while requiring twenty-three minutes to recover would mean nobody ever completes anything.
Five. The practical case for batching survives all of this, and on our own arithmetic it holds at every switch cost we tested, including a switch cost of one minute.
Our Grades For These Claims
Applying the scheme from the first article in this series.
Grade A for the 2008 findings, from an abstract obtained verbatim from the publisher's digital library and confirmed against a bibliographic service.
Grade A for the study's design, from a copy of the paper hosted by the first author at her own university.
Grade A for the misattribution, which is directly checkable: a page states the claim and names the paper, and the paper's abstract does not contain it.
Grade D for the twenty-three minute figure itself, which we did not trace to any paper we obtained and therefore report only as a claim in circulation.
Grade A for our own arithmetic, which is exact, and which we checked after making a claim we had not printed the supporting row for.
A Note On Method
Everything here is verified to August 2026.
We obtained the 2008 abstract verbatim from the publisher's digital library[1] and confirmed it against a bibliographic service which supplies an additional sentence[2].
We obtained design details from a copy of the paper hosted by the first author at her own university[3].
Two sources are practitioner pages rather than research[4][5], flagged at every use. They are the evidence for how the finding circulates, not for what is true.
We did not obtain the paper's results section, so we report no effect sizes from it, and we did not obtain any source for the twenty-three minute figure, which we therefore do not endorse or refute.
All arithmetic is ours. The switch costs and frequencies are invented, because no source we obtained supplies a switch-cost figure at all.
This article discusses research on work interruption. It is not medical, psychological or management advice.
The Paper Everyone Cites
The source.
Mark, G., Gudith, D., and Klocke, U. (2008), The cost of interrupted work: More speed and stress, Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, CHI 2008, Florence, 107–110, DOI 10.1145/1357054.1357072[1].
Four observations, ours.
It is a conference paper of four pages, which is the normal format in its field and is worth knowing when weighing it against a journal article.
The title contains the finding, and the finding is not the one the paper is usually cited for. More speed and stress, in that order, is an accurate summary of the abstract.
The venue is a computing conference rather than a psychology journal, which explains the framing around system design that appears in the abstract's final sentence.
And it is heavily cited, with citing works appearing in our sources as recently as 2026[1], which is the reason a misreading of it matters.
Two reasons a business publication should care about a computing conference paper, ours. It is the empirical basis for a very large amount of workplace advice, including advice sold to firms as policy.
And the advice is frequently expensive to follow. Meeting-free days, communication policies and focus tooling are real commitments, and a firm making them deserves to know what the underlying study reported.
What It Actually Found
The abstract, in full as we obtained it.
"We performed an empirical study to investigate whether the context of interruptions makes a difference. We found that context does not make a difference but surprisingly, people completed interrupted tasks in less time with no difference in quality. Our data suggests that people compensate for interruptions by working faster, but this comes at a price: experiencing more stress, higher frustration, time pressure and effort."[1]
A bibliographic service supplies a further sentence: "Individual differences exist in the management of interruptions: personality measures of openness to experience and need for personal structure predict disruption costs of interruptions."[2]
Four observations, ours.
The word "surprisingly" is the authors' own, and it tells you they expected the opposite too.
The primary research question was about context, meaning whether an interruption related to your current work is less disruptive than an unrelated one, and the answer was no.
The speed finding is reported as an unexpected secondary result, which is worth registering. It was not what the study set out to test, and unexpected results in four-page conference papers deserve the caution this series applies generally.
And the last sentence is the one that never travels. The effect depends on who you are, and two specific personality measures are named as predicting it.
Faster, Not Slower
The finding stated as plainly as we can, because it is the reverse of the received version. Ours.
Four observations.
People completed interrupted tasks in less time. Not slightly less badly than feared; less time than the uninterrupted comparison.
With no difference in quality, which forecloses the obvious objection that they rushed and produced worse work.
The mechanism the authors propose is compensation. People work faster because they know they are behind, which is an entirely intuitive response and does not require any bias to explain.
And that makes the popular version of this literature not an exaggeration of the finding but an inversion of it, which is a different and more serious category of error than the ones this series usually documents.
Two distinctions worth holding, because the difference matters for how much to worry.
An exaggeration preserves the direction. If a real effect of two minutes is reported as ten, a reader still learns that interruption costs time and overestimates by how much.
An inversion does not. A reader who learns that interruption costs enormous time from a study reporting it saves time has learned something the evidence contradicts, and no amount of discounting for hype recovers the truth.
The Price That Was Paid
What the compensation cost, which is the part worth acting on.
The abstract lists it: "more stress, higher frustration, time pressure and effort."[1]
The paper's own methods section records that these were measured with a standard instrument, the NASA Task Load Index, on twenty-point scales, with a stress measure added to the six standard ratings[3].
Four observations, ours.
These are self-reported ratings rather than physiological or output measures, which is the standard approach and is a real limitation.
But the finding is coherent and is the one a business should attend to. The output held up and the person paid for it, which is precisely the trade a busy firm makes without noticing.
It also explains why the popular inversion is so persistent. Interruption genuinely feels expensive, and the feeling is real even though the time cost the study measured went the other way.
And it is a finding about sustainability rather than about a single day. A day of interrupted work gets done; what the study measured is what it takes out of the person doing it, and nothing in a four-page paper tells you what a year of that does.
Which identifies the study's real limitation, and it is not the one usually raised. The design measures a session, and the concern is cumulative, so the most important question about interrupted work is one this study was not built to answer.
That is not a criticism of the paper, which claims nothing about accumulation. It is a caution about the inference a reader will make anyway, since a single session finding gets read as a claim about working life.
The Moderators Nobody Quotes
The sentence that appears in the abstract and in almost no retelling.
It reports that "personality measures of openness to experience and need for personal structure predict disruption costs of interruptions."[2]
The paper's methods record that openness was measured using a subscale of the NEO Personality Inventory Revised, in a German translation, and that need for structure was measured with ten items from a Personal Need for Structure subscale[3].
Four observations, ours.
We did not obtain the direction or size of these relationships, only that they were found, and we will not guess which way they run.
The existence of moderators means an average effect is a poor guide to any individual, which is the caution the sixty-second article made at length and which applies here directly.
For a business it converts a general finding into a specific question. Some people on your team are far more disrupted by interruption than others, and a blanket policy will be wrong for somebody either way.
And it is a reminder about what gets dropped in transmission. The qualifier sits in the abstract, exactly as the parenthetical in the eighty-third article did, and exactly as rarely survives.
What They Actually Did
The study, from a copy of the paper hosted by the first author at her own university.
Participants worked in a simulated office environment, on a task of writing and responding to a number of emails[5]. The paper records that "Interruption frequency was set to two minutes, based on observations from" an earlier study and pilot experiments, and that "the experimenter adjusted the length of time for interruptions to try to make the times as equal as" possible[3].
Four observations, ours.
This is a laboratory simulation, not a study of real work, and every caution this series has raised about that applies.
The task is email, which is unusually favourable to the speed finding. Email is highly interruptible by nature and resuming it costs little, which is not true of drafting a set of accounts.
Interruptions every two minutes is a severe rate, and the finding that people got faster under it is more surprising for that.
And the generalisation problem runs in a specific direction here. If the speed effect depends on the task being easily resumable, it should not be expected for deep work, and the paper does not claim it would.
Two consequences for a reader deciding what to make of this, ours. The speed finding is the one least likely to generalise, because it plausibly depends on the resumability of email specifically.
And the stress finding is the one most likely to generalise, because there is no obvious reason compensating under pressure would feel better on harder work. That is the reverse of which finding travels, and it is our inference rather than anything the paper says.
The Twenty-Three Minutes
The figure in circulation.
One practitioner page states: "it takes an average of 23 minutes and 15 seconds to fully refocus on a task after an interruption," attributing it to a 2024 work by the same first author[5].
Both sources here are practitioner pages, not research, and we cite them as evidence of circulation only.
Four observations, ours.
We did not obtain any source for this figure and take no position on whether it is correct.
The more careful of our two practitioner sources attributes it to a 2024 work, which is a different and much later output than the 2008 paper.
The precision is worth noticing. Twenty-three minutes and fifteen seconds is stated to the second, which is a level of precision that usually indicates a mean from a specific dataset rather than a rule of thumb.
And a mean of that kind would need a definition. What counts as "fully refocused" determines the number entirely, and we could not obtain the definition.
Attributed To The Wrong Paper
The checkable error, and it is the reason this article exists.
A second practitioner page states: "A University of California study found that after each interruption it takes over 23 minutes to refocus," and then supplies the citation: "Cite: Mark, Gloria & Gudith, Daniela & Klocke, Ulrich. (2008). The cost of interrupted work: More speed and stress."[4]
Again, a practitioner page rather than research.
Four observations, ours.
That is a specific and checkable claim: a number, attributed to a named paper, with a full citation including the DOI.
The abstract of that paper does not contain the figure, and its headline finding is that interrupted tasks were completed in less time.
We cannot say the number appears nowhere in the paper, because we did not obtain its results section, and we state that limit plainly.
But the paper is four pages and its abstract reports the opposite direction, so a finding of that magnitude sitting unmentioned in the abstract would be remarkable. We think the attribution is wrong and we cannot prove it absolutely.
A Source Describing A Different Study
A second error, in the more careful of our two practitioner sources.
It describes the 2008 paper's method as: "Observational study in clinical settings measuring the impact of interruptions on nurses administering medications and doctors making diagnostic decisions."[5]
The paper itself describes participants writing and responding to emails in a simulated office environment[3], with interruptions delivered by instant message and personality inventories administered afterwards.
Four observations, ours.
Those are not the same study. One is a laboratory simulation with student-style participants and an email task; the other is an observational study of clinicians.
A clinical interruption literature does exist, and our own sources reference a study of hospital doctors using prescribing systems[1], so the description appears to have attached to the wrong paper rather than being invented.
What makes this notable is that it is the more careful of the two sources. It correctly reports the speed finding, correctly notes the stress cost, correctly attributes the twenty-three minutes elsewhere, and then describes the wrong methodology.
And it is the kind of error that strengthens a claim rather than weakening it. Nurses and medication errors is a more consequential setting than an email simulation, and a reader would weight it accordingly.
Which is a pattern worth watching for generally, ours. Errors in transmission are not random in direction, and the ones that survive are disproportionately the ones that make a finding more striking.
The Two Numbers Cannot Both Hold
Our own arithmetic, and it requires nothing but division.
The two figures are usually paired. One practitioner source pairs them explicitly, reporting that workers switch tasks every three minutes on average, then noting that if refocusing takes twenty-three minutes, "we never reach full cognitive recovery before the next interruption arrives."[5]
Take that literally. You are interrupted, you need 23 minutes of clear time to recover, and you get 3.
You complete 13.0 percent of the required recovery. Then you are interrupted again, from a state that was never restored.
Four observations.
Recovery is never completed. Not once, in any working day, ever. The steady state is permanent non-recovery.
If both figures measured the same underlying quantity, every knowledge worker would operate permanently at zero focus and nothing would ever be finished.
Things are finished. Accounts get filed, code ships, contracts get drafted. So the two numbers do not measure the same thing, and the pairing is invalid regardless of whether either figure is individually correct.
And the practitioner source drew exactly the wrong conclusion from its own observation. It treated the incompatibility as an alarming finding rather than as a sign the numbers had been misused.
That is worth dwelling on, because the source was one step from the right answer. It performed the calculation, saw the impossible result, and reported the impossibility as the point.
The step it did not take is the one this series takes constantly. When a calculation produces an absurdity, the first hypothesis should be that an input is wrong, not that the world is absurd.
What The Incompatibility Proves
Being precise about what our arithmetic does and does not establish. Ours.
Four observations.
It does not show that either figure is wrong. Both could be accurate measurements of different quantities.
That distinction is worth insisting on, because the temptation here is to conclude too much. An impossible pairing tells you about the pairing and not about either figure, and a reader who now dismisses both has overdrawn on our argument.
It does show that they cannot both be what the pairing assumes: a recovery requirement and an interruption interval, in the same units, on the same clock.
The most likely resolution, and this is our inference and not a finding, is that a figure like the twenty-three minutes measures elapsed time before returning to the original task, during which other work is done. That is a very different quantity from a period of degraded capability.
And under that reading the number stops being alarming and becomes ordinary. Taking twenty minutes to get back to a document because you handled three other things first is a description of a normal morning, not a productivity crisis.
Two tests that would distinguish the readings, ours, if anyone had the data. Was the person idle during the interval, or working on something else? If working, the time is not lost and the framing is wrong.
And was performance on the original task degraded on return, or merely delayed? Those are different claims requiring different measurements, and the popular version asserts the first while a resumption-time figure could only establish the second.
The Charitable Reading
What can be said for the popular version, since we have spent several sections against it. Ours.
Four observations.
The underlying concern is legitimate. Interruption costs something, the paper says so, and the thing it costs is stress and effort rather than nothing.
The 2008 speed finding rests on an email task in a laboratory, and a reader who doubts it generalises to complex work has the paper's own design on their side.
And a figure we could not trace is not a figure we have refuted. It may be perfectly sound in its own source, measuring exactly what its author says it measures.
What we can defend is narrower and still worth saying. The most-cited paper does not support the claim most often attached to it, and the two most-quoted numbers cannot be paired the way they routinely are.
Which is enough to change what a firm should do with the literature. Use it to justify protecting attention and not to quantify what interruption costs, because the quantification is the part that does not hold up.
What Actually Survives
Our reading, stated directly.
Five statements.
Interrupted tasks were completed in less time with no quality difference, in a laboratory email task with interruptions every two minutes.
The cost was stress, frustration, time pressure and effort, self-reported on a standard workload instrument.
The effect varies by person, with two named personality measures predicting disruption costs, and we do not know in which direction.
A widely circulated figure is attributed by at least one source to a paper whose abstract reports the opposite direction, and by a more careful source to a different work entirely.
And none of the above tells you what an interruption costs your business, which is a question with an arithmetic shape and no measured input.
What A Switch Cost Does To A Day
Our own arithmetic, on invented figures, because no source we obtained supplies a switch-cost number. An eight-hour day is 480 minutes.
At 10 switches a day costing 2 minutes each: 20 minutes lost, 4.2 percent of the day.
At 20 switches at 2: 40 minutes, 8.3 percent. At 40 at 2: 80, 16.7. At 20 at 5: 100, 20.8. At 40 at 5: 200, 41.7. At 40 at 10: 400, 83.3 percent.
Four observations.
Forty switches at five minutes costs 41.7 percent of the day, which is a serious figure and one that requires no impossible claim to reach.
It is also nothing like the incompatible pairing above, and that is the point of computing it. A defensible model of the cost produces alarming numbers on its own.
Every figure here is invented. The switch cost in particular is a quantity we could not source, and the table shows a structure rather than a measurement.
And the structure is what matters, because it identifies which variable to attack. The loss is the product of two numbers, and one of them is far easier to change.
Two notes on which one, ours. The switch cost is a property of the work and of the person, and reducing it means changing how tasks are set up or who does them, which is slow.
The number of switches is a property of the schedule, which is changeable this afternoon at no cost. That asymmetry is why the next section is about batching rather than about focus training.
What Batching Saves
The consequence. Ours, same invented figures: 40 interruptions a day, 5 minutes of switch cost each.
Handled as they arrive, in 40 separate switches: 200 minutes lost.
Consolidated into 20 blocks: 100 lost, saving 100 minutes. Into 10: 50 lost, saving 150. Into 5: 25 lost, saving 175. Into 4: 20 lost, saving 180 minutes, which is exactly three hours. Into 2: 10 lost, saving 190.
Four observations.
Batching forty interruptions into four blocks saves three hours a day without answering one fewer message.
The saving comes from the number of switches, not the number of interruptions, which is the distinction the whole exercise turns on.
The returns diminish sharply. Going from 40 to 10 saves 150 minutes; going from 10 to 2 saves only 40 more, so most of the benefit is available from a modest change.
And that matters practically, because the modest change is the achievable one. Checking messages ten times a day is a policy a person might keep; twice a day usually is not.
Two supporting points, ours. A policy that fails is worth less than a weaker policy that holds, and the arithmetic says the weaker one captures most of the benefit anyway.
And the diminishing returns give permission that people rarely grant themselves. You do not need to become unreachable to capture five sixths of the available saving, which removes the main objection to trying.
The Handling Time Never Moves
The part that makes batching a free saving rather than a trade. Ours, invented figures: 40 interruptions at 3 minutes each to actually handle.
In 40 blocks: 120 minutes handling plus 200 switching, 320 minutes total.
In 10 blocks: 120 plus 50, 170 total. In 4: 120 plus 20, 140. In 2: 120 plus 10, 130 minutes.
Four observations.
Handling time is constant at 120 minutes in every row. The same messages get the same attention.
Every minute saved comes from switching less, not doing less, which is why this is not a productivity trade-off in the usual sense.
The cost of batching is real and appears nowhere in this table. It is delay, and a business that answers everything within four hours instead of four minutes has changed something its customers may notice.
And that gives the actual decision, which is not about focus at all. Is your response time a feature you sell, or a habit you acquired? The arithmetic prices the habit and cannot price the feature.
One way to answer that question with evidence rather than instinct, ours. Ask whether any customer has ever mentioned your response speed, in a review, a referral or a renewal conversation.
If several have, it is a feature and the switching cost is a cost of selling it. If none ever has, you have been paying three hours a day for something nobody noticed, which is the more common finding in our experience and is worth checking rather than assuming.
And It Barely Matters What The Number Is
Testing whether our conclusion depends on the figure we invented. Ours.
The saving from reducing 40 switches to 4, at various switch costs:
At 1 minute: 36 minutes a day. At 2: 72. At 3: 108. At 5: 180. At 8: 288. At 10 minutes: 360 minutes a day.
Four observations.
Even at a switch cost of one minute, the saving is over half an hour a day, which is roughly three working weeks a year.
So the recommendation does not depend on the size of the number we could not source, which is fortunate, because we could not source it.
That is a general property worth stealing for other decisions. When an input is unknown, test whether the conclusion survives its plausible range rather than picking a value and defending it.
And if the conclusion flips somewhere in that range, you have learned that the number matters and must be measured. Here it does not flip anywhere, which settles it.
Two further points about that habit, since it is the most transferable thing in this article. It converts a missing number from a blocker into a parameter, and a great many business decisions stall on an input nobody has.
And it tells you when to stop researching. If the answer is the same across the whole plausible range, obtaining the exact figure has no value, and the hours you would have spent finding it are better spent elsewhere.
A Claim We Made Before Printing It
A note on our own process, in the series' usual habit.
We wrote that batching into four blocks saves three hours, and then noticed our table had rows for five blocks and three blocks and not for four. The claim was asserted from a table that did not contain it.
On checking, four blocks saves exactly 180 minutes, which is exactly three hours. The claim was correct.
Three observations, ours.
This is a reporting fault rather than an arithmetic one, and it happened to come out right, which is the least instructive way for it to resolve.
We record it because the check was worth running either way, and a series that documents only the errors that turn out to be wrong is documenting selectively.
And it is the same failure mode as the previous article's second correction. A sentence written before the number was looked at, which appears to be our characteristic error and is now on the record twice.
One structural fix we would draw from that, ours. Print the row before writing the sentence, which sounds trivial and is exactly the discipline both errors bypassed.
Three Errors, One Paper
Collecting what this article found, because the pattern is worth naming. Ours.
Four observations.
A figure the paper's abstract does not contain is attributed to it by citation, complete with authors, venue and DOI, which is the form that makes a claim look checked.
The paper's method is described as a clinical observational study when it was an email task in a simulated office, by a source that got everything else about it right.
And two figures that cannot be paired are paired, with the impossibility noticed by the source doing the pairing and read as alarming rather than as a warning.
What is instructive is that none of these required carelessness about the topic. All three sources are trying to be accurate, cite their work, and have engaged with the material, and the errors are of a kind that reading one abstract would have caught.
One implication we would draw, ours. Citing a source is not the same as having read it, and a full citation with a DOI is the strongest available signal that somebody has, which is precisely why it travels.
Why Inversions Survive
Our reading of the mechanism, offered as reasoning rather than evidence.
Four observations.
The received version is more useful to more people than the finding. A number quantifying lost focus supports a recommendation; a finding that people compensate and pay in stress does not sell software.
It also matches the felt experience. Interruption feels expensive, and a figure confirming that will be adopted more readily than one complicating it, which is the eightieth article's point about which evidence gets checked.
And the inverted version is more actionable, which is not nothing. A reader wants to be told what to do, and "your interruptions cost 23 minutes each" tells them, while the real finding leaves them with a health observation.
Which is why we have built the arithmetic below rather than stopping at the correction. A reader who came for a reason to protect their time deserves to leave with one, and the defensible version supplies it.
And it is worth being explicit that we are not arguing interruption is fine. The defensible case against constant switching is strong enough not to need an inverted finding propping it up, which the next sections demonstrate on arithmetic alone.
The Owner-Operator Problem
Where this lands for a small business. Ours, and not management advice.
Four points.
An owner is the default interrupt handler for their own business. Every question that nobody else can answer arrives at one desk, and the arithmetic above describes that desk.
The batching remedy is harder for an owner than for an employee, because some of the interruptions are customers and some are staff who are blocked, and a blocked employee costs more than the switch does.
Which suggests the useful move is not batching everything but separating the classes. A blocked employee is genuinely urgent; a supplier email almost never is, and treating them identically is what produces forty switches.
And the honest observation is that this is a delegation problem wearing a focus problem's clothes. The number of questions only you can answer is a fact about how the business is set up, not about your calendar.
Two things that reduce the count rather than rescheduling it, ours and unglamorous. Write down the question the next time somebody asks you something only you can answer, and after a month you will have a list of the decisions that are undocumented rather than genuinely judgment calls.
Most such lists are shorter than the owner expects and mostly consist of thresholds. What can I discount, what can I refund, what can I order without asking are answerable once and in writing, and each answer removes a recurring interruption permanently.
The Finding To Take Most Seriously
Returning to what the paper actually reported, because it is the part with the most weight. Ours.
Four observations.
The measured cost was stress, frustration, time pressure and effort, and a business owner reading a productivity literature is usually looking for time and will skip past it.
The trade the study describes is one a firm makes silently. The work still gets done, so nothing appears in any report, and the cost is carried by whoever did it.
Which makes it invisible to exactly the measurement a business would run. Output held; the person absorbed the difference, and output is what gets tracked.
And we would state the boundary this publication always states. Sustained stress at work is a health matter and not a productivity one, and a person who recognises the description here should raise it with a doctor rather than with a scheduling app.
One further note for owners specifically, since the structure of the problem differs. An employee working under sustained interruption has somebody to raise it with, and an owner frequently does not.
That absence is worth naming rather than working around. The people who could tell you the pace is unsustainable are the ones who work for you, and they are the least likely to say so unprompted.
Bibliographic Note
The series keeps a count.
A practitioner newsletter names the paper's second author as "Daniel Gudith"[5] where the paper itself and every other source give Daniela Gudith[3].
Three observations, ours.
A dropped letter in a surname is minor and would probably not defeat a search, since the other two authors and the title are correct.
It sits alongside two substantive errors from practitioner sources in this article, which is a higher density than we usually find, and all three are on pages written to be useful rather than to be checked.
That brings the running count of bibliographic variants across this series to thirty.
Two observations at the round number, ours. Thirty variants across eighty-five articles is roughly one every three, which is higher than we expected when we started counting and lower than the error rate this particular article found.
And the distribution has shifted. The early variants were mostly page numbers in academic reference lists; the recent ones are increasingly on commercial pages, which is where a business reader actually encounters research and where nobody is checking.
What To Do
Do not budget time for lost focus on the basis of the twenty-three minute figure. We could not trace it to a paper, and the study most often cited for it reports interrupted tasks finished faster.
Count switches, not interruptions. On our own arithmetic the loss is switches multiplied by switch cost, and only the first is easily changed.
Batch into about ten blocks rather than two. Going from 40 switches to 10 captures 150 of the 180 minutes available, and is a policy someone might actually keep.
Separate blocked colleagues from everything else. A blocked employee costs more than the switch does; a supplier email does not.
Decide whether fast response is a feature you sell. Batching costs response time, which the arithmetic cannot price and your customers can.
Expect people to differ. The abstract reports two personality measures predicting disruption costs, and a blanket policy will be wrong for someone.
Watch for the cost that does not appear in output. The study found the work got done and the person paid in stress, effort and frustration, which no report will show you.
Check a citation before repeating a number. Two of the three errors documented here would have been caught by reading the abstract of the paper being cited.
The Limits Of This Analysis
Several caveats matter. This article discusses research on work interruption and is not medical, psychological or management advice; the applications are our own reasoning and untested. Everything is verified to August 2026. We did not obtain the 2008 paper's results section, only its abstract from the publisher's digital library, a further abstract sentence from a bibliographic service, and method passages from a copy hosted by the first author; we therefore report no effect sizes from it, including the size of the speed difference, which we describe only as the direction the abstract states. Because we did not read the full paper, we cannot state absolutely that the twenty-three minute figure appears nowhere in it, and we have said so in the body rather than overclaiming. We did not obtain any source for the twenty-three minute figure itself, take no position on whether it is correct, and note that the more careful of our practitioner sources attributes it to a 2024 work we did not see. Two of our five sources are practitioner pages rather than research, flagged at every use; they are evidence about how a finding circulates and about nothing else. The 2008 study is a four-page conference paper reporting a laboratory simulation with an email task, its speed finding was an unexpected secondary result rather than what it set out to test, and its cost measures are self-reported workload ratings. All arithmetic is ours and every figure in it is invented: no source we obtained supplies a switch-cost number, so the tables show a structure rather than a measurement, and we have tested the conclusion across the full plausible range rather than defending any single value. And this article records one claim we asserted before printing the row that supported it, which on checking proved correct.
Frequently Asked Questions
What did the 2008 study actually find?
So interruptions are harmless?
Where does the twenty-three minute figure come from?
Why can the two popular numbers not both be right?
How much does switching actually cost?
Does batching actually help?
Does that conclusion depend on your invented number?
References
- Publisher digital library record for Mark, G., Gudith, D., & Klocke, U. (2008), The cost of interrupted work: More speed and stress, Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, CHI 2008, Florence, 107–110, DOI 10.1145/1357054.1357072, reproducing the abstract: that the authors performed an empirical study to investigate whether the context of interruptions makes a difference; that they found context does not make a difference but that surprisingly people completed interrupted tasks in less time with no difference in quality; and that their data suggests people compensate for interruptions by working faster, but at the price of experiencing more stress, higher frustration, time pressure and effort. The same record carries the paper's reference list and a list of citing works extending to 2026, and separately references a study examining the effect of interruptions and task complexity on prescribing error rates among hospital doctors using computerised order entry systems. Note: the publisher's digital library record. Our source for the abstract. We did not obtain the paper's results section and therefore report no effect sizes from it, and cannot state absolutely that a given figure appears nowhere in the full text. dl.acm.org
- Bibliographic service record for the same paper, reproducing the abstract and supplying a further sentence absent from our other reproduction: that individual differences exist in the management of interruptions, and that personality measures of openness to experience and need for personal structure predict disruption costs of interruptions. Note: a bibliographic service record. Our only source for the personality moderators. We obtained that these relationships were found and NOT their direction or size, and have not guessed at either. semanticscholar.org
- Copy of the same paper hosted by the first author at her own university, reproducing method passages: that interruption frequency was set to two minutes, based on observations from an earlier study and on pilot experiments; that the experimenter adjusted the length of time for interruptions to make the times as equal as possible; that workload was measured using the NASA Task Load Index with a stress measure added to the six given rating scales of mental and physical demands, performance, time pressure, effort and frustration, rated on a twenty-point scale; that openness to experience was measured using the Openness to Actions subscale of a German translation of the NEO Personality Inventory Revised; and that ten items of a Personal Need for Structure subscale were also administered. The same copy gives the second author's name as Daniela Gudith. Note: a copy hosted by the paper's first author at her own institution. Our source for the study's design and instruments. ics.uci.edu
- Practitioner page stating that a University of California study found that after each interruption it takes over 23 minutes to refocus, and supplying as its citation Mark, Gloria, Gudith, Daniela and Klocke, Ulrich (2008), The cost of interrupted work: More speed and stress, Conference on Human Factors in Computing Systems Proceedings, 107–110, DOI 10.1145/1357054.1357072. Note: a practitioner page, NOT research, flagged at every use in this article. Cited solely as evidence that the twenty-three minute claim is attributed to this specific paper, whose abstract reports that interrupted tasks were completed in less time. We could not locate the figure in any part of the paper we obtained. highperformanceroutines.com
- Practitioner pages discussing the same research, one reporting that it takes an average of 23 minutes and 15 seconds to fully refocus on a task after an interruption and attributing this to a 2024 work by the paper's first author rather than to the 2008 paper; reporting that workers switch tasks every three minutes on average, citing a 2014 study; observing that if both figures hold then full cognitive recovery is never reached before the next interruption arrives; correctly summarising the 2008 speed and stress findings; and describing the 2008 study's design as an observational study in clinical settings measuring the impact of interruptions on nurses administering medications and doctors making diagnostic decisions. A second such page describes participants completing a task of writing and responding to a number of emails in a simulated office environment, and names the second author as Daniel Gudith. Note: practitioner pages, NOT research, flagged at every use. Cited as evidence of how the finding circulates. The clinical-settings description does not match the paper's own account of an email task in a simulated office, and the second author's given name is rendered without its final letter. get.convictional.com
This article discusses research on work interruption and is not medical, psychological or management advice. The 2008 paper's results section was not obtained, so no effect sizes are reported and no absolute claim is made about what its full text contains. No source was obtained for the twenty-three minute figure, on whose correctness this article takes no position. Two of the five sources are practitioner pages rather than research and are flagged at every use. All arithmetic is the authors' own and every figure in it is invented, since no source obtained supplies a switch-cost number.