Switching between tasks: what the studies measure about the cost, and how a crew lead batches work

A cook is halfway through a ticket when the server asks about table six. A foreman is checking a delivery when someone wants to know where the second pallet goes. Each question takes twenty seconds. Most leads would tell you those twenty seconds cost far more than twenty seconds, and the internet has numbers ready: every switch costs 23 minutes, and multitasking eats 40% of your day.

Those two numbers don't say what people think they say. The first is not a recovery time. The second is a remark, and a sum shows when it could hold. The studies behind them do find real costs, but mostly not in lost minutes: they show up as lost places, wrong steps, more stress, and jobs that sit open for longer. Those are the costs a lead can do something about.

Here is what each study measured, with the figures, and then what they suggest for a ticket rail or a site. No study we found tested a kitchen rail or a foreman's day, so that last part is our reading of the evidence, and we say so where it matters.

What the quoted numbers measure

The lab cost: about a second a switch

Joshua Rubinstein, David Meyer and Jeffrey Evans ran four experiments, with 12 to 36 young adults in each according to the American Psychological Association's release. People either did one task over and over, or alternated between two: sorting geometric shapes by rule, or solving arithmetic problems. The switch cost is the extra time per switch when alternating.

The averages hide two findings that matter more. Harder rules cost more: in Experiment 1 the switch cost was 545 and 582 ms with simple rules, and 1,386 ms with complex ones. A visible cue seems to cut the cost: in Experiment 2 showing a task cue saved 467 ms on average (a noisy estimate, at the edge of significance), and switching between addition and subtraction with the signs shown cost only 63 ms.

Now do the sum for a working day. A hundred switches at 1.1 seconds each is under two minutes. So the lab cost, taken alone, doesn't come close to explaining a bad shift.

Try it the other way round. 40% of an eight-hour day is 192 minutes, or 11,520 seconds. At 1.096 seconds a switch, that takes about 10,500 switches: one every 2.7 seconds, all day. That is roughly what the people in the lab were doing. In Experiment 1, switching took 26% to 45% of the time per card in the alternating blocks (545 of 2,094 ms, 582 of 1,830, 1,386 of 3,080; our arithmetic, not the authors'). So a figure near 40% can hold when you switch on nearly every move. It is not a measure of a normal working day, and the figure does not appear as a result in the paper.

23 minutes: a gap, not a recovery time

This number is usually told as "it takes 23 minutes to get your focus back". Here is what Gloria Mark and her colleagues reported.

Four things follow. It's a gap, not a recovery. They timed how long before someone returned to the interrupted work, and in that time people were working on about two other things. The spread is huge. A standard deviation more than twice the mean means a few very long gaps pull the average up, so a typical gap is probably much shorter (our inference; the paper gives no median). The 23:15 figure is from an interview. We searched the 2005 paper and the 2008 paper often credited for it, and neither gives that number. They watched desk workers, not a line or a site.

The same paper has a number worth keeping: these workers stayed about 11 minutes in one area of work before switching or being interrupted, and 57.1% of those stretches were interrupted. And in 90.1% of the same-day returns, people went back on their own, without anyone else prompting them. So going back to a job depends largely on someone remembering it.

The surprise: interrupted people worked faster

In 2008 Mark, Daniela Gudith and Ulrich Klocke gave 48 people, mostly German students, batches of emails to answer in a role-play job. In two conditions they were interrupted every two minutes by phone or instant message. The time below leaves out the time spent on the interruptions themselves.

Interrupted people finished about 10% faster with no more errors, and rated their stress about a third higher (6.92 to 9.46 is 37%; to 9.13 is 32%). This was after only about 20 minutes. The authors' reading is that people make up for interruptions by working faster, and pay for it in stress. Over one task that looks like a good deal. We don't know from this study what it does over a ten-hour double.

Where it really costs: losing your place

Erik Altmann, Greg Trafton and David Hambrick gave 300 students a task with seven steps in a fixed order, then interrupted them briefly with a small typing task.

Other kinds of mistake did not rise, and the effect was limited to the first move after the interruption. So the damage was specific: people did the step correctly, but it was the wrong step. A three-second interruption is enough to lose your place.

The nearest thing to crew work we found is a hospital ward. Johanna Westbrook's team followed 98 nurses on six wards in Sydney through 4,271 medication administrations, over 505 hours. Over half of them (53.1%) were interrupted.

Each interruption went with 12.7% more clinical errors, and experience did not protect people: experienced nurses were no safer from clinical errors and had more procedural failures. It's an observational study, so a busy ward may produce both interruptions and errors. But the job looks a lot like checking an allergen order or a lockout: a fixed sequence, done by hand, interrupted in person.

Attention residue: finishing helps, parking helps too

Sophie Leroy (2009) asked why people do badly on the task after a switch. In two studies, with 84 and 78 students, people did a word task and then appraised a CV. Those who had finished the first task under time pressure showed fewer signs of still thinking about it, and did better on the CV than those who hadn't finished or who finished without time pressure. Leroy called the leftover attention attention residue. We looked at this from the angle of the finished task in Why finished work feels lighter.

A crew lead can't always finish first, which makes the follow-up more useful. Leroy and Theresa Glomb (2018, four studies) found that expecting to come back to an interrupted task under time pressure hurt performance on the interrupting job. A brief ready-to-resume plan, noting where you were and what you will do next, reduced the residue and removed that drop.

The cost nobody counts: too many open jobs

This cost needs no psychology at all. Suppose one person has five one-hour jobs, and switching is free. Do them one after another, or give each 15 minutes in turn:

The work takes five hours either way, but the average job finishes 50% later when you rotate, and that is before adding any switch cost. A customer waiting on job A, or the next trade waiting on it, feels that delay.

Italian judges supply the real-world test. Cases are assigned by lottery, so judges have similar workloads. Decio Coviello, Andrea Ichino and Nicola Persico found that judges in Milan who ran more trials at once took longer to finish them, with no gain in quality as measured by appeals: by a later paper's summary, 1% more juggling meant 2% longer cases. Then Robert Bray and the same authors persuaded six judges in Rome to schedule hearings for their oldest open case first. Against 44 other judges over eight years, they closed 11% more cases a day, cases took 111 days (12%) less, and appeals fell 3.8%. Six judges chosen by persuasion is a small group, though the authors' placebo checks support the result.

One caution against taking this too far. Bradley Staats and Francesca Gino used two and a half years of data from a Japanese bank's home-loan line: focus on one kind of work within a day raised productivity, but variety across days raised it too. So batch the day, not the whole career. People need some change, and the same line of research reports waiters more focused with more tables to look after (Tan and Netessine, 2014, as summarised by Bray and colleagues).

Does batching work? Some evidence, mostly on email

Direct tests of batching mostly concern email, so treat them as a neighbour of crew work, not the same thing.

Kushlev and Dunn (2015): 124 adults spent a week checking email three times a day and a week checking as often as they liked. Actual checks were 4.70 against 12.54 a day. They received and answered about the same number of emails, and daily stress was lower (d = 0.37), a small to medium effect. Their stress recalled at the end of the week didn't change, and neither did most other measures.
Mark and colleagues (2016): 40 office workers tracked with logging software and biosensors for 12 working days. Batchers rated themselves more productive on long email days, but there was no evidence of less stress.

So: fewer check-ins didn't lose any work in the first study, and the stress benefit is modest and not guaranteed.

The cost of a switch is rarely the second it takes. It is the wrong step after it, and the job left open behind it.

How a crew lead batches work

None of these steps has been trialled on a crew. Each one is our reading of the studies above, and the tally in step 1 will tell you whether it is worth doing on yours.

1. Count one shift. Keep a tally of every time someone stops you, and mark each as one of three kinds: it couldn't wait, it could have waited for the next check-in, or the answer should have been written down already. Here is an example with made-up numbers for a lead with a crew of five, on an eight-hour shift:

The questions still have to be answered, so this doesn't save all that time. What it changes is how long the lead goes between stops (480 minutes divided by 36, then by 9). Count your own numbers before you believe ours.

2. Write down what never waits. Injury, gas, allergens, a customer at the counter, anything unsafe. Everyone interrupts for those, every time. Batching only works if nobody hesitates on this list.
3. Set the check-ins. Three a shift is a reasonable start, since that is what the email study used. Keep them short; our ten-minute pre-shift huddle can be the first.
4. One current job each. One is the aim and two the limit, as in our task board for a five-person team. The job's name goes on the board, so nobody has to ask.
5. Finish the nearly done before starting the new. This is the idea behind the judges' rule, and it clears the small jobs that sit at 90%.
6. Park it in one line. Before you leave a job: where you stopped and the next step. This is the ready-to-resume plan, and it also gets you back, since mostly nothing else will prompt you.
7. Group like with like, and change it up across days. Same kind of work together in a day, different kinds on different days.

At the ticket rail

A line cook with five pans going isn't switching in the lab sense all the time, since a pan cooks without attention. The switching happens when the cook's eyes and hands are on one plate and the next demand arrives. The rail already does what the cues did in the lab: it shows what comes next, without anyone having to hold it in their head. So keep questions to the expeditor, not the cooks, and let tickets be the single channel. Protect the sequences where a wrong step does harm: an allergen plate is plated by one person, start to finish, with nobody talking to them. If they are interrupted anyway, go back one step instead of carrying on. Group prep by station before service, and leave the side jobs that can wait for the after-rush check-in.

On a site

A foreman is often interrupted by questions someone could answer with the drawing in hand. Morning huddle, one check before lunch, one mid-afternoon, and everything else that can wait goes on the job as a comment, with a photo if it's something you can see. A comment doesn't ping anyone's phone, so anything urgent stays a shout or a call, which is the right way round. Do all the material runs on one trip. Start a task only when the last one is finished or properly parked, and treat lockouts, lifts and anything with a fixed order like the allergen plate. If a job has to stop at day's end, the note is what the next person picks up; our shift handover piece has the rest of that routine.

What the research does not show

It doesn't show a fixed cost per switch. The lab figures run from under 100 ms to over a second depending on the tasks, and they come from young adults doing simple rule tasks.
It doesn't show that an interruption takes 23 minutes to recover from. No study here measured that.
It doesn't show 40% lost in real work. We could not find a study measuring that on a working day.
It doesn't test kitchens or building sites. The closest workplace evidence is on nurses, office workers, judges and bank staff.
It doesn't prove interruptions cause the errors on the ward. That study observed them together; the lab work points the same way but with different tasks and people.
It doesn't show batching always helps. On email the two studies differ on stress, and a crew's questions can carry information a lead needs now.
It doesn't say that all switching is bad. The bank data favours variety across days. Some questions save the job.

Sources

Rubinstein, J. S., Meyer, D. E. & Evans, J. E. (2001). Executive control of cognitive processes in task switching. Journal of Experimental Psychology: Human Perception and Performance 27(4), 763-797. doi:10.1037/0096-1523.27.4.763
American Psychological Association (2001). Is multitasking more efficient? News release, with the sample sizes.
American Psychological Association. Multitasking: switching costs. The 40 percent remark.
Mark, G., Gonzalez, V. M. & Harris, J. (2005). No task left behind? Examining the nature of fragmented work. CHI 2005, 321-330. PDF
Robison, J. (2006). Too many interruptions at work? Interview with Gloria Mark, Gallup Business Journal, June 8, 2006.
Mark, G., Gudith, D. & Klocke, U. (2008). The cost of interrupted work: more speed and stress. CHI 2008, 107-110. PDF
Altmann, E. M., Trafton, J. G. & Hambrick, D. Z. (2014). Momentary interruptions can derail the train of thought. Journal of Experimental Psychology: General 143(1), 215-226. PDF
Westbrook, J. I., Woods, A., Rob, M. I., Dunsmuir, W. T. M. & Day, R. O. (2010). Association of interruptions with an increased risk and severity of medication administration errors. Archives of Internal Medicine 170(8), 683-690. doi:10.1001/archinternmed.2010.65
Leroy, S. (2009). Why is it so hard to do my work? The challenge of attention residue when switching between work tasks. Organizational Behavior and Human Decision Processes 109(2), 168-181. doi:10.1016/j.obhdp.2009.04.002. Study details as reported by the British Psychological Society's Research Digest.
Leroy, S. & Glomb, T. M. (2018). Tasks interrupted: how anticipating time pressure on resumption of an interrupted task causes attention residue and low performance on interrupting tasks and how a "ready-to-resume" plan mitigates the effects. Organization Science 29(3), 380-397. doi:10.1287/orsc.2017.1184
Coviello, D., Ichino, A. & Persico, N. (2015). The inefficiency of worker time use. Journal of the European Economic Association 13(5), 906-947. doi:10.1111/jeea.12129
Bray, R. L., Coviello, D., Ichino, A. & Persico, N. (2016). Multitasking, multiarmed bandits, and the Italian judiciary. Manufacturing & Service Operations Management 18(4), 545-558. PDF
Staats, B. R. & Gino, F. (2012). Specialization and variety in repetitive tasks: evidence from a Japanese bank. Management Science 58(6), 1141-1159. doi:10.1287/mnsc.1110.1482
Tan, T. F. & Netessine, S. (2014). When does the devil make work? An empirical study of the impact of workload on worker productivity. Management Science 60(6), 1574-1593. doi:10.1287/mnsc.2014.1950
Kushlev, K. & Dunn, E. W. (2015). Checking email less frequently reduces stress. Computers in Human Behavior 43, 220-228. PDF
Mark, G., Iqbal, S. T., Czerwinski, M., Johns, P. & Sano, A. (2016). Email duration, batching and self-interruption: patterns of email use on productivity and stress. CHI 2016, 1717-1728. doi:10.1145/2858036.2858262