A quiet on-call night should be free. Nobody called. Nothing broke. You slept in your own bed the whole night through, uninterrupted. Two separate sleep-lab studies put that assumption directly to the test, on nights specifically engineered so that no call would come, and both found a real cost anyway.
This matters because on-call cost, in most engineering organizations, gets accounted for entirely by page count. Zero pages this week reads as zero cost this week, which is a reasonable-sounding assumption right up until you actually go looking for evidence that it's true. Two research groups did go looking, independently, using different methods, and neither one found the free night the assumption predicts. Readers who reach on call through a growth or RevOps role will want XenGrowth's work on go-to-market systems alongside this.
The study that split the objective and subjective story apart
Both studies were designed specifically to remove the confound that makes on-call research hard to interpret in the real world: if you compare weeks with actual pages to weeks without, any difference could be the sleep disruption from the page itself, not the state of being on-call in general. By engineering nights where the phone simply never rings, both research teams isolated the one variable this post is actually about — reachability itself, with the interruption surgically removed from the picture.
Ziebertz and colleagues ran 96 students through two consecutive nights: an ordinary reference night, and a night where participants were told they could be called at any point and would need to complete about 30 minutes of online tasks if they were. No one actually got called. The design isolates exactly one variable: the state of being reachable, with the interruption itself removed from the equation.
Actigraphy — a wrist-worn accelerometer used to objectively estimate sleep — found no significant within-person difference between the two nights. If you stopped reading there, you'd conclude on-call with no actual calls is a non-event. The self-report data says otherwise: on the on-call night, participants reported longer sleep onset latency, more awakenings, and more time spent awake after initially falling asleep.
Measure | Reference night | On-call night (no call) |
|---|---|---|
Objective sleep (actigraphy) | Baseline | No significant within-person difference |
Self-reported sleep onset latency | Baseline | Longer |
Self-reported awakenings | Baseline | More |
Self-reported wake after sleep onset | Baseline | More |
That's a genuinely awkward pair of findings to hold at once, and the honest reading isn't 'the subjective data is wrong' or 'the objective data is wrong.' It's that actigraphy measures gross movement patterns, not the felt experience of lying awake anticipating a possible call, and the two don't have to move together. A night can look identical on a wrist sensor and feel measurably worse to the person who lived through it. The XenGrowth resource library works through the operations side of this in more operational detail.
The study that measured what waking up felt like
Sprajcer and colleagues took the question into a fully controlled sleep laboratory. Twenty-four men spent an adaptation night, a control night, and two on-call nights — on one, they were told they'd have to give a speech immediately on waking (high-stress); on the other, they were told they'd just read silently (low-stress). Again: nobody was actually woken by a real page. The manipulation was purely what participants believed might be asked of them.
Compared to the control night, participants felt significantly sleepier upon waking during both on-call nights. The detail that should reframe how you think about a 'low-severity' rotation: there was no significant difference in that sleepiness between the high-stress and low-stress framing. Believing you might have to give a speech didn't cost meaningfully more than believing you might have to read quietly. Simply being on-call did the damage; what you were on-call for barely moved the number.
Anticipating a high-stress task, versus a low-stress one, did not make waking up feel significantly worse. Being on the hook for anything at all was already most of the cost.
One further wrinkle worth naming honestly: spatial performance in the same study was actually faster when participants were on-call than during the control condition. Not every measure in this line of research points the same direction — some vigilance-linked measures can improve under mild anticipatory arousal even as subjective sleep quality worsens. The evidence supports a real cost; it doesn't support treating every possible metric as uniformly worse under on-call conditions.
Putting the two studies side by side
The two studies used different populations, different environments, and different measurement approaches, which makes their agreement more meaningful than either result alone. Ziebertz's field study used real students in their own beds with real actigraphy watches; Sprajcer's used a fully controlled sleep laboratory with a manipulated cover story about what waking would require. Neither design is a stand-in for the other, and neither is a perfect model of a real engineer's on-call week — but both, independently, found being reachable degrading something measurable about the night, even with zero actual calls. On AI agents and marketing automation specifically, XenGrowth on AI agents and marketing automation is worth reading.
Dimension | Ziebertz et al. (2017) | Sprajcer et al. (2018) |
|---|---|---|
Setting | Field study, participants' own beds | Controlled sleep laboratory, time-isolated |
Sample | 96 students | 24 men, aged 20-35 |
What varied | Reference night vs. simulated on-call night (no call) | Control night vs. two on-call nights (high- vs. low-stress framing) |
Objective measure | Actigraphy — no significant difference found | Sleep-lab measures of sleep inertia and spatial performance |
Subjective measure | Self-reported sleep onset latency, awakenings, wake-after-sleep-onset — all worse on-call | Self-reported sleepiness on waking — worse on both on-call nights vs. control |
Did task severity matter? | Not directly tested | No — high-stress and low-stress on-call nights did not differ significantly |
Read across that table, the two studies converge on something neither one could establish alone: it's not the objective sleep architecture that reliably breaks under on-call anticipation, and it's not the severity of what you're on-call for. It's the subjective experience of the night and the state of waking up, apparently driven by the mere fact of being reachable rather than by anything that actually happened.
Why 'nothing happened, so it was fine' is the wrong test
It's worth connecting this to a broader pattern in sleep research: Van Dongen and colleagues' randomized trial of restricted time in bed found something structurally similar, in a different setting entirely. People whose sleep was chronically restricted for 14 days reached measurable impairment on attention and working memory — and their own subjective sleepiness ratings did not track that decline. They did not feel as impaired as they had become. That mismatch between how something feels and what it's actually doing to you shows up again here, just through a different mechanism: on-call anticipation rather than restricted time in bed.
The practical upshot is the same in both cases. 'I feel fine' and 'nothing happened' are not reliable evidence that a quiet on-call rotation was free. Neither study found a dramatic, obviously visible cost — nobody described these nights as catastrophic. What both found was a real, measurable, and consistent one, on nights specifically designed to have nothing else going wrong.
Why this is easy to dismiss, and why that's a mistake
The natural objection is that these are lab and short-field studies, one to two nights long, and a real on-call rotation runs for a week or more at a stretch. That's a fair limitation to name honestly — neither study followed anyone through a sustained multi-week rotation, and it's possible the effect habituates with repeated exposure, or alternatively compounds. Nobody has run the longer version of this experiment cleanly enough to say which. What the existing data does rule out is the comfortable assumption that a single quiet on-call night is a neutral event by default, which is the assumption most rotation scheduling quietly makes. Whatever happens over a longer stretch, it isn't starting from zero cost on night one. XenGrowth on AI search, GEO and discovery covers the AI search, GEO and discovery side of this.
There's also a selection problem worth naming in how these findings tend to get received inside engineering organizations specifically. On-call scheduling is usually optimized around a single, easily measured variable: how many people are available to be paged, and how evenly the burden is split by headcount. Sleep quality on the non-paged nights isn't in that spreadsheet at all, because nobody who built the rotation was thinking about it as a real cost — not out of carelessness, but because it doesn't show up anywhere the way a page count does. The two studies in this post are useful specifically because they measured the thing the spreadsheet was never built to see.
What this means for how rotations get designed
Stop counting on-call cost by pages received. Both studies found a real cost with zero actual calls — the state of being reachable is the variable, not the interruption count
Don't assume a 'low-severity' rotation is close to free. Sprajcer's data found no significant difference in cost between high- and low-stress anticipated tasks — the mere obligation did most of the damage
Recognize that self-reports and 'objective-looking' data can genuinely diverge, and that doesn't mean one of them is wrong. Ziebertz's actigraphy and self-report told different stories about the same nights, and both were real measurements of something
Rotate on-call duty with the assumption that quiet weeks still carry a cost, rather than crediting them as free recovery time in scheduling math
If someone says a quiet on-call week 'was totally fine,' treat that the way Van Dongen's research suggests you should treat any self-report of feeling unaffected: informative, but not proof, especially over a sustained rotation rather than a single night in a lab
None of this argues that on-call rotations are inherently unworkable — plenty of systems genuinely need someone reachable. It argues against treating a quiet night as a free one when it comes to planning how often someone should carry that reachability. The sleep-lab data says the phone doesn't have to ring for the night to already have a price on it.
If a rotation schedule is currently built entirely around minimizing pages, it's worth adding a second, much simpler question alongside it: how many nights per month does any given person spend reachable at all, regardless of whether anything ever comes through. That second number is the one both studies suggest is actually driving the cost, and it's the one almost nobody currently tracks. It costs nothing to start tracking it, and it's a better predictor of who's quietly burning out on a rotation than the page count ever was.
Further reading from XenGrowth
The XenGrowth resource library — what you'll learn: how the commercial side of this work is run, across search, automation and revenue operations.
XenGrowth on AI agents and marketing automation — what you'll learn: how the teams who own AI agents and marketing automation plan and measure it.
XenGrowth on AI search, GEO and discovery — what you'll learn: how the teams who own AI search, GEO and discovery plan and measure it.
Where this work meets go-to-market
For the marketing and revenue operations view of on call, see XenGrowth's growth operations team.
Four questions on the two studies this post is built on. Nobody in either study was actually called during the night in question — that's the whole point.
