Summary or TL;DR – Since my kids were born, I have used time tracking as a two-week diagnostic exercise to identify areas where I could be more efficient with my time. Having had a brush with burnout, I decided to track a whole 18 months. This first post covers the background of the exercise, some of the early findings and, roughly, what 3,200 working hours achieved.
My grandfather had a saying: “You can be a good person, a good husband, or a good lawyer. You don’t have the time to be all three.” As I didn’t like to write, I became a scientist (oh, the irony!), and I have spent years trying to prove my grandfather wrong. For a long time I had no idea whether that was a reasonable ambition or just a slow way of losing. So I started counting.
Why I started counting
I began tracking my time after my first son was born. Until then I’d been the person who stayed in the lab for 12 to 14 hours a day. As a dad, that stopped being an option. So, I needed to cram those 12 hours into 8(ish).
The exercise was always the same. Run a time tracker for a couple of weeks, log everything, look at the result, optimise. I repeated it periodically, never on a schedule. Each time it was triggered by the same feeling: that I wasn’t handling my duties, or my research, as efficiently as I felt I should be.
It did change things. After one cycle in 2016, I realised that seeing two team members a day was more disruptive to my work than seeing all of them on the same day. I moved to a single team member day. Since 2020 (as a result of another cycle), I shifted to dividing the team across 2 days. Small changes, but always with noticeable impact.
Where the log outperforms the memory
My personal feeling is that scientists overestimate their useful hours in the lab. However, I cannot find any data to back that up, and in fact, the general evidence is a lot more varied. The only thing I could find evidence for is that memory is a poor accountant of working time. In one study, students’ retrospective estimates of time spent on tasks spread over several days overshot in 78% of cases, at a median of 1.45 hours reported for every hour actually spent [1]. Self-reports deviate most from diaries for people with long hours and flexible schedules [2], and fragmented work makes people less accurate at judging how long a task took [3]. The error doesn’t always run in one direction, either: German teachers’ daily diaries added about two hours a week compared with a single estimate [4].
So the safe summary is that estimates go wrong in long, flexible, fragmented work, which is the very nature of academic work.
A tool for learning, not for supervision
I recommend Toggl openly to colleagues and students. Not as a supervisory tool, but as a learning tool for the person doing the tracking. Done properly, which means logging every break however short, even two weeks is very informative. Two things show up quickly:
Avoidance. The gap before you start the hard task, the email you check instead, the item that gets pushed to Thursday every week. Once it’s on the screen it’s hard to call it anything else, and then you can do something about it. That opens a whole path towards efficiency.
Switching. When the blocks get shorter and shorter, it’s also time to look at it in detail. Frequent swapping between tasks can raise cognitive load: interruptions raised effort and worsened emotional state even when performance held [5], and a review of 247 publications maps how and when this happens [6]. The evidence isn’t unanimous. One randomised study of 152 knowledge workers found that the task-switching group reported less depletion and higher motivation [7]. A log won’t settle that argument, but it lets you see your own switching rate and judge what it does to you. In a small exploratory study, graduate students who logged their working hours began to consider breaks and emotions, not only output [8].
When two weeks stopped being enough
Twice in recent years I brushed against burnout. Each time I ran the familiar two-week log, and each time it told me nothing I could act on. Two weeks is long enough to catch a habit, but it hadn’t been long enough to catch a pattern.
So I went longer. I planned a year, and extended it to 18 months so that I could introduce changes and measure their impact. The changes turned out to be simple: more science (an activity that for me generates energy rather than simply taking it).
A note on what the log counts. It records task time only: no breaks, no lunch, no movement between sites, no commute. Every number below is therefore a floor.
Eighteen months, counted
Between January 2025 and July 2026 I logged roughly 3,200 hours. Five things stood out:
The week. In a normal working week, after netting out the days I wasn’t supposed to work, I logged 43.5 hours against a 38-hour contract, 14% over. Measured from first task to last task, the week stretches to about 50 hours.

Logged time versus time at work, in hours per week. January 2025 to July 2026. The grey part of each bar is logged task time; the full bar height is the span from first to last task – so time at work. The line marks the 38-hour contract.
The days off. There were 202 days that were nominally off: weekends, declared leave, university closures. I worked on 76 of them (a total of 290.7 hours). That’s 9.1% of everything I logged. Nearly one working hour in ten fell on a day I wasn’t meant to be working.
Where it goes. Institutional administration took 31% of my logged time (in fairness I was a member of the Faculty Board and the vice-dean of Internationalization at the time), group supervision 21%, teaching and assessment 20%. My own research took 4.5%! I count the time I spend writing up and remediating students’ manuscripts separately (another 3.4%), because it isn’t research output.
The shape of a day. The median block of work lasted 43 minutes. On a median day I had 7 blocks across 4 different projects, and 46% of days touched five or more projects. (Yes, I put all of this into AI to help make sense of it).
Deadlines decide. About 64% of my hours went on work with a hard external deadline: teaching, committees, grants, refereeing, email. Only 11% went on work with no external deadline at all: my own research, my own writing, my own learning. The two compete. In the weeks with the most deadline-driven work, only 5% of my time was deadline-free; in the lightest weeks it was 22%, a 4.4-fold difference. Across 74 working weeks the correlation between the two was r = −0.33, and 22% of weeks contained no deadline-free work at all. That is what the long window bought me: a pattern that only shows across many weeks, at the weekly scale. Research in Belgium has one hard deadline a year, the grant season, and is otherwise “infinitely” “postponable”. The obvious caveat with this is that the job I was hired for (BOF-ZAP – Research-focused group leader) was the one being pushed aside for the deadlines.
More science, more energy.
In January to March 2025 (so, Q1) I logged 12.3 hours of my own scholarly work. In the same months of 2026 it was 48.9 hours. This does not include the time spent on manuscript revision (another 50 h in Q1/26) – not as rewarding as doing the science, but often still better than purely administrative tasks. This extra science time got clawed from education (new course was developed for Q1/25 and only needed minor adjustments for Q1/26).
Nonetheless, if I look at the 6 months the data overlap, the extra science didn’t always come from clawing back time from other activities. I took fewer leaves (3 days of holiday in Q1+Q2/26) and worked 25 weekend days. Despite the extra hours, I feel more energised.

Hours per category, H1/25 versus H1/26. Hours logged in the same window (6 January to 16 July) of 2025 and 2026.
What this is, and isn’t.
Because I am discussing workload (and will continue for at least 2 more posts), I also need to add some framing here. This blog is not a complaint. It’s one person, at one institution, in one country, at one career stage, in one discipline. Time tracking (and subsequent analysis) changes behaviour, and 18 months of logging is a selective habit. The classifications rest on my own tagging.
I’m also not offering this as a complaint about any institution. The accounting problem it points to is documented internationally. Miller followed a single Australian academic for three months against a nominal 40:40:20 workload (teaching : research : admin+service) model and found something closer to 40:25:35 [9]. I’m one more data point, just a longer one on a 22:38:40 workload (according to the buckets defined in [9]).
So was my grandfather right?
This is a third of the data needed to answer that. But from these data, I have learned how many hours the attempt cost, and where those hours went. I also learned that I need to do more science.
References
- Hart et al. (2025) Time-on-task estimation for tasks lasting hours spread over multiple days. PLOS One.
- Otterbach et al. (2010) How Accurate are German Work-time Data? A Comparison of Time-diary Reports and Stylized Estimates. Social Indicators Research.
- Barrero et al. (2008) Work pattern causes bias in self-reported activity duration. Occupational and Environmental Medicine.
- Kreuzfeld et al. (2022) Teachers’ working time as a risk factor for their mental health. BMC Public Health.
- Zijlstra et al. (1999) Temporal factors in mental work: Effects of interrupted activities. Journal of Occupational and Organizational Psychology.
- Puranik et al. (2019) Pardon the Interruption: An Integrative Review and Future Research Agenda for Research on Work Interruptions. Journal of Management.
- Nuhn et al. (2025) Effects of Task Switching on Depletion, Motivation, and Creativity. Journal of Personnel Psychology.
- Chow et al. (2023) Feeling Stressed and Unproductive? A Field Evaluation of a Therapy-Inspired Digital Intervention for Knowledge Workers. ACM Transactions on Computer-Human Interaction.
- Miller (2019) Where does the time go? An academic workload case study at an Australian university. Journal of Higher Education Policy and Management.
Leave a Reply