Reduce Burnout in Engineering On-Call Without Raising Risk

Question

Engineering on-call work should protect systems without exhausting the people who support them. Experts in the field share practical ways to reduce burnout, from clearer responder roles to smarter incident paging. These strategies help teams focus on real risks, pause unsafe releases, and move lower-priority notices out of the overnight queue.

Teri Maltais · Answer

I recommend splitting on-call duties into a primary responder and a backup, with alerts routed to the primary first and the backup only engaged after a set time or when severity reaches a predefined threshold. The single change I made was enforcing that primary/backups have distinct responsibilities so only one person needs to be on high alert at a time. We also formalized a protected recovery period after serious night incidents so the responder can recover without undermining coverage. Clear escalation rules and severity thresholds keep coverage safe while reducing continuous high-alert fatigue.

Amit Agrawal · Answer

Burnout caused by on-call trainings can rarely be attributed to the commitment of the engineers; it can rather be considered a fault in the way we treat people's attention as something we can use purely as needed. During my experience of over twenty years in charge of delivery teams globally, I saw that the greatest threat to reliability is not the slow response, but rather the tired one. Therefore, the change we implemented in order not to lose speed was the shift to a worldwide dispatch of the teams instead of the local rotation. With the help of our teams in different time zones in Europe, India, and the USA, we can ensure that the main responder is almost always working during the times when he/she is usually working.

As for the most effective change I made to minimize the problem of overhead calls, it was the introduction of automated filtering and alert suppression system depending on impact threshold related to the end-users. Earlier our engineers used to receive alerts on all technical problems, including minor ones, leading over time to alert blindness - the problem of people being unable to focus on the messages coming through the system. We changed the progress monitoring process in such a way that apart from the most significant disruptions receiving alerts at a midnight is impossible since any event that occurred and did not have any impact on the end-users is logged until morning instead.

Steven Mitts · Answer

Reducing on-call fatigue starts with improving the quality of alerts, not simply adding more coverage.
In my companies, the biggest operational risk is not usually that something breaks. It is that the team gets trained to ignore noisy systems. Once people stop trusting the alert layer, reliability is already compromised.
The single change I made was simple: we stopped paging people for "interesting" problems and only paged for actionable, customer-impacting issues. That discipline came from building and scaling entrepreneurial ventures, where small teams cannot afford to waste attention on noise. In a startup, every alert has an opportunity cost: it pulls someone away from product, customers, sales, or the next critical decision.
If an alert could not clearly answer three questions — what is broken, who is affected, and what action needs to happen now — it came out of the overnight on-call path and moved into a daytime review queue. That one shift reduced fatigue because engineers were no longer waking up for warnings, duplicate notifications, or issues that could safely wait. It also made coverage safer because the truly urgent alerts became easier to see.
My rule for founders is: don't solve alert fatigue by asking tired people to be tougher. Solve it by designing a smarter operating system. Better thresholds, cleaner ownership, automated triage, and fewer false positives protect your team and your customers better than brute-force coverage ever will.
— Steven Mitts, Founder & CEO, Steven Mitts Services

Hasan Can Soygök · Answer

I changed our process to prioritize stability with a simple error-budget gate: when a release is spending trust faster than it adds value, we halt new feature rollouts and focus on fixes. We use customer-visible incident spikes and change failure rate as the signals that trigger the gate. That single change reduced on-call churn because engineers stopped chasing frequent, recent-release incidents and could focus on scheduled remediation. It kept coverage safe by preventing risky changes from reaching more users and by making alerts more meaningful so the on-call team could concentrate on real reliability issues.

Kyle Barnholt · Answer

The alert threshold shifted toward sustained conditions that genuinely affected customer experience across services. The change looked simple but it demanded careful discipline across every operational team daily. Many teams confuse visibility with urgency and create pages for events needing observation instead. Meaningful alerts should support timely action without causing unnecessary interruptions during routine operations consistently.

Before adjusting the threshold every incident received careful review with one guiding question first. Did each alert lead to a decision that prevented meaningful customer harm before escalation. Signals without clear value were grouped with related activity or moved into quieter monitoring. The result created fewer interruptions stronger trust and faster responses whenever important alerts appeared.

Ranjith Raghunath · Answer

The approach that has helped us manage this work best is making it a natural step in the engineer career cycle. Brand-new hires generally need more understanding of our systems and practices before they can handle on-call work, but after they've been with us for a year or two and are looking for the next step on the ladder, on-call work is perfect. It requires more commitment, great communications skills, and flexibility, all of which are useful skills for team leads or even eventual CTOs.

Victor Smushkevich · Answer

Page only when a human has to do something right now. Everything else waits for a morning digest. That's the single change I'd defend, and it goes straight at fatigue, because broken sleep is what wears people down on a rotation.

The pattern I keep running into on small teams is an alert list that grew one incident at a time. Each alert made sense the day someone added it. Nobody went back and asked whether it still needed a person awake. So the pager fires on a queue that's slow but healing itself, and whoever's on call learns to distrust it.

Coverage stays safe if you sort every alert by one question: what does the person do when it goes off? If the answer is nothing until morning, it moves to a daily digest someone reads over coffee. If the answer is restart this or roll that back, script it and let the machine try first, then page only when the fix fails. The alerts that touch users directly, like sign-in or payments failing, stay loud and stay paged.

After each rotation I'd read through every page that fired. Any page that needed no action gets demoted or deleted before the next person takes the pager.

Reduce Burnout in Engineering On-Call Without Raising Risk

Assign Distinct Responder Roles

Deploy Global Dispatch and Impact Filters

Page Only for Actionable Incidents

Halt Risky Releases Through Error Budgets

Elevate Thresholds for Genuine Harm

Make On-Call a Career Milestone

Route Deferrable Notices to Morning Digests

Related Articles