Follow-the-Sun On-Call: What Actually Works Below 8 Engineers

"Follow-the-sun" gets pitched as the obvious answer to on-call for any distributed team: spread your engineers across time zones, hand off the pager as each region's workday ends, and nobody ever gets paged at 3am. In practice, most teams that try this with the headcount they actually have end up with something worse than a single-timezone rotation — a rotation that looks distributed on paper but still pages the same one or two people overnight, because the math of real coverage doesn't work out the way the pitch implies.

What Follow-the-Sun Actually Requires

Genuine follow-the-sun coverage means that at every hour of the day, UTC, at least one person on the rotation is inside their own reasonable working hours — no one is ever the person who has to wake up for a page. That requires enough people, spread widely enough, that their working-hours windows actually tile the full 24-hour clock with no gaps. Here's what that looks like for a real four-person team spanning San Francisco, London, and Singapore, using a standard 9am–6pm local on-call window for each person:

Lay those three windows (in each city's winter, standard-time offset) on a 24-hour clock and they tile it completely, with a clean handoff at each boundary: San Francisco hands off to Singapore around 01:00–02:00 UTC, Singapore hands off to London around 09:00–10:00 UTC, and London hands back to San Francisco around 17:00–18:00 UTC. Three people, three regions spread roughly 8 hours apart, and the pager genuinely never has to wake anyone. This is the case follow-the-sun advocates point to, and it's real — it just needs a specific spread most small teams don't have by default, and the handoff times themselves drift by up to an hour depending on the calendar (see which regions shift their clocks and when). Check any of these pairings live with the San Francisco ↔ London or London ↔ Singapore converters.

Why It Breaks Below About 8 Engineers

The three-region model above needs three people who can each personally cover roughly nine hours a day — which means, in practice, at least two people per region so nobody is on call solo every single week indefinitely. That's six people minimum for genuine follow-the-sun, and most teams running it sustainably (with vacation coverage, sick days, and reasonable rotation frequency) land closer to eight. Below that, one of three things happens instead:

• The "third region" is one person, so that person is on call every single day indefinitely rather than in rotation — which isn't distributed coverage, it's a single point of failure with a nicer name.
• The regions are too close together to actually tile the clock — three people across San Francisco, Chicago, and New York cover a combined 9am–9pm Pacific window at best, which is one region's workday stretched out, not three independent ones.
• Coverage looks complete on the schedule but isn't in practice, because the handoff hours land in each person's own fringe or rest time — the person "covering" 02:00–04:00 UTC is technically in their working-hours window on paper, but it's 6am for them and they're asleep.

None of this means small teams can't have sane on-call. It means the fix isn't more time zones — it's a different model that doesn't pretend to eliminate overnight pages and instead makes them rare and fairly distributed.

The Alternative: Severity-Based Paging

Instead of trying to make sure a human is always awake somewhere, separate incidents by how much an immediate human response actually changes the outcome, and only page live for the ones where it does.

• SEV1 — customer-facing outage, data loss risk. Pages immediately, wakes whoever's on call regardless of their local hour. This is the honest cost of running a service — someone occasionally gets woken up, and the fix is compensating and rotating that fairly, not pretending it never happens.
• SEV2 — degraded but not down, has a workaround. Auto-escalates only if unacknowledged after a set window (commonly 15–30 minutes), and only pages a second person if the first doesn't respond. Most SEV2s resolve fine within a slightly longer window and don't justify an overnight wake-up on a hair trigger.
• SEV3 — everything else. Goes into a queue reviewed at the start of the next business day, wherever that day starts. No page, no rotation involvement at all.

This does the actual job follow-the-sun is trying to do — minimize needless overnight wake-ups — without requiring headcount most teams don't have. A three-person team on a severity-based model with honest, infrequent SEV1s produces far fewer 3am pages than the same three people stretched across three time zones pretending to have full coverage.

Handoff Hygiene, When You Do Have Multi-Region Coverage

If your team is large enough for genuine follow-the-sun, the rotation itself is only half the system — the other half is what gets written down at each handoff. A region going off shift without a clean handoff just moves the problem: the next person starts their shift debugging from scratch instead of picking up context that already exists. At minimum, a handoff needs:

1. What's currently open, with severity and current status — not "still looking into it," but the specific hypothesis being tested and what's been ruled out.
2. What's been tried already, so the next person doesn't repeat a dead end.
3. Anything auto-resolved or flapping during the shift that's worth watching, even if it didn't page.

This is the same discipline as a decision doc that survives a time zone gap applied to incidents instead of decisions — the goal is that the next person never needs to wait for a live conversation with someone who's now asleep.

Rotation Fairness for On-Call Specifically

The same weighted-pain logic from the fair rotation formula for meetings applies here, with higher stakes: a page during someone's rest hours costs meaningfully more than one during their fringe hours, and a lot more than one during their working hours. Track not just whose turn it is, but how many overnight pages each person has actually absorbed over the last quarter — turn-based rotation and pain-based rotation diverge fast once a few real incidents land on the same unlucky person's week.

Setting This Up for Your Team

1. Plot every engineer's city on the SyncHours overlap planner and check how much of the 24-hour clock your actual team spread covers.
2. If it's under ~18 hours of combined working-hours coverage, don't build a follow-the-sun schedule — build a severity-based paging policy instead.
3. If it's close to full coverage, confirm the handoff windows land in each person's real working hours, not just their nominal region's daytime.
4. Either way, track overnight pages per person over time, not just whose turn is next — that's the number that actually predicts burnout.