Follow-the-Sun Coverage and the Cost of Every Handoff

Follow-the-sun scheduling is often described as a way to eliminate night work: three regions, each covering its own daylight hours, and the pager passes westward as the day ends. The description is accurate about the goal and misleading about the mechanism. Covering a full day with daylight shifts requires more than three time zones’ worth of people, a deliberate overlap at each boundary, and a handoff artifact good enough to survive being read by someone who has been awake for two hours while the author has been awake for fourteen.

The arithmetic of a day without nights

Twenty-four hours divided into three eight-hour shifts gives one shift per region. That division is necessary and not sufficient, because a handoff at the boundary requires both the outgoing and incoming responders to be present at the same time. If the outgoing shift ends at the moment the incoming shift begins, the transfer has zero duration, and the incoming person starts with whatever was written down and nothing else.

The overlap is what turns a schedule into a handoff. Thirty to sixty minutes is the usual range: enough to walk through open incidents, confirm the state of anything in flight, and ask questions. The overlap can be arranged in two ways. Either each shift is longer than the coverage it provides, for example a nine-hour shift with one hour shared at the boundary, or the shifts stay at eight hours and the overlap is subtracted from coverage, which leaves a gap that a third person must fill. The first arrangement costs each responder an extra hour; the second costs additional headcount. Neither is free, and the choice is usually invisible in the schedule diagram.

The total person-hours for one layer of coverage are therefore at least 24 plus the overlap, not 24. With two one-hour overlaps, the daily requirement is 26 person-hours, or 182 per week, against the 168 that continuous coverage nominally needs. The overhead is small in percentage terms and large in scheduling terms, because it has to be staffed by people in specific time zones at specific local times.

Three regions is a minimum, not a target

With three regions spaced roughly eight hours apart, each region can cover its own working day, and the rotation moves around the globe. With two regions, each must cover twelve hours, which means one of them works a night half. The night half is the thing follow-the-sun was supposed to remove, so a two-region arrangement is really a twelve-hour shift schedule with a geographic label.

Three regions also impose a depth requirement that is easy to underestimate. Each region needs enough people to cover its shifts every day, including weekends and holidays, and to absorb vacation and illness. A region staffed by a single person is not a region; it is a single point of failure that happens to be in a different time zone. The scheduling arithmetic that applies to a small team in one place applies to each regional team separately, and the total headcount for a global rotation is the sum of three viable local rotations, not a division of one.

This is the economic argument that decides most cases. Follow-the-sun does not reduce the total hours required to cover a week; it redistributes them into daylight. What it buys is the elimination of night shifts and the improved judgment that comes with working when rested. What it costs is duplication of local depth and the overhead of the overlaps. For a team that cannot staff three viable local rotations, a single-region schedule with an explicit, documented out-of-hours fallback is more honest than a global rotation that exists on paper and collapses when one person is on leave.

Handoff times belong in UTC

The handoff instant should be defined in UTC, with the local times derived from it, rather than defined three times in local wall-clock terms. The reason is daylight saving. Regions do not change their offsets on the same dates, and some do not change at all. A handoff set at 08:00 local in each region will drift relative to the others by an hour for parts of the year, and the overlap that was carefully sized in the design will sometimes be two hours and sometimes zero.

Defining the boundary in UTC keeps the interval between handoffs constant. The local time of the handoff then shifts by an hour twice a year in the regions that observe daylight saving, which is a visible, predictable change that can be communicated in advance. The opposite arrangement produces an overlap that silently disappears, which is discovered during an incident.

The time zone data itself is not static. Offsets and daylight-saving rules change by political decision, and the IANA time zone database is updated to reflect those changes. Systems that store a fixed offset, or that compute local times from a hardcoded table, will be wrong after the next change. Storing the zone identifier and resolving through an updated database is the only arrangement that survives; the same rule applies to the recurrence rules that generate the schedule in the first place, discussed in rotation coverage and handoff.

What an incident loses at each boundary

A handoff during a quiet shift costs the overlap. A handoff during an active incident costs the overlap plus the reconstruction of a hypothesis. The outgoing responder knows which theories have been eliminated, why a particular mitigation was not attempted, and what the last log line actually meant. Very little of that is written down, because it was built up incrementally and feels obvious at the time.

The incoming responder starts from the written handoff, the timeline, and whatever the outgoing person manages to say in the overlap. The gap between the two mental models is where incidents stall: the incoming person repeats an investigation that was already ruled out, or abandons a promising thread because its context did not transfer. Each boundary adds a fixed cost, and the cost is larger when the incident is complex.

The countermeasure is to keep one owner across boundaries whenever the incident is active, with the incoming responder acting as support rather than as a replacement. The owner stays engaged through the handoff, briefs the incoming person, and remains reachable for a defined period after their shift ends. This is a real cost to the outgoing person, and it should be bounded and compensated rather than assumed. The alternative, a strict boundary at which the owner disappears, produces the repeated-investigation pattern that makes long incidents feel endless, and it is why the command handover is treated as an explicit event rather than a change of shift.

The night shift does not disappear, it moves

In a three-region arrangement, nobody works a night shift in their own time zone. In a two-region arrangement, half of each region’s schedule is a night shift, and the fact that it is daylight somewhere else does not change the physiology of the person working it. The rotation is only as good as its weakest local schedule.

There is a second, subtler version of the same problem. The region that covers the hours when the fewest people are awake locally also tends to be the region with the least local depth, because it is often the newest or smallest office. That region then absorbs a disproportionate share of incidents that began elsewhere and were handed over unresolved. The distribution of incident load across regions is worth measuring separately from the distribution of pages, because a region can have few pages and still spend its entire shift on inherited incidents; the per-rotation measures in measuring alert fatigue need a regional split to show this.

Measuring whether it works

The quantities that matter are the number of handoffs per incident, the time spent in handoff, and the rate at which work is repeated across a boundary. The first is a property of shift length and incident duration; the second is a property of the artifact and the overlap; the third is the one that reveals whether the handoff is doing its job.

A simple proxy for repetition is the number of times the same hypothesis appears in the timeline with no new evidence between appearances. That is a countable event if the timeline records hypotheses, and it is more informative than any satisfaction survey. A rotation with a high repetition rate needs a better handoff artifact, not more people.

The handoff artifact itself is the same one described for single-region rotations: open incidents and their state, silences and expiry, changes in flight, known degradation, and the contacts who are actually reachable. The difference is that in a global rotation the artifact is read by someone who cannot walk over to the author’s desk, so the cost of an omission is higher and the tolerance for ambiguity is lower.

The economics of follow-the-sun are therefore not a comparison of page counts. They are a comparison between the cost of night work, measured in judgment and attrition, and the cost of duplicated regional depth and handoff overhead. For a service whose incidents are frequent and long, the handoff cost dominates and a well-staffed single region with a humane schedule may be better. For a service whose incidents are rare but whose response must be immediate at any hour, the global rotation is the only structure that meets the commitment without destroying the people who hold it.