Developer Offshore logoDeveloper OffshorePowered by StealthAgentsScope a developer
Table of contents

Developer Offshore research

On-call load distribution in distributed offshore support

An analysis of on-call load in distributed offshore support: what load includes, how to measure it without surveillance, time-zone effects, escalation quality, fatigue, and retention.

Use this report with the Research library and related developer guides to turn evidence into a bounded work brief.

On-call load distribution in distributed offshore support

Key stats

  • 4 load dimensions: alerts, interventions, decision requests, and after-hours pages
  • 2 fairness views: total load and unsocial-hours load
  • 3 quality signals: escalation accuracy, runbook use, and repeat pages
  • 1 rule: load is reviewed by rotation and service, not by individual ranking

Key takeaways

  • Measure load across alerts, interventions, decision requests, and unsocial-hours pages.
  • Report total load and unsocial-hours load separately so fairness is visible.
  • Pair volume with escalation quality, because a noisy rotation is not the same as a busy one.
  • Treat a rising after-hours share as a retention risk, not only a scheduling problem.
  • Review the rotation monthly and change the service or the routing before adding people.

Question and scope

This report examines how on-call load is distributed across a buyer's engineers and an offshore support provider. The unit is one rotation period, usually a week, for a defined service or service group. The scope covers paging volume, interventions, decision requests, and the share of work that falls outside the responder's normal hours. The analysis is about the design of the rotation and the service, not about the diligence of individual responders. It uses aggregate signals that a buyer and provider can review together.

The question is whether a buyer can see an unfair or unsustainable distribution before it becomes attrition, missed pages, or a contract dispute. On-call load is often discussed anecdotally, with the loudest week standing in for the pattern. The analysis looks for a small set of measures that make the pattern visible without turning the rotation into a surveillance system. The report also considers what the data cannot show, including the informal help that never becomes a ticket. That gap is a reason to pair the numbers with a short conversation.

What load includes

Counting pages alone misses much of the work. A responder may receive few alerts but spend hours on decision requests, customer follow-ups, and manual checks that never generate a page. Those tasks consume attention and are often invisible in the tooling. A useful definition of load includes them.

Include four dimensions: alerts that require action, interventions such as restarts or rollbacks, decision requests from the buyer or support desk, and after-hours pages. The first three describe volume and the fourth describes personal cost. Report them separately so a rotation with many daytime requests is not confused with one that wakes people at night. Agree the definitions before collecting the first week, or the categories will overlap. A simple shared sheet is enough to start.

Measuring without surveillance

Use aggregate counts from systems the team already uses. The paging tool, the incident tracker, and the change log can supply most of the data. Avoid screen monitoring, keystroke logging, or anything that feels like watching individuals. The goal is to understand the rotation, not to grade the responder.

Report at the rotation and service level, not the person level. If a person is consistently overloaded, that should surface through the rotation design and the team's own review, not through a public ranking. Aggregate data protects privacy and reduces the incentive to hide work or avoid asking for help. Share the aggregate with the team and let them challenge the classification. The people on the rotation usually know which pages were avoidable.

Time zone and handoff effects

A follow-the-sun rotation can distribute load evenly on paper and unevenly in practice. The night shift may receive the same number of pages but with less support, slower access to the buyer, and a higher cost to the person. Count the unsocial-hours share alongside the total, and note when a handoff leaves an open decision. Record the time of each handoff and the number of items still open at that point. That small field turns an impression about fairness into something the team can inspect.

Handoffs also shift load. A responder who inherits a messy incident carries work that another shift created. Record the handoff time and the number of open items transferred. If one shift consistently inherits more open work, the handoff format or the upstream triage needs attention. The schedule may be fair while the work is not.

Escalation quality versus volume

A high page volume is not automatically a failure, and a low volume is not automatically healthy. The more useful question is whether the pages were appropriate. A page that could have been an email, a dashboard check, or a next-day ticket is noise. A page that arrives too late is a missed signal. Both distort the load.

Sample a handful of pages each month and classify them as actionable, avoidable, or late. Pair the classification with the runbook used and the escalation path taken. A rotation with many avoidable pages needs better alerting, not more people. A rotation with many late pages needs earlier thresholds, not a longer on-call list.

Fatigue and retention signals

On-call fatigue shows up in small ways before it shows up in a resignation. Responders stop volunteering for hard incidents, take longer to acknowledge pages, or trade shifts more often. Those signals are easy to miss in a delivery review that focuses on tickets and releases. The rotation review should ask about them directly.

A rising after-hours share is a retention risk, especially when it persists for several weeks. It may be a seasonal spike, a service that needs repair, or a routing rule that sends too much to one time zone. Distinguish the cause before changing the schedule. Adding people to a noisy service spreads the pain without fixing it.

Contract and staffing implications

A support contract should state the coverage window, the expected response, the escalation path, and how after-hours work is handled. Without those terms, the offshore provider may absorb load that the buyer assumed was covered, or the buyer may pay for coverage that does not match the service's real behavior. Put the coverage terms in the same document as the response targets so they are read together. Then review the actual pages against what the contract assumes. A mismatch is a reason to renegotiate the model, not to blame the responder.

Review the contract against the measured load each quarter. If the service now pages mostly at night, the coverage model may need to change. If the buyer's own engineers take most of the day pages, the split may no longer match the intent. The contract should follow the work, not the other way around.

A practical protocol

Collect the four load dimensions for one rotation cycle. Report the total and the unsocial-hours share by rotation and service. Sample ten pages for appropriateness. Review the result monthly with the support lead and the buyer's service owner, and record one change with an owner.

Keep the review short and focused on the service. Ask whether the pages were actionable, whether the runbook was used, and whether the escalation reached the right person. If a change is made, measure the same fields for a month. The protocol is working when the after-hours share and the avoidable-page count both move in the right direction.

A worked example

A service pages an average of eighteen times a week. The night rotation receives six pages, but those six account for most of the after-hours interventions because the buyers are offline. The daytime rotation receives more pages but can resolve them with a quick message. The total looks balanced and the experience does not.

The team separates the after-hours share and finds that two alert rules produce most of the night pages. One is a duplicate of a dashboard warning, and the other fires before the mitigation has time to work. They adjust the rules and add a handoff note for open decisions. The total falls slightly and the night load falls sharply.

Adoption sequence

Start with one service and one rotation cycle. Collect the four load dimensions and ask the responders to flag pages they considered avoidable. Share the aggregate with the team before changing any rule. The first review is for agreement on the definitions, not for a target.

In the second month, change one thing: an alert rule, a routing path, or a handoff note. Measure the same fields and compare. Keep the change if the after-hours share or the avoidable-page count improves. If it does not, revert it and try the next cause rather than adding more process.

Counterevidence and limits

Not all after-hours pages can be removed. Some services genuinely need immediate attention, and a provider that never pages at night may be missing real problems. The goal is to remove avoidable load, not to achieve zero pages. A quiet rotation still needs a tested escalation path.

Load data also depends on honest classification and stable tooling. A change in the paging system can make the trend incomparable. Record the tool and the definition version with the numbers. The review should treat the data as a conversation aid, not as a contractual score without context.

Evidence table

Signals, inspection points, and accountable owners
SignalWhat to inspectOwner
OutcomeAcceptance evidence for the bounded taskTask reviewer
ControlAccess, test, and approval boundaryInternal owner
HandoffOpen risks and next decisionNext owner
Good distributed work is observable at the handoff: the result, evidence, limitations, and next owner are explicit.

Frequently asked questions

Should on-call load be measured per person?

Measure at the rotation and service level. Person-level rankings discourage help-seeking and can turn a scheduling problem into a performance issue.

Is a low page volume a good sign?

Not by itself. It may mean alerting is too quiet or problems are being found late. Pair volume with appropriateness and defect signals.

How do we handle after-hours work in the contract?

State the coverage window, expected response, escalation path, and how after-hours work is compensated or rotated. Review it quarterly against the measured load.

Sources

  1. NIST SP 800-61 Rev. 2, Computer Security Incident Handling Guide
  2. Google SRE Workbook, Incident Response
  3. CISA Incident and Vulnerability Response Playbooks

Related research