On-call compensation

The common models

A flat stipend is a fixed amount per on-call day or week, whatever happens. It is simple to run and predictable for both sides. Its weakness is that it pays the same for a silent week and a week of 3 a.m. pages, so the person who is woken most often feels underpaid first.

Standby pay is an hourly rate for the hours someone must stay reachable, usually a fraction of their normal rate. It recognises that being on call limits what people can do with their evening. It needs a clock: someone has to record when the shift starts and ends, and payroll has to handle it.

Pay per incident is an amount, or time in lieu, for each page answered outside working hours, sometimes with a minimum per call-out. It rewards the work actually done, and it makes the cost of noisy alerts visible in the budget. It can also create the wrong incentive, so the definition of an incident needs to be tight.

Time off in lieu is a day off after a heavy week, or hours back for hours worked at night. It protects recovery. It does not pay for the constraint of being reachable, so it works best beside one of the models above.

Most mature setups combine a stipend for being reachable with extra pay or time back for incidents actually worked.

What on-call costs the business

On-call has a cost whether or not you pay for it. Paid, it is a line in the budget. Unpaid, it shows up as attrition, slower hiring, and engineers who stop volunteering for the services the business depends on most. A compensation model is a choice about which of those costs the company carries.

Fairness is the retention lever. People accept a demanding rotation when the load is shared evenly and the pay reflects it; they leave when the same few people absorb the night pages. Tracking pages per person per rotation makes that imbalance visible before it turns into a resignation. The signs to watch for are set out in on-call burnout.

Noise is the cost nobody budgets for. Every page that did not need a person is paid, or unpaid, disruption, so cutting flaky alerts lowers the bill and the burnout at the same time. It usually pays back faster than raising the stipend.

Customers are on the other end of this. A rotation that is rested, fairly paid and clear about who answers is the one that picks up the page at 3 a.m. and fixes the problem, which is what a customer’s renewal depends on. The escalation policy decides who is woken; the compensation decides whether they will still be there next year.

Check the law first

Employment law can decide whether standby time counts as working time. In the EU, courts have held that standby with tight constraints, such as having to respond within minutes, can count as working time. In the EU this matters first for the limits on working hours and rest periods. How standby is paid is left to national law and employment contracts.

In the US, the question is usually whether the employee is exempt from federal overtime rules, which is not the same as being paid a salary. For a non-exempt employee, time spent engaged to wait, where the restrictions are so tight that the time is not really their own, can count as work. Some states add stricter rules. Rules differ by country, by state and by contract, so check with an employment lawyer before you set a policy. This page is not legal advice.

Two questions cover most of the ground. How quickly must the person respond, and how far from work can they be? And do they carry a phone, or must they stay at a desk? The tighter the answers, the more likely the time counts as work.

Fairness is part of the pay

• Rotation size. A rotation of four means on call every fourth week. Below that, no stipend makes it sustainable. How a small team can run one is covered in on-call for a team of five.
• Follow the sun. Teams in several time zones can hand over so that nobody is paged at night.
• Recovery. A rule that someone paged in the night starts late the next day costs little and is noticed.
• Swaps and opt-outs. Let people trade shifts, and make it easy for someone with a new baby or an exam to step out for a while without a negotiation.
• Visible load. Publish pages per person for each rotation. Fairness that people cannot see is not fairness they believe in.

Choosing a model

Start from how often people are actually paged, then pick the model that matches.

1. Rarely paged out of hours. A flat stipend for the shift, and time in lieu on the rare night worked. Cheap to run and enough to acknowledge the constraint.
2. Paged sometimes. A stipend, plus pay or time back per incident. The variable part keeps the total honest.
3. Paged often. Treat the pages as the problem before the pay. Cut noise, add people to the rotation, and only then settle the pay. If the pages are all real, the team needs more engineers on the rotation, not a bigger stipend.

Whatever you choose, write it down: who is covered, what the stipend or rate is, what counts as an incident, how time in lieu is booked, and who reviews the arrangement and when. An arrangement that lives in a manager’s head is unfair by default. Reading the numbers behind it is the MTTA and MTTR work: a rising MTTA often means the rotation is tired.

Mistakes that cost more than they save

• Paying only for pages. The constraint of being reachable is the price of the job, quiet weeks included. Pay for that as well as for the work.
• Forgetting the follow-up. An incident does not end when the page is acknowledged. The fix, the handover and the write-up take time too, so count them in the definition of an incident.
• Leaving out the escalation path. The people at the end of an escalation chain, often senior engineers and managers, get woken as well. If they are outside the scheme, the most experienced people carry the load unpaid.
• Changing the deal without notice. People plan their year around the arrangement. Announce a change before the new rota starts, not after the first bad night.