On-call compensation in 2026 is still the most awkward conversation in engineering management. Three years ago we published a guide to how the tech industry pays for on-call, and it remains one of the most read things on this blog. The reason is not that it was clever. It is that almost nobody publishes real numbers, so engineers and managers alike end up negotiating in the dark.

This is the 2026 update. The framework from 2023 held up better than we expected: the industry still splits into companies that treat on-call as extra work deserving extra pay, and companies that treat it as part of the job covered by a high base salary. What has changed is the size of the numbers, the sharpness of the legal picture in Europe, and the fact that on-call load itself has grown. This post gives you current benchmarks, the five models teams actually run, what the law now requires, and a method for setting a rate you can defend.

What has changed since our 2023 on-call compensation guide

Start with what did not change, because it matters. In 2023 we described seven approaches that collapse into two paradigms: on-call is additional, or on-call is the job. That is still the correct top-level split, and if you are trying to understand why two companies pay so differently for apparently identical work, this is almost always the answer.

Three things are genuinely different in 2026

The pay is more standardized. In 2023, asking five companies about on-call rates got you five incompatible answers. In 2026 there is a recognizable market rate in most regions, and a recognizable structure: a flat availability stipend plus separate pay for time actually worked. Engineers now come to interviews with comparables.

The legal position in Europe firmed up. The European Court of Justice line on when standby counts as working time has been applied consistently enough that it is now a compliance question rather than an academic one, and German case law has continued to refine it.

The load went up. This is the part most compensation discussions miss. The infrastructure your team depends on has been failing more often, not less. In August 2026 alone, GitHub logged more than twenty incidents in under a month, and ThousandEyes counted 534 global network outage events in a single week. Paying the same stipend for a rotation that now generates twice the pages is a real terms pay cut, and engineers notice.

On-call compensation benchmarks for 2026

Here are the numbers, with sources. Treat them as anchors rather than gospel: rates vary enormously by country, seniority, criticality of the systems, and how aggressive the response time requirement is.

Europe: the weekly stipend is the norm

European tech companies overwhelmingly pay a flat amount per week of rotation. The figures below come from the Pragmatic Engineer's ongoing collection of on-call compensation data, compiled by ilert in its 2026 on-call compensation guide.

Outside the euro zone the same shape holds at lower absolute levels: Mastercard in the UK pays roughly 470 pounds per week, Wise around 300 pounds, and PayPal in Germany reports about 350 dollars per week. The spread between Klarna at 500 euros and Zalando at 1,050 euros for broadly similar work is the single most useful fact in this post. There is no objectively correct rate. There is only what your local market and your own load justify.

The United States: stipend plus incident pay

US practice is less uniform, partly because far more engineers are salaried and exempt, which removes the overtime mechanism entirely. SHRM data indicates roughly 70 percent of North American IT, technology and life science organizations provide some supplemental compensation to technical employees required to remain on-call outside working hours. Among those paying a flat amount, reported averages land near 250 dollars per week, or broken down, about 23 dollars per weekday, 45 dollars per weekend day and 50 dollars per holiday.

The more common structure at engineering-led companies is a hybrid: a weekly stipend in the 200 to 500 dollar range for carrying the pager, plus pay at 1.5 to 2 times the normal hourly rate for time actually spent on incidents, plus compensatory time off after a severe or overnight event. For non-exempt technical staff, 1.5 times the standard hourly rate for hours worked while on call is the typical figure.

Percentage of base: the model used at the top end

A third structure pays a fraction of your normal hourly rate for every hour of availability, scaled to how restrictive the rotation is. This is the most economically honest model, because it prices the actual constraint on your life.

Two published examples illustrate it well. Google's Tier 1 SRE rotations, which carry a five minute response expectation, compensate 40 minutes of every on-call hour outside office hours, effectively 66 percent of base hourly rate. Tier 2 rotations with a 30 minute response expectation pay 20 minutes per hour, or 33 percent. AWS teams on Tier 0 services in the EU report 25 percent of base pay per out-of-hours on-call hour, plus half a day of paid time off for each Saturday or night-time page.

Notice what those numbers encode. A five minute response requirement means you cannot leave the house, cannot drink, cannot take your kids to the park. It is priced at double a 30 minute requirement, which is roughly right. If your policy demands a five minute acknowledgement and pays a flat 200 dollars per week, you are not running a compensation model, you are running an unpriced liability.

And the companies that pay nothing

Apple, Airbnb and Netflix are the standard examples of firms that do not pay for passive standby. The argument is coherent: total compensation is at the very top of the market, on-call is disclosed as part of the role, and the base salary is understood to cover it. This is a defensible position if, and only if, three conditions hold. The base really is top of market. On-call load is genuinely light. And the expectation was clear before the person accepted the offer. Most companies that copy this model satisfy none of the three, and simply use it as cover for not paying.

Five on-call compensation models, and what each one signals

Rates are the visible part. The model you choose sends a louder message about how you view the work.

Flat weekly stipend

One number per week of rotation, regardless of how many pages arrive. Simple to administer, easy to budget, trivially understood by everyone. Its weakness is that it prices availability and ignores actual disruption. A quiet week and a week with six overnight pages pay identically, which feels unjust precisely when people are most exhausted.

Stipend plus incident pay

The stipend covers the constraint on your time, and an hourly multiplier covers work actually performed. This is the model most engineering organizations converge on, and for good reason: it separates two genuinely different costs. It requires you to track incident time honestly, which is a healthy forcing function, because a team that cannot say how many hours it spent responding last quarter cannot argue for headcount either.

Percentage of base hourly rate

Availability is paid at a fraction of normal rate, scaled to response time strictness. The fairest model on paper and the best at aligning incentives, since a tighter SLA costs the company more. Its drawback is complexity, and it produces uncomfortable conversations when two engineers on the same rotation earn different amounts for the same shift.

Time off in lieu

On-call is repaid in rest rather than money. A common formulation is eight hours on-call earning four hours of paid leave, with a full day off after a significant overnight incident. This is popular in Europe and genuinely valuable, since sleep debt is the actual harm. It fails when the culture makes taking the time off impractical. Unusable leave is not compensation.

No additional compensation

Covered above. Legitimate at the very top of the market with light load and clear disclosure. Rarely legitimate otherwise.

Whichever you pick, the evidence points at one thing mattering more than the model: consistency. Data cited by Blameless tied compensation clarity to retention more strongly than any other on-call variable. Engineers tolerate a mediocre rate applied predictably far better than a generous rate applied by favour.

The legal picture in 2026

This section is general information, not legal advice, and the details vary by jurisdiction. Check with employment counsel before writing policy.

Europe: standby can be working time, but pay is a national question

The European Court of Justice has held that standby time counts as working time under the Working Time Directive when the constraints imposed on the worker are so significant that they objectively prevent the worker from freely organizing that time and pursuing their own interests. A requirement to reach a site within eight minutes has been found to meet that bar. Where the constraints do not prevent the worker from pursuing their own interests, only time linked to actual provision of services counts.

The crucial nuance, and the one most often misread: the Directive does not govern remuneration. That falls outside EU competence. Member states may lawfully provide that pay during working time differs from pay during rest. So "this is working time" and "this must be paid at normal rates" are two separate questions with two separate answers.

The practical consequence is about rest, not money. If your standby qualifies as working time, it counts toward maximum weekly hours and minimum rest periods. Rotations that look affordable can quietly become unlawful on hours.

Germany: the useful public benchmark

Germany is worth knowing even if you do not operate there, because the public sector collective agreement (TVöD) gives a rare published rate card. Standby shifts of 12 hours or more are paid at twice the hourly rate for the day on weekdays, and four times on weekends and public holidays. Shorter standby windows earn an additional 12.5 percent of hourly rate per hour on call.

Alongside that, the ICT trade body Bitkom recommends capping on-call at 56 days per calendar year per person and guaranteeing at least eight consecutive hours of rest per shift. Those two guardrails are a good sanity check on any rotation anywhere. If your engineers are on call more than roughly a quarter of the year, the problem is your rotation size, and no stipend fixes it.

The United States: the engaged-to-wait test

Under the FLSA, the question is whether on-call time is spent predominantly for the employer's benefit. Passive on-call generally counts as rest time provided the employee can use it effectively for personal purposes. Once paged and working, those minutes become compensable working time for non-exempt employees. Most software engineers are exempt, which means none of this creates an obligation, and on-call pay becomes purely a policy and retention decision. Some states, notably California, layer on additional reporting time rules worth checking.

How to set your own on-call rate

Benchmarks tell you what others pay. They do not tell you what you should pay. Here is a method that produces a number you can defend to both finance and the rotation.

Step one: measure the actual burden. Pull the last two quarters. For each rotation, count pages per shift, how many arrived outside working hours, median minutes to resolve, and how many shifts included at least one overnight interruption. Most teams have never looked at this and are startled. You cannot price a burden you have not measured.

Step two: price the constraint separately from the work. The stipend pays for the restriction on someone's life: not drinking, staying near a laptop, declining plans. Set it by how strict your response requirement is. The Google tiering is a good mental model, where a five minute expectation is worth roughly double a 30 minute one.

Step three: price the interruptions. Use an hourly multiplier for time actually worked, 1.5 times as a floor and 2 times for overnight, plus guaranteed recovery time after a night incident. Recovery time should be automatic, not requested. Anything an exhausted person has to ask for, they will not ask for.

Step four: sanity check against the market and the law. Compare with the benchmarks above for your region. Check the rotation against the Bitkom-style guardrails: no more than about a quarter of the year on call, minimum rest guaranteed, weekly hours legal once standby is counted.

Step five: write it down and apply it identically. Publish the policy. Same rate for the same rotation regardless of level or negotiating skill. Review it twice a year against actual page volume.

The checks worth running before you sign off on any policy:

Is the rate indexed to load? If page volume doubled since the rate was set, the rate has effectively halved. Review it on a schedule, not when someone complains.
Does the response requirement match the pay? A five minute acknowledgement expectation with a token stipend is the most common mismatch in the industry.
Is recovery time automatic? Compensatory rest that requires a request will not be taken.
Is the rotation large enough? Fewer than six people means someone is on call every sixth week or worse, and no rate compensates for that indefinitely.
Is it applied consistently? Clarity and consistency correlate with retention more strongly than the absolute number.
Have you counted standby toward working hours? In Europe this can be a legal obligation, not a courtesy.
Do candidates hear the real expectation before they accept? Undisclosed on-call load is the fastest route to early attrition.

Compensation is necessary, and it is not sufficient

Paying people properly for on-call is the right thing to do and it will not, on its own, save a bad rotation. The survey evidence on burnout is consistent and grim: a Catchpoint and DevOps Institute study of more than 500 on-call engineers found 67 percent reporting burnout symptoms tied directly to paging load, and broader engineering surveys have put burnout in the mid sixties percent range. Replacing a senior SRE is commonly estimated at 150,000 to 200,000 dollars in recruiting, onboarding and lost productivity. Against that, a 1,000 euro weekly stipend is cheap, and so is fixing the alerts.

The highest leverage work is usually reducing the pages rather than pricing them. Every alert that fires without requiring human action is a tax you are paying twice, once in the stipend and once in the attrition. Teams that audit their alerts quarterly, delete the ones nobody acts on, and route the rest to a genuinely correct owner reduce the compensation problem by shrinking it.

Tooling matters here in a specific and limited way. The rotation has to be visible, fair, and easy to swap. Most of the resentment we see in on-call is not actually about money, it is about opacity: not knowing who is on next, discovering you are on call during a holiday you booked, or having a swap fall through because it lived in someone's direct messages. Pagerly keeps schedules, overrides, escalation and incident channels inside Slack and Microsoft Teams, which makes the rotation legible to everyone and makes swaps a two click operation rather than a negotiation. That does not replace paying people. It removes the friction that makes the pay feel inadequate.

The takeaway

If you take four things from this update: the European market rate for a week of on-call sits roughly between 500 and 1,050 euros, with 800 to 1,000 euros as the common band at well-funded companies. The US norm is a 200 to 500 dollar weekly stipend plus 1.5 to 2 times hourly for incident work. Percentage-of-base models priced to response strictness, in the 25 to 66 percent range, are the fairest structure available. And paying nothing is defensible only at genuinely top-of-market total compensation with light, clearly disclosed load.

Three years on from our 2023 guide, the industry has more data and better norms, but the underlying question is unchanged and still mostly ducked: is being reachable at 3am a part of the job, or extra work? You do not have to answer it the way we would. You do have to answer it explicitly, write it down, and apply it the same way to everyone. The companies that do that keep their engineers. The ones that leave it ambiguous pay for it eventually, just not in stipends.