The cost of running your own 24x7 rota

A year is 8,760 hours. One full time engineer, after holiday and public holidays, covers about 1,856 of them. That is roughly 21 per cent of the year, from one salary.

Divide the first number by the second and continuous cover needs 4.72 people. Round up, because people are not divisible, and the honest floor for 24x7 is five. Five, with nobody spare, nobody ill, and nobody on the beach.

The plan that gets written instead

Almost nobody writes a business case for five engineers. What gets written is a case for one, sometimes two, with a rota to be worked out later. The rota is always worked out later.

What arrives instead is an arrangement. Two or three engineers take alternating weeks. It works, in the sense that the phone gets answered. It also means a third of the year on call for each of them, forever, on top of the job they were actually hired to do.

No individual decision here is unreasonable. The company needed cover, the engineers volunteered, the alternative was worse. Collectively they have built a rota that depends on nobody ever leaving.

What the business case leaves out

The spreadsheet usually contains salaries. The things it tends not to contain:

Recruitment, at a market rate for a skill set that is competitive to hire. Notice periods, during which the rota is already short. Onboarding, because an engineer who does not know the estate is not cover, they are an escalation path with a laptop. Training and certification. Management time. Tooling. And the quiet one: the cost of the rota being a single resignation away from collapse.

That last one is not a cost until it is the only cost that matters. A five person rota that loses one person is a four person rota, which is a different job, which is why the second person leaves.

Cover is not the same as capability

There is a version of this problem that more headcount does not solve at all. Somebody answering the page at three in the morning needs to know which signal means something, where the runbook is, and what the blast radius of a restart looks like on that particular service. Cover without that knowledge is a person awake and worried.

This is the part that makes the arithmetic misleading in both directions. Five engineers who have never operated the estate are not cover. One engineer who has operated it for three years is not cover either, because there is only one of them.

The case for building it

Some companies should build it, and it is worth saying so plainly.

If operations are the product, the capability belongs in house. If the estate is genuinely unusual, the learning curve is steep enough that owning it is cheaper than explaining it. If the company is large enough that five engineers is a rounding error against the platform team it already has, the marginal cost of proper cover is low and the control is worth having.

A team already staffed for out of hours does not need a managed service to hold the pager. It may need help with the backlog underneath, which is a different purchase.

The case against

For most companies at Series A or B, the five people do not exist and will not be approved, so the real comparison is not five engineers against a managed service. It is three tired engineers against a managed service, and the three tired engineers are also the people building the product.

That is the trade that gets made by default, without anybody deciding it. The rota is not a line in the budget. It is a slow tax on the roadmap, paid in the mornings after.

For what it is worth, we built the other answer: Critical Support puts a team on the pager instead, at around 75% less than building it, at the standard of an in house 24/7 team. Critical Cloud reduces mean time to resolve incidents by 40 to 60 per cent. For teams already running Datadog, we operate the cloud inside the environment they already own, which is the Datadog cloud MSP route.

The rota calculator on that page runs the same arithmetic as this piece, with the salary set to whatever the real number is.