Who Fixes It at 02:00

Everyone can answer who builds it. Almost nobody can answer who is awake when it breaks.

In an earlier piece we set out four control points that decide whether you own your stack: where your data sits, what you can revoke, whether you can leave, and whether you can operate it. 🧱

Three of those can be answered in a meeting. The fourth cannot, and it is the one that quietly decides the other three.

The question was: when something breaks at 02:00, who fixes it, and do they work for you?

This is our answer, written down, including the parts that are uncomfortable.

🌙 What 02:00 actually looks like

It is rarely dramatic. Production does not usually fall over in one piece. What happens is smaller and more awkward:

• An identity provider stops issuing tokens, so nothing is technically down, but nobody can log in.
• A nightly data pipeline finishes without error and writes yesterday's numbers into today's dashboard.
• An AI agent starts answering confidently from a stale index, and every answer is plausible.
• A certificate expires on an internal service that four other services quietly depend on.

None of these page loudly. All of them are worse by morning. And every one needs somebody who understands the system, not somebody who can restart it.

🕰️ Follow-the-sun is a structure, not a rota

Plenty of vendors will tell you they are 24x7. Ask the follow-up question: is the person on at 02:00 doing their job, or doing a night shift?

That distinction is the whole thing. A night shift is somebody awake at the wrong time, working alone, cautious, and inclined to wait until morning rather than act. Over months it produces tired people and slow decisions.

Follow-the-sun means the work moves to wherever it is currently working hours. Our teams sit in Germany and in India. When it is 02:00 in Frankfurt it is 05:30 in Bengaluru, and the people picking it up are at the start of a normal working day, with colleagues beside them and the authority to make a call.

That is not a scheduling trick. It changes what the person on the other end is able to do.

🎯 What Class-A operations means when something is actually broken

The phrase is easy to put on a slide, so here is what we mean by it in practice.

• The responder can decide. They can roll back, fail over, revoke a credential or stop a pipeline without waking anyone for permission. An escalation path that requires three approvals at 02:00 is not an escalation path.
• The action is auditable. Every intervention is logged, including what an automated agent did on its own. In a regulated environment, "we fixed it" is not an answer. "Here is what changed, when, by whom, and why" is.
• Change management still applies at night. The temptation at 02:00 is to do whatever stops the alarm. Regulated change and release discipline does not get suspended because it is dark.
• The handover is written, not remembered. The value of follow-the-sun collapses if the next region starts by reconstructing what happened. The shape of the handover matters more than the speed of the fix.
• Somebody owns it by name. Not a queue, not a rota, a named owner who sees it through to the post-incident review.

⚖️ Why we run it across two countries rather than one

We are an Indo-German group, and that is a delivery decision rather than a cost one.

From the German side comes regulatory discipline: decades of banking, insurance and audit work, and the habit of designing for someone who will ask awkward questions about it a year later. From the Indian side comes engineering depth and the capacity to staff real coverage rather than a thin veneer of it.

Operations is exactly where those two have to meet. A control boundary nobody can operate is theatre. An operating team with no governance model cannot be let near a regulated workload. You need both, and they are difficult to buy separately.

🤖 Where AI helps at night, and where it does not

Our support desk puts an AI assistant first and a human immediately behind it. That split is deliberate.

• AI is good at triage, correlating alerts, summarising what changed in the last hour, and answering the routine question at 02:00 that does not need a person.
• AI is not good at deciding whether to fail over a payment system. That is a judgement call with commercial consequences, and it belongs to a person who can be held to it.

The failure mode we design against is an AI that sounds confident about an incident it has misread. Speed at night is worth very little if the first action is the wrong one.

🔍 Four questions to ask your current provider

1. At 02:00 on a Sunday, is the person who answers in their working day, or on a night shift?
2. What can they do without waking somebody more senior? Name the actions.
3. Show me the audit trail from the last real incident, including anything automated.
4. Who owns it by name until the post-incident review is signed off?

If the answers are vague, the 24x7 in the contract is a billing arrangement rather than an operating model. 🌍

📅 Talk to us

We run managed operations for AWS, Azure, Google Cloud and on-premise workloads, across Europe and India, for clients in banking, insurance, healthcare, public sector and manufacturing.

Book a 30-minute operations review. We will walk your escalation path with you and tell you honestly where it breaks at 02:00.

Managed services, cloud and AI: mejuvante.ai/enquiry-1