🇸🇬 HireDeveloper.sg

Our Email Died Through a Whole Working Day and Nobody Owned It: the 3 Roles I Am Hiring After the Microsoft Outage

Server room corridor lit in red, representing the Exchange Online infrastructure failure that spilled into Asia Pacific business hours on 13 November 2026
Bryan

Bryan

Delivery & Offshore Teams Expert · November 15, 2026 · 9 min read

TL;DR

  • •What happened: A Microsoft 365 failure centred on Exchange Online began on 12 November 2026 and rolled into Asia Pacific business hours the next day, taking Teams, Copilot and parts of SharePoint with it. The Singapore government confirmed its own email was down.
  • •The real finding: The companies that kept working were not the ones with better architecture. They were the ones where somebody was allowed to make a call without written approval from a person they could not email.
  • •What I am hiring: Three roles, none of them a second cloud provider: an incident owner for third party failures, an engineer who treats the vendor as a dependency with a contract, and a workplace engineer who can run an out-of-band channel.

What I remember about Friday morning is not the outage. It is a message in a group chat, from a delivery lead, asking whether she was allowed to tell a client that a scheduled report would be late. She could not reach the person who normally approves that, because reaching him required email. So she waited, and the client found out from the absence of the report. Nothing in our architecture failed that morning. Our decision-making did, and that is a staffing problem rather than an engineering one.

What Actually Broke, in Order

The failure began on Thursday 12 November and was still being worked through on the Friday. Exchange Online was the centre of it: Microsoft service health notices described users unable to reach mailboxes through several connection methods, and the company tracked the incident as EX1550543 with a companion identifier, MO1550821, for the dependent services. That dependency list is the part worth staring at, because it is where a mail outage becomes a company outage: Teams, Copilot, Planner, To Do and some SharePoint functions all degraded behind the same root problem.

On scale, the trackers tell a consistent story. Reports across five Microsoft services reached roughly 19,650 a minute within two hours, against fewer than 15 a minute on an ordinary Thursday. In the United States, Outlook reports passed 130,000 around the early afternoon. Then it moved west to east: coverage of the Friday spillover into Asia Pacific business hours noted problems reported in India, Australia and Japan in local morning, and Singapore reports peaked at about 567 around half past nine, with the government confirming its own email was affected.

On cause, be careful what you repeat. Microsoft stated the problem was technical and unrelated to a cyber attack or to AI agents, and pointed to an issue arising from an earlier network capacity upgrade that surfaced following a recent network change. Reporting during the event also described a process on affected servers becoming overwhelmed and ceasing to handle requests, with a restart of that infrastructure as the remedy. Those two accounts are not contradictory but they are not the same statement either, and the company has said it will publish a full post incident report with a root cause. Until then, anybody presenting a confident causal story, including the one you are reading, is working from partial information.

Our expert take

The detail I would put in front of a board is the extension of email timeouts to 48 hours during recovery. It sounds like housekeeping and it is actually the most informative public signal of the week, because it tells you the operator expected a backlog measured in days rather than hours. Any continuity plan written on the assumption that a mail outage is a two hour inconvenience was invalidated by that single configuration change. If your plan has a recovery time objective for a vendor you do not operate, ask yourself where that number came from. In most companies the honest answer is that somebody felt it was about right.

Why This Landed Harder in Singapore Than the Numbers Suggest

Two structural reasons, neither of them about Microsoft.

The first is timing. Because the incident began during the American afternoon and recovery rolled through Asia overnight, Singapore teams did not experience an outage with a beginning and an end. They started a working day inside somebody else partial recovery, which is the worst phase to be in: services intermittently available, mail arriving hours late and out of order, and no clean signal telling you whether to invoke a fallback or wait ten more minutes. Teams that had rehearsed a binary switch, up or down, had no rule for the state they were actually in.

The second is organisational shape. A lot of Singapore companies run lean local teams against regional or global IT functions, which is efficient for about 363 days a year. During a multi-region vendor incident the local office has the problem and the authority sits in a different time zone, behind the communication channel that just failed. That is the gap my delivery lead fell into, and I would bet it accounted for more lost output across the island on Friday than the actual unavailability did.

Our expert take

The partial recovery state deserves its own line in your plan, and almost nobody has one. Every continuity document we see treats a dependency as binary, available or unavailable, because that is easy to write a rule for. Friday was neither: mail arriving late and out of order is arguably worse than mail not arriving at all, since staff keep assuming the channel works and keep sending into it. If you write one sentence into your plan this month, make it the rule for degraded rather than down. Something as blunt as “if a message is more than fifteen minutes late, the channel is treated as failed” is cruder than the reality and still better than deciding in the moment.

How a mail outage becomes a company outageLayers 1 and 2 have an owner. Layers 3 and 4 usually do not.1. Mail and identity path degradedVendor owns this. You cannot fix it and should not try.2. Collaboration tools followChat, meetings, files, task tools. Workplace IT owns this.3. Approvals and decisions stallNo owner in most companies. This is where the day is lost.4. Customers find out from silenceNo owner. The reputational cost sits entirely here.A simplified teaching model. Your dependency graph and your contractual position differ, and the vendor post incident report supersedes any of this.

The 3 Roles I Am Actually Hiring For

1. An incident owner for things you do not operate

Not a new team and not a network operations centre. One named person, with a named deputy, who is accountable for third party service incidents and holds the authority to declare the fallback plan active without approval from anyone. They publish the internal announcement, they own the customer-facing message, and critically they decide when the organisation stops waiting and starts working differently.

The authority is the whole job. A policy document that requires sign-off from someone unreachable is not a continuity plan, it is a description of one. We screen for this in a way that surprises candidates: ask them to describe a decision they took during an incident that they were not formally entitled to take, and what happened afterwards. The people who have genuinely held this responsibility have that story and some residual discomfort about it. The ones who have only written the runbook do not. If you are building out the surrounding rota, the mechanics are in building an on-call rotation for a Singapore engineering team.

2. A platform engineer who treats the vendor as a dependency with a contract

Most teams can tell you their own service availability. Very few can tell you, without looking, which third party services sit on the critical path of a customer-visible workflow, what each one promises contractually, and what the agreed remedy is when it is missed. That mapping is a real engineering artefact and it rots within a quarter if nobody owns it.

The hire I want here is senior enough to read a service agreement and technical enough to instrument the dependency. The screen that works is a single question: what did you measure to know your vendor was down before the vendor status page said so. Everybody who has lived through this has built a crude synthetic check, usually in an afternoon, usually embarrassing, and it is the single highest return piece of code in their monitoring stack. The surrounding discipline sits in our method for an engineering SLO and error budget policy, and for the reliability role itself in hiring a site reliability engineer in Singapore.

3. A workplace engineer who can stand up an out-of-band channel

The least glamorous of the three and the one that would have saved my Friday. Somebody whose job includes a communication path that does not depend on the identity provider carrying your mail: a second channel, held warm, with the current phone numbers of the twenty people who matter, and a rehearsed way to reach your largest customers without sending an email.

This role is usually filled by whoever is best at IT support, which is not the same skill. What you need is someone who thinks in dependencies and will notice that your status page authenticates against the thing that is down, that your emergency contact list lives in a document behind the same sign-on, and that nobody has tested the second channel since the day it was created.

Our expert take

If budget allows one of these three and not all of them, take the first. The incident owner is the only role that converts a vendor failure from an open-ended wait into a sequence of decisions, and the other two produce artefacts that nobody acts on without it. We have reviewed a reasonable number of continuity plans for Singapore employers this year and the pattern is consistent: the documents are fine, the dependency maps are mostly accurate, and there is no named person with permission to say the words “we are switching to the fallback now”. That sentence, said by somebody allowed to say it, is worth more than the rest of the plan.

Did Friday expose a gap in your team

Tell us what stalled and who was waiting for whom. We will tell you honestly whether it is a hiring gap, an authority gap or a process you can fix next week for nothing.

Discuss it with us

What I Am Not Doing, and Why

I am not moving mail to a second provider, and I would push back hard on anyone proposing it this month. Running mail, identity and collaboration across two vendors roughly doubles the integration surface, the licence cost and the administration, and it adds failure modes of its own, all to hedge an event that arrives once every year or two. The arithmetic does not work unless you are regulated into it.

I am also not writing a longer continuity document. The one we had was adequate and it did not help, because nobody had the authority it assumed and nobody had rehearsed the ambiguous state. Rehearsal is cheap: pick a Tuesday, declare mail unavailable for ninety minutes, and watch what stops. You will discover the three things that actually matter faster than any tabletop exercise, and the finding is usually not technical.

What I am doing, beyond the three hires, is deleting one sentence from our process: the one requiring written approval for client communication during an incident. That sentence existed to prevent a small reputational risk and on Friday it created a larger one. If your own process has an equivalent, this is the week it is cheapest to change, because everybody remembers why.

Four responses, ranked by cost per unit of protectionSecond provider for mailDoubles integration, licence and admin costAdds new failure modes of its ownJustified mainly when a regulator requires itNamed incident ownerOne person, one deputy, real authorityConverts waiting into decisionsHighest return of the fourDependency map with checksKnow you are down before the status page says soNeeds an owner or it rots in a quarterAn afternoon of code, reviewed monthlyRehearsed out-of-band channelSecond path, warm, not behind the same sign-onNinety minute drill, once a quarterNearly free, almost never doneRelative costs, not a recommendation for any specific company. Regulated workloads, contractual commitments and tenant design change the ranking.Verify your own obligations before removing any approval step described in this article.

Where This Read Stops

Three caveats, since this is being written while the incident is still being cleaned up.

  1. The cause is not settled and I am not the one to settle it. The accounts available now come from vendor status updates and contemporaneous reporting. The post incident report will be more accurate than anything in this article and may contradict parts of it.
  2. Downdetector figures are a floor, not a measurement. They count people who chose to report, which systematically undercounts corporate users whose first instinct is to message their own IT function. Treat the Singapore peak of 567 as evidence that it happened here, not as a measure of how much it cost.
  3. The three roles assume a company of a certain size. Below roughly thirty people, all three are one person who already works there and needs permission rather than a job title. The hiring argument starts to bite when the local office has its own customers and the approval chain leaves the country, and the Singapore cloud-first guide covers the architectural half of the same question.

Frequently Asked Questions

What actually failed in the November 2026 Microsoft 365 outage?
Exchange Online sat at the centre of it. Microsoft service health notices described users being unable to reach their mailboxes through several connection methods, and the failure propagated to everything depending on that mail and identity path: Teams, Copilot, Planner, To Do and some SharePoint functions. Microsoft tracked the incident as EX1550543 for Exchange Online and MO1550821 for the dependent services. On cause, the public statements shifted as recovery progressed. Microsoft said the problem was technical rather than related to a cyber attack or to AI agents, and attributed it to an issue arising from an earlier network capacity upgrade that surfaced after a recent network change. Separate reporting during the event described a process on affected servers being overwhelmed and stopping request processing, with the infrastructure restarted as a result. The company said it would publish a full post incident report with a root cause once recovery completed, and that report is the document worth reading rather than any of this.
How bad was it in Singapore specifically?
Bad enough to cost a working day in a lot of offices, and the Singapore government confirmed its own email was affected and that it was working with Microsoft on a resolution. Downdetector reports from Singapore peaked at around 567 at about half past nine in the morning before tapering, which badly understates the real impact for two reasons: most affected staff never file a report, and a mail outage hits every function in a company at once rather than one product team. Globally the same tracker recorded reports across five Microsoft services reaching roughly 19,650 a minute within two hours, against fewer than 15 a minute on an ordinary Thursday. The disruption then rolled into Asia Pacific business hours on the Friday, with problems reported in India, Australia and Japan in local morning, which is why so many Singapore teams started their day inside somebody else recovery window.
Is the answer to stop using a single cloud provider?
Almost certainly not, and the multi-cloud reflex is the most expensive wrong answer available. Running mail, identity and collaboration across two providers roughly doubles your integration surface, your licence cost and your administrative overhead, and it introduces failure modes of its own in exchange for protection against an event that lands once every year or two. The cheaper and more effective posture is to assume the provider will fail and design for the hours in which it is failing: a communication channel that does not depend on the same identity provider, a documented way to reach customers when your mail is down, and a clear rule about which decisions people may take without written approval. That is a planning and staffing problem rather than an architecture problem, and it costs a fraction of a second vendor.
Which engineer actually owns a vendor outage, in practice?
In most Singapore companies under about two hundred people, nobody does, and that is the gap the week exposed. Application teams own their own services and genuinely cannot fix Exchange Online. The IT or workplace function owns the tenant but is usually staffed for provisioning and support rather than for incident command. So the outage lands in the space between them and gets handled by whoever happens to be senior and awake. The fix is not a new department. It is to name one person accountable for third party service incidents, give them the authority to declare the fallback plan active without asking permission, and make sure the status page, the customer messaging and the internal announcement are things they can publish themselves. One named owner with authority beats a well written policy with no owner, every time.

Hiring for resilience rather than for headcount

We screen platform, reliability and workplace engineers for Singapore employers, and we will tell you when the role you are describing is really an authority problem you can fix without hiring anyone.

Discuss it with us