When our service desk told me emails were coming back late on Friday afternoon, I asked, only half joking, whether we could opt out until Monday. I could make light of it; plenty of others could not. The Microsoft 365 outage that began on November 12, 2026 (the early hours of Friday, November 13, in Australia) disrupted Outlook, Teams and other services around the world for the best part of a day. In Britain, pharmacies lost access to NHS email and struggled to reach GP practices. In Australia, small businesses could not send orders and some fell back on fax machines. Government agencies from Japan to the United States reported disruption.
Email was not simply unavailable as a communication tool. For some organizations, it was the mechanism that delivered purchase orders, converted voicemail into actionable messages, or triggered the next step in a business process. The systems at either end could be perfectly healthy, but the process connecting them had stopped. Depending on an organization’s infrastructure and integration, the problem spread from email into many other Microsoft products.
There will be plenty of technical analysis of what went wrong at Microsoft. I am more interested in what the outage revealed about the organizations that rely on its services, and whether their resilience arrangements worked when they were needed.
Financial-services firms in many jurisdictions are subject to operational resilience requirements, including Australia’s CPS 230, the UK’s operational resilience framework and the EU’s DORA. The terminology and obligations differ, but the common concern is an organization’s ability to keep delivering important services through disruption.
Being unregulated did not spare anyone the impact. Operational resilience matters to all organizations. Operational risk, business continuity and operational resilience professionals will no doubt be reviewing how they responded, how well their plans and contingencies held up, and the capability of their response. Even if the effect on your organization was limited, the outage is an opportunity to learn.
Is your resilience capability ready for the next shared-provider outage? Download The complete guide to achieving operational resilience for practical guidance on severe but plausible scenarios, impact tolerances and third-party dependencies.
Automation and integration
Automation and integration are not new, but they are evolving. Canberra Health Services converted its voicemails, as well as faxes, into email. Members of the public who knew email was down had no way of knowing their voicemails would not be received. Online contact forms may also be a front end for an email notification, a dependency equally invisible to the customer.
Email can be built into a range of other processes. An incoming email may trigger a linked process; Microsoft does this natively with Power Automate, as do tools such as n8n and Zapier. It can also work the other way, with employees given a dedicated email address for logging requests in other tools. There are reports of support desks going quiet because lodging tickets by email had become the norm, and employees no longer remembered how to reach the web portal. Nor could support desks reply to tickets received before the outage.
It is also hard to ignore how similar disruptions could affect AI as agentic solutions scale. I have not (yet) heard of a large-scale AI-enabled business process failing spectacularly, but Microsoft 365 connectors were unavailable to Claude, ChatGPT and other AI platforms. In a back-and-forth chat, that failure is obvious. In a multi-agent orchestration, where agents choose which connected tools to use, the response to a tool failure must be designed deliberately, before an agent tries to hand off to another part of the process.
Managing third-party reliance
You may already treat email and office tools as a critical resource supporting your critical operations, and by extension treat Microsoft (or an alternative provider) as a material service provider.* Others may see them as “hidden infrastructure,” assumed to be always running or not critical.
Material service providers are classified by how directly they influence your infrastructure. Here, the disruption did not affect just you; it affected the whole ecosystem. It harks back to the CrowdStrike outage in 2024. In some respects, customers and the public may give you some latitude when everyone is affected. But for the most critical services, customers and regulators expect you to keep delivering through disruption and to avoid harm.
Perhaps you or your IT team are unhappy with the speed or effectiveness of Microsoft’s response. You might consider switching to another provider, but make sure it is a rational business decision:
- Will the switch change your risk profile? Will your resilience improve?
- Are there operational benefits or lasting cost savings that you will realistically achieve?
- What will switching cost, not just in money but in the change-management effort required of your people, and in opportunity cost?
This is not a case for staying with Microsoft; it is a case for avoiding recency bias. You can outsource a service, but you cannot outsource the risk. Providers of this kind of infrastructure understand the importance of reliable service, but they cannot reasonably know which of your services they support. That understanding, and managing the consequences of disruption, remains with your organization.
I am an advocate for working with your material service providers on continuity, but you also have to be realistic. Microsoft is not going to join your business continuity planning sessions.
Managing severe but plausible scenarios
Operational resilience regimes and guidance focus on severe but plausible scenarios, and on testing your capability against them. A prerequisite is understanding your own operations. Without that understanding, you may wrongly assume that you can stay within tolerance during those scenarios.
The obvious question is: “Did we know every process that a failure of email (or whichever tools failed in your environment) would affect?” You should, of course, update your resource dependencies based on real-world observations. But “Is our documentation up to date?” is not the most important question.
Up-to-date documentation matters only if your people have the capability to act on it. That includes leaders’ ability to respond, to activate incident management effectively and to adapt in the moment. I am an advocate for dependency mapping, but I am also a realist: chasing a (probably unattainable) 100% accurate map brings diminishing returns. I would rather have a sufficiently accurate dependency map that people understand and have tested than a theoretically complete one that provides false confidence.
Testing communication channels
This incident also highlighted another part of business continuity planning: do not assume that the resources your response needs will be available. I would expect modern plans to consider alternative communication channels, but it is worth checking that yours do.
The outage may have given you experience to draw on, but your review should go further than a desktop exercise where someone says “We’d use this channel instead” and everyone nods in agreement:
- Who decides on that channel?
- How do other people know that it is the chosen channel?
- Will all the right people have access to that channel? Is there an address book of those people?
- Do third parties need to know about this channel?
- Do the primary and secondary channels share a dependency?
Regulated entities may need to consider whether they will have adequate records of those communications, especially for customer interactions involving regulated products and services. At the very least, they need a clear and ideally predetermined decision on the trade-off between potential customer harm and record-keeping (unless their regulator has already set out these boundaries).
Restoration versus recovery
Continuity events do not end when the resources themselves are restored. Some services were disrupted for 24 hours or more, unable to act for a full business day because incoming orders and requests had not arrived. Those services may then have been flooded with emails, which can take hours or even days to clear.
A related problem is matching requests originally sent by email with requests repeated through other channels. Once the backlog arrives, some organizations may need to spend time sorting through those requests to avoid duplication.
Next steps for your organization
Treat this outage as a free test while the detail is fresh. Check where email and other shared tools sit in your dependency maps, confirm your plans name the alternative channels and who activates them, and plan for the backlog as well as the restoration.
Protecht’s operational resilience and business continuity capability helps you map critical operations to the resources and providers they depend on, test severe but plausible scenarios and evidence your recovery capability in one connected platform.
See how it works:
* In this article, we have used the terms material service provider and critical operations. If applicable, adapt these to your local regulatory context, such as important business services and critical third parties.


