What Microsoft's preliminary report says
Microsoft has put out its preliminary post-incident report (PIR) on the Microsoft 365 outage that ran from 4:20pm UTC on 12 November to 4:10pm UTC on 13 November. The report is dated 15 November, and admins can read it in the Microsoft 365 admin center under incidents EX1550543 and MO1550821, according to Office Watch's summary. In New Zealand time, the disruption started before the working day on Friday 13 November and finished at about 5am on Saturday.
The report puts the failure down to a fix Microsoft wrote in July 2026 for the networking gateway servers that carry Microsoft 365 connections. Under some loads, the fix caused lock contention in an internal Windows process. The contention used up CPU and memory until the servers stopped responding. The change went out worldwide over several weeks and then ran for more than three months with no visible problems. Microsoft's lab tests ran on weaker hardware with fewer cores, and the defect gets worse on newer machines with more cores. Microsoft says the defect doesn't affect ordinary Windows deployments, and customers don't need to change their tenants.
Two other conditions lined up with the defect. About 15 hours before the outage, Teams traffic failed over to the primary gateway, which also serves Exchange Online. On top of that, nearly 14% of gateway capacity was already offline because of an unrelated problem the day before. When servers started failing, automated recovery moved their connections onto healthy servers, and those servers then failed as well. Within two hours more than 80% of the gateway servers were unresponsive. The boundaries meant to keep a failure inside one region didn't hold, because of a separate fault.
Recovery dragged because waiting traffic knocked restarted servers straight back over. Microsoft's usual rollback, flighting and feature flag tools also couldn't quickly reverse a change deployed months earlier. Engineers ended up building an update that put the connection-handling code back to its earlier known-good state, then patched and restarted the servers in large batches. Teams, SharePoint, OneDrive and Office apps could move their traffic off the faulty gateway and recovered sooner. Exchange Online couldn't make that move at its scale, so email was out longest.
Why the root cause matters more than the outage
New Zealand felt the outage directly. The NZ Herald reported that property settlements were badly disrupted because law firms couldn't receive bank drawdown details. Fletcher Building couldn't use its electronic signing system, and a Hamilton District Court teleconference didn't go ahead. Most commentary at the time was about backup channels. The PIR raises a harder question for IT teams about how they manage change.
Most change processes we see in NZ organisations assume a bad change shows itself quickly. The post-implementation review closes the change record within days, and the backout plan assumes the previous version is one step away. Microsoft's July change passed testing and ran cleanly for a full quarter. When it finally failed, the fix was to re-engineer and redeploy across the fleet. During a major incident, the service desk's first move is to look at recent changes. A change from July wouldn't have been on that list.
The trigger matters for problem management too. Microsoft says the remaining gateway capacity would have coped on its own. The outage came from a dormant defect meeting a failover and reduced capacity at the same moment. Smaller environments fail the same way. A firmware update on a firewall pair can sit unnoticed until one unit dies and the survivor takes the full load. A database patch can wait for the end-of-month batch run. Root cause analysis that stops at "what changed this week" will miss these failures.
Test fidelity is the other uncomfortable lesson. Microsoft has far more test hardware than any New Zealand business, and its lab still didn't match production. Mid-sized firms usually test on an undersized virtual machine or a development tenant with a handful of users. That's enough to check functionality. It tells you very little about behaviour under real load.
The timeline in the report also shows how long a large operator took to stop making changes. The high-priority alert fired at 4:20pm UTC, and deployments and configuration changes were paused at 7:00pm UTC. Many NZ firms have no written rule for when a degraded service or an active failover should freeze other changes. Without one, the decision gets made under pressure, often too late.
What's still unknown
Microsoft said it would publish the final PIR, with a full root cause analysis and longer-term repair items, by 20 November. Neither the incident record nor the public summaries we checked had confirmed that the final report was out. Microsoft's own commitment is a preliminary PIR within two business days and a final PIR within five.
Compensation is also unresolved. Microsoft's Australia and New Zealand spokesperson declined to comment on compensation. Craig Young of the Information Technology Users Association told the NZ Herald that customers should expect compensation, "especially if any loss can be clearly shown." Firms planning a claim will need their own records of the impact on their business.
Next steps
- Download Microsoft's PIR from Service health history in the admin center, or through the Microsoft Graph incident report endpoint that 365Fanatic describes. Attach it to your own incident record, and do the same with the final version when it's published.
- Run an internal review of 13 November. Record when your team first noticed the problem, how staff were told, and which workarounds people used. MBIE's Glen McCloy warned businesses not to send confidential customer information through personal or unapproved accounts. Find out whether staff did.
- Keep change records searchable for at least six months. Tag changes that touch shared infrastructure such as gateways, firewalls, identity and databases. Add "changes older than 30 days on affected components" to your major incident triage checklist.
- Write a change freeze trigger into policy. Any active failover, lost redundancy or degraded service should pause non-urgent changes on related systems until someone with authority lifts the freeze.
- Load test changes to shared infrastructure on production-like hardware, or with production-like traffic, before rollout. If you can't, roll out in stages with a longer watch period and a tested way to reverse the change.
- Check your MSP and SaaS contracts for PIR obligations. Microsoft's two-day and five-day targets give you a reasonable benchmark to ask suppliers for.
These are standard ITIL 4 change enablement and problem management practices, and most firms have them on paper. The gap is usually in how they're run day to day. If you want a second opinion on your change and incident processes, our ITIL 4 framework service covers exactly this. The Microsoft report shows that a change passing its review only tells you it worked under the conditions you tested.
Sources
- Microsoft 365 outage: what went wrong, in plain English - Office Watch
- Microsoft Reveals How a Hidden Windows Bug Took Down Microsoft 365 Services Worldwide - Cyber Security News
- Microsoft 365 outage 12 November 2026: post-incident report - 365Fanatic
- Microsoft reveals 80% server failure behind global outage - AOL
- EX1550543 incident record - Cloud Admin Hub
- Microsoft 365 outage leaves NZ businesses scrambling as email and Teams fail - NZ Herald
- Microsoft Reveals Root Cause Of Massive Microsoft 365 Outage - Wccftech
Want to know what this means for your organisation? Our team in Auckland can show you your top three opportunities in a free 30 minute assessment.
Get my free assessment How we help