Phones kept most organisations working when Microsoft 365 went down
Microsoft 365 email was down for about 24 hours from 12 to 13 November 2026. The organisations that kept serving people mostly picked up the phone, told the public which channels still worked, and used tools that didn't depend on Microsoft. Here's what happened, why it spread, and what to set up before the next one.
The full timeline, services and organisations affected are on the outage record. Every figure here links to its source.
One fault in email took calendars, meetings and sign-in codes with it
Microsoft says traffic shifts during routine network maintenance exposed a latent bug in the systems that set up secure connections. New connections couldn't complete, requests piled up and the servers ran out of capacity. 23
Lots of Microsoft 365 leans on Exchange, the system that runs email. Teams needs it for meetings and calendars. 619 Every Microsoft 365 group gets a shared Outlook inbox and calendar beside its Teams, Planner and SharePoint pieces. 620 Copilot reads mail and calendars to answer questions. 622 So when Exchange stopped, everything that needed it stopped too. Teams chat and SharePoint files were reported working.
Fixing it took a day. Microsoft says restored servers "can become unresponsive again once they begin receiving traffic". 614 Kentik measured Exchange traffic about 85% below normal while devices opened up to 70% more connections than expected. 75 Cloudflare saw requests to Exchange triple, which it read as apps retrying. 615
Login codes sent by email turned the email outage into a wider lockout. Businesses lost access to finance and other apps that send a code by email before letting you in. 554
Microsoft's full post-incident report hadn't been published as at 15 November. 23
51 organisations in 13 countries kept serving people, and how
Named by the organisation itself or a news outlet. Filter by country.
What the calm ones had in common
A phone line outside Microsoft. Councils in Sydney, Victoria, Scotland and London, banks in Manila, hospitals in Bristol and Arnhem and traders in Tokyo and Singapore all fell back on the phone. It worked because their phones didn't run through Microsoft 365. NTT East in Japan shows the limit: its lines rang, but the back office behind them was down until evening.
A list of working channels, published early. Redbridge Council in London put every working channel, with numbers and hours, on its website at 9am. Argyll and Bute, East Lothian and Scottish Borders councils did the same. Organisations without a list told people to call back later.
Messaging apps for urgent work. CMC Markets in Singapore used WhatsApp. Argyll and Bute was the only public body found offering residents a WhatsApp text service. Health sector members used WhatsApp and Signal beside paid alert tools.
A second platform that was already set up. Colorado School of Mines already had Google Workspace and sent staff straight to Google Chat and Docs. Basingstoke and Deane's online forms kept working because they don't run on Microsoft.
Switching to another Microsoft tool rarely helped. Two tools that sign in through the same account system tend to fail together.
Nobody has a credible cost figure for this outage yet
The estimates published so far don't show their working. In Australia they range from "tens of millions" (RMIT's Mark Gregory) 593 to "billions of dollars" (RSM's Ashwin Pal). 592 Morningstar DBRS put insured losses at US$300 million to US$500 million, which is an insurance figure and can't be added to productivity losses. 591
To estimate your own cost, these are the downtime studies that publish a method:
- Uptime Institute: share of major outages over US$100,000, 57%. 595
- Uptime Institute: share of outages over US$1 million, one in five. 595
- ITIC: firms where one hour of downtime exceeds US$300,000, over 90% of mid-size and large enterprises. 596
- ITIC: firms putting one hour at US$1 million to over US$5 million, 41%. 596
- Splunk and Oxford Economics: annual downtime cost, Global 2000, US$400 billion, 9% of profits. 597
- Splunk and Oxford Economics: average annual downtime cost per company, US$198 million. 597
- Splunk and Oxford Economics: annual downtime cost, Global 2000, US$600 billion. 598
- Splunk and Oxford Economics: cost per minute of downtime, US$15,000. 598
- Splunk and Lonergan Research: average unplanned downtime loss, 500 or more staff, 251,000. 599
- Splunk and Lonergan Research: national annual downtime cost, AU$86 billion. 599
- New Relic and ETR: high-impact outage cost per hour (mean), US$1.85 million. 600
- Parametrix: CrowdStrike direct loss, Fortune 500 excl. Microsoft, US$5.4 billion. 602
- Parametrix: CrowdStrike average loss per Fortune 500 company, US$44 million. 602
- CyberCube: CrowdStrike insured loss, standalone cyber, US$400 million to US$1.5 billion. 603
- Lloyd's and AIR Worldwide: economic loss, top cloud provider down 3 to 6 days, US$15 billion (US$11 billion to US$19 billion). 604
Microsoft's big outages keep coming from changes and latent bugs
Root causes in Microsoft's own post-incident reports for earlier outages.
What to set up before the next one
Five staff
- Keep Outlook's offline copy and pin the OneDrive folders people use every day. The email guide shows how.
- Leave Microsoft's sign-in resilience defaults on. They're on by default for every policy. 564
- Write one rule on paper: during an outage, don't sign out, don't change passwords and don't restart unless you must. Most sign-ins during an outage reuse an existing session. 564
- Pick one way to talk that isn't Microsoft. CISA suggests an end-to-end encrypted app such as Signal. 583 Keep an offline contact list of staff, key clients and suppliers. 585
- Get status from outside your own account. A free StatusGator plan covers 3 services. 580
- If email is the business, keep a standby mailbox elsewhere. Zoho Mail's free plan covers up to 5 users 576 and Google Workspace Business Starter is A$9.90 a user a month. 575
Fifty staff
- Everything above, plus two emergency admin accounts kept outside normal sign-in rules. The break-glass guide has Microsoft's steps.
- Look at an email continuity service. Mimecast, Proofpoint Essentials, Barracuda and Hornetsecurity all sell one and none publishes a price. Ask how long it runs, whether it starts by itself and whether you can see mail from before the outage. 572
- Use a status monitor that alerts by text message or a chat tool outside Microsoft. Paid plans start at US$22 a month. 581
- Run one practice session a year with a "Microsoft 365 down for 24 hours" scenario. NCSC's Exercise in a Box is free. 587
Five hundred staff
- Everything above, plus an email continuity service with tested failover.
- Pull Microsoft's service health into your own alerting through Microsoft Graph. 579
- Write and test a formal contingency plan with named backup channels, using the joint guidance from CISA, ASD's ACSC, NCSC UK and others. 584
- If the case stacks up, run a second email platform. Google documents dual delivery with Gmail as the main system. 574 Microsoft has no documented pattern for a standby second tenant.
The free outage playbook walks through each step, with a one-page plan template.
What experts said
- "Resilience shouldn't depend on whether you happen to have a second system." ACS (Dr Prins Ralston) 605
- "technology is great until it isn't" Ai Group (Innes Willox) 592
- "We're so dependent upon these tech companies." University of Melbourne (Andrew Cullen) 606
- "Use upcoming enterprise contract renewal cycles to critically assess vendor concentration risk." Forrester (Balaouras and others) 509
- "It shows how much of the working day now runs through one provider's email" Ookla (Luke Kehoe) 591
- "Enterprises have priced Microsoft 365 as a licence" Greyhound Research (Sanchit Vir Gogia) 609
- "turns an isolated control-plane bug into a systemic societal disruption" University of East London (Prof Rabih Bashroush) 593
- "Millions of organisations have chosen the same handful of cloud providers" University of St Andrews (Dr Stephen McQuiston) 593
Check what's down right now
The live status board reads each provider's own status page. The outage leaderboard shows who has gone down most since 2020.