This website uses cookies

Read our Privacy policy and Terms of use for more information.

At 16:17 UK time on Thursday 12 November, Outlook users began to find they could not open their mailboxes. Just over two hours later, at 18:26, Microsoft opened a second incident covering every Microsoft 365 service that depends on Exchange Online. That list included Teams calendars, external sharing from SharePoint, Planner, and Microsoft 365 Copilot Chat. Microsoft did not report 'sustained service recovery' until 22:00 on Friday 13 November, almost 30 hours after the first failures.

Microsoft has been clear about what the outage was not. Its statement on Friday evening said: 'This was a technical issue not related to cyber or AI / agents.' The company traced the incident to 'an issue during an earlier network capacity upgrade that surfaced following a recent network change', apologised, and promised a full post-incident report. We could not find a published report when we checked on 14 November.

For project managers, cost teams, contractors, designers, and clients, the lesson is in how ordinary the failure was and how much was sitting behind it. Contract correspondence travels by email. The AI assistant many firms have just paid for lives in the same tenant as the mailbox. On Thursday afternoon both stopped together.

What happened, and when?

The timings here come from azureblog.co.uk, a UK Microsoft admin site that logged each of Microsoft's service health posts as they appeared. Microsoft tracked the event under two references: EX1550543 for Exchange Online and MO1550821 for the wider Microsoft 365 incident. A separate and shorter Teams incident, TM1550488, delayed chat messages in the Americas between 14:45 and 15:53 the same afternoon and was resolved before the main event began.

Microsoft first described the mailbox problem as affecting the Americas, then said it could affect users in any region. Its early updates pointed to 'elevated load'. By the evening it had changed that reading: 'We determined the elevated load isn't driven by an increase in traffic'.

A number of servers had become unexpectedly unavailable, which pushed their work onto the ones that were left. Overnight, Microsoft said a process that sets up secure connections on those servers was being overwhelmed, and it began reverting a recent network configuration change.

Servers that were brought back kept failing once traffic returned. At 06:57 on Friday the company reported only 'marginal success'. It extended the email timeout window to 48 hours so that delayed messages would be held and delivered.

Microsoft has not said how many users were affected. The Downdetector figures differ by snapshot and only count people who chose to report. azureblog.co.uk puts the peak for Outlook at about 130,000 reports around 18:00 on Thursday.

Forbes cites 48,124 reports from US users at 13:21 EST. Al Jazeera reported that the tracker had logged reports from 74 countries and territories within about an hour. The Associated Press noted that Microsoft said in January it had more than 450 million paid Microsoft 365 commercial subscriptions.

What stopped and what kept working

Microsoft's admin centre post said features that 'interact with Exchange Online and email services are impacted', while the parts of each service that do not depend on Exchange kept running. In practice, as azureblog.co.uk recorded it, that meant the following.

  • Mailboxes could not be reached through Outlook on the web, desktop Outlook or mobile mail, and email delivery was delayed.

  • Teams calls, meetings and chats mostly kept working, but calendar scheduling, invites and meeting edits could fail.

  • Files stored in SharePoint could still be opened, though sharing them with people outside the organisation could fail.

  • Microsoft 365 Copilot Chat might not load or answer, and might not reach email, calendar or files.

  • Planner and To Do were on the affected list, and Microsoft Graph requests for mail, calendar and contacts could time out.

The Graph point matters to anyone running automations or third-party tools against a mailbox. The Associated Press reported that OpenAI and Anthropic each said they had problems connecting with Microsoft 365 tools during the outage.

The UK felt it too. According to the same AP report, the British government said it was 'working closely' with Microsoft and had 'robust contingency plans' in place, police forces said they could not receive crime reports online, and the National Pharmacy Association warned of prescription delays because NHS email was affected.

The assistant went down with the mailbox

Nothing suggests AI caused this, and Microsoft says plainly that it did not. The AI point is about dependency. An assistant that drafts from your email, calendar and files is only available while those services are.

❝

Services from email to AI virtual assistants connected to Microsoft can stall for users, grinding many organizations to a halt.

Jake Moore, global cybersecurity advisor at ESET (as told to Forbes)

On a live project that halt has a contractual edge. Notices, payment notices, instructions and RFI responses usually go by email, and many of them carry time limits. Most contracts and appointments say how a notice must be given and when it takes effect. Whether a message that sat in a queue overnight was given on Thursday or on Friday is a question for your contract's notice provisions and for whoever advises you on them. It is worth asking this week, while people can still remember what they sent and when.

Walter Pasquarelli, an affiliate researcher at the Bennett School of Public Policy at the University of Cambridge, told Al Jazeera that organisations should plan for this kind of disruption, and framed the planning question like this.

❝

What is the minimum viable service or the minimum viable product that we need to keep up if there's a global outage?

Walter Pasquarelli, Affiliate Researcher, Bennett School of Public Policy, University of Cambridge

For a project team the answer is short. You need to be able to issue and receive notices, see the current registers, and reach the other parties. It was also an odd week for the AI supply chain more widely: Nvidia's chief executive resigned, and Notion disclosed that customer data had been compromised through a breach at an AI supplier, as we cover elsewhere this week.

The service credit is small. Claim it anyway

Microsoft's Service Level Agreement measures uptime per tenant, per calendar month. Microsoft's own training material gives the credit table as 25% of the monthly fee for the affected service when uptime falls below 99.9%, 50% below 99% and 100% below 95%.

Bar chart of Microsoft 365 service credit tiers: 25% credit when monthly uptime is below 99.9%, 50% below 99%, 100% below 95%

Service credit by monthly uptime. Source: Microsoft Learn, example SLA credit table. Chart: Project Flux.

azureblog.co.uk works through the arithmetic. November has 43,200 minutes, so the average user can lose about 43 minutes before uptime drops below 99.9%. In its worked example, a 400-user firm that lost Outlook on the web for 22 hours would sit at about 96.9% and could claim 50% of its Exchange Online fees for the month.

Three conditions apply. The credit is not automatic: your organisation has to submit a claim, and Microsoft decides whether it is owed. It covers only the service that missed its level.

Microsoft's Partner Center guidance uses the example of a suite customer who had a SharePoint outage and is credited 'only for SharePoint and not the customer's entire plan'. And if you buy through a Cloud Solution Provider, the partner files the claim with the tenant ID, the incident ID and an email from your own domain asking for the credit.

That guidance says claims for services other than Azure must be submitted within one month of the end of the billing month in which the incident occurred. azureblog.co.uk reads the SLA as setting a deadline of 31 December 2026 for this outage. It also quotes the SLA as saying service credits are 'your sole and exclusive remedy' for availability problems, so do not expect the credit to cover the cost of a lost day.

Takeaway

A day without email is survivable. A day without email on the day a notice is due, with no agreed alternative route and the registers locked in the same system, is a different problem. The fix is cheap and mostly administrative.

  • Check the notice provisions in your live contracts and appointments, and agree a second route with the other parties where the contract allows it.

  • Keep an offline or separately hosted copy of the key registers and the contact list, refreshed on a schedule.

  • Agree one out-of-band channel for the project team, and decide who sends the first message.

  • Ask IT or your Microsoft partner to file the service credit claim, quoting EX1550543 and MO1550821, before the end of December.

If you want next week's version of this, the story that touches your tools before your suppliers tell you about it, the Project Flux newsletter is free at projectflux.ai.

All content reflects our personal views and is not intended as professional advice or to represent any organisation.