On Thursday, November 12, Outlook mailboxes in the Americas started failing a little after 11 a.m. Eastern. Within hours, Teams, Copilot, SharePoint and OneDrive were struggling too. Downdetector logged around 130,000 reports from 74 countries and territories. In the UK, some health services went back to paper records, and police asked people to call instead of email. Services didn't really start recovering until around midday on Friday the 13th. You couldn't write that date into a script without someone calling it lazy.
We made a short video about it (above). This is the longer version: what broke, what Microsoft did about it, and the part most teams skip when they plan for outages.
What broke
Microsoft ruled out an attack early on, calling it "a technical issue not related to cyber or AI / agents."
The failure was a chain. Servers failed their automated health checks and dropped out of service. The servers left standing couldn't carry the load. A process that sets up secure connections got overwhelmed, so servers stopped answering requests. When engineers brought servers back, some failed again under live traffic, which pushed the overload onto whatever was left.
When we made the video, Microsoft hadn't named a root cause. It has since said the trouble began with "an issue during an earlier network capacity upgrade that surfaced following a recent network change." In other words, a problem that sat quietly for a while, until something else changed around it. If you've run production systems for any length of time, that pattern will feel familiar.
How they brought it back
The recovery is the most useful part of this story, because Microsoft described it as they went. They rebooted infrastructure, throttled traffic, rerouted around failures and added capacity. None of that is exotic. It's the standard playbook, applied carefully so that freshly restored servers didn't get flattened by all the traffic queued up behind them.
One detail stood out. Delayed emails were held for up to 48 hours instead of being dropped. Mail that couldn't be delivered during the outage was kept and retried. That one decision is the difference between "email was slow on Thursday" and "we lost a day of customer messages."
The part most teams skip
Every company plans for its own outages. Fewer plan for someone else's. Yet most workflows depend on services the team doesn't run: email, an identity provider, a payments API, a model endpoint. When one of those stops answering, the question isn't whether you'll be affected. It's what your workflow does while it waits.
Usually the honest answer is that it hangs. A step calls a service, the service never answers, and the run just sits there. Nobody gets an error, so nobody does anything. Hours later, someone finds a pile of half-finished work and has to work out which parts actually happened.
Three decisions fix most of this. They're cheap to make before an outage and expensive to make during one.
How long should this step wait? Pick a number. "Forever" is also a choice, just a bad one.
What happens when the wait runs out? Retry, fail loudly, or hand off to a backup process. Which one depends on the step. A notification can be retried later. A payment probably shouldn't be retried blindly.
What gets kept so nothing is lost? Microsoft's 48-hour mail hold is a good model. If you can't finish the work now, store it and pick it up when the service is back.
Where Unmeshed fits
Unmeshed handles this at the step level. Every step in a workflow can have its own timeout, and when the timeout runs out, you choose what happens next: retry, fail, or run a backup process. That won't prevent outages. Nobody can promise that, not even Microsoft. What it does mean is that no step is left hanging forever while you wait for someone else's system to come back.
Plan for the outage before it plans your Friday.
Top comments (0)