
If you don't tell your outage story, someone else will
Microsoft, the FAA, Chase and United all had outages in one week. Only one explained itself. Here's how to tell your own outage story when it's your turn.
If you don't tell your outage story, someone else will
On this page
It’s been a week in the tech world. Microsoft 365 had an outage on Thursday. Flights were grounded nationwide on Friday. On Monday your banking app told you some services weren’t available, and if you were flying United that afternoon, you sat at the gate a while longer. Somewhere in there a relative sent you a headline asking if America’s tech is falling apart.
I get why it feels that way. Four high-profile outages in five days looks like a pattern, and media coverage leaned into it. The AP asked “Could everything be broken?” CNN noted that Chase’s outage was “the third straight weekday with a massive technical failure.” NBC News’s morning newsletter just asked, “What’s with all the tech glitches?”
Having spent the majority of my career commanding incidents like this, I’m a little less worried than the headlines. These were four separate incidents that happened in the same week. What made it feel like one crisis had less to do with the technology and more to do with who explained what happened, and who didn’t.
That’s the lesson for folks in this industry, even if none of these were your outage. The next one might be. When it is, the story about it gets written in the first few hours, with you or without you. So I want to walk through what each organization actually said, what these outages really have in common, and what you can set up now so your company tells its own story when it’s your turn.
What actually happened
Here’s each outage, in the words of the organization that had it. Notice how much the last two columns vary.
| Outage | When | How long | What they said caused it | What they published |
|---|---|---|---|---|
| Microsoft 365 (Exchange Online) | Thu Nov 12 to Fri Nov 13 | 23 hours, 50 minutes | A latent bug in a July 2026 update to their networking gateway servers, set off by extra Teams traffic and capacity that was already offline | A 7-page preliminary Post Incident Report (dated Nov 15), with a final report promised for Nov 20 |
| FAA nationwide ground stop | Fri Nov 13 | A bit over 4 hours, more than 12,000 flights canceled | A flood of connection attempts on their telecom network (FTI) during router work by contractor L3Harris. Root cause still under investigation | A Q&A page (Nov 15) and press briefings |
| JPMorgan Chase | Mon Nov 16 | Monday morning until about 6:30pm ET | “A software bug in an on-premise hardware system,” not cyber-related | Short statements on X and to reporters |
| United Airlines | Mon Nov 16 | About 90 minutes | “A technology issue that delayed flight departures” | Short statements. The detail came from a pilots’ union memo instead |
Microsoft and the FAA at least told us something about the mechanism. Chase gave us one sentence. United gave us less than that. Keep that spread in mind, because it explains most of what happened next.
Same week, different causes
I’ve seen this movie before, just on a smaller screen. In one incident I had two services start throwing errors at the exact same moment, and almost everyone on the call assumed one cause. I knew those services shared very few dependencies, and their failures didn’t look alike. So I split them into two incidents with their own responders and cross-linked them. They turned out to be two separate problems.
My rule since then is simple. Starting at the same time doesn’t mean sharing a cause. What would have changed my mind that day was a shared dependency throwing its own errors, matching error signatures, or one change that touched both. Applying that test to this week, none of the four outages passes it:
- Different systems. Microsoft’s problem was in their own gateway servers. The FAA’s was in a telecom network that L3Harris owns and operates. Chase pointed to a bug in an on-premise hardware system. United’s pilots’ union pointed to the airline’s weight and balance system. Nobody has named a shared vendor, network or dependency.
- Different clocks. Microsoft’s bug had been sitting in production since July. The FAA’s Transportation Secretary said a system in Amarillo that L3Harris had replaced about three weeks earlier also started flooding the network. Things that break in the same week usually started months apart.
- Not an attack. Microsoft said throughout that they weren’t aware of any malicious activity. Chase said there was no evidence their issue was cyber-related. United said there was no indication of a cybersecurity issue.
Even the press ended up walking back its own framing. The same AP story that asked whether everything could be broken called the official explanations “far more prosaic.” CNN pointed out that the incidents don’t share the same scope or root causes, and NBC News wrote that the four appeared to be unrelated.
Big, complicated organizations (let’s be honest, all organizations) have incidents constantly. Most weeks you just don’t hear about four of them at once.
What they do have in common
None of these outages caused another. But if you’ve been doing incident response for a while, a few things in these stories should look pretty familiar. If any of these had been my incident, here’s what I’d want to dig into in the retrospective:
- Changes were at the center. Microsoft’s trouble started with a July update to their gateway servers. The FAA’s started during router work at a remote site in El Dorado, Arkansas, part of a copper-to-fiber conversion L3Harris says is more than 70% done. None of that is surprising. Google’s SRE book says roughly 70% of outages come from changes to a live system. I wouldn’t spend any time on who made the change. I’d ask why one router at one remote site could destabilize radar and radio feeds across the country.
- The bug needed the right conditions. Microsoft’s defect only showed up on high-core-count hardware, and their load testing focused on machines with fewer cores. It went unnoticed for about four months. Then three things lined up on the same day: extra Teams traffic, about 14% of gateway capacity already offline, and the bug. Take any one of those away and maybe there’s no outage. Sidney Dekker calls this drift into failure. Nothing looked broken until the margins ran out.
- Failover moved the problem instead of removing it. Teams failed over automatically onto the gateway Exchange Online also used, which added the load that tipped it over. Microsoft’s automated recovery then made things worse. And Exchange Online, unlike Teams, couldn’t fail out to the second gateway at all. Redundancy you can’t use for every service is redundancy with an asterisk. Chase showed the same thing from a different angle: other banks were affected only on transactions involving Chase. Your blast radius doesn’t stop at your own front door.
- New layers on old foundations. The FAA is mid-migration from copper to fiber, which means old and new equipment running side by side. I lived a smaller version of this at Braintree, where my first major incidents happened in the middle of a move from two on-prem data centers to AWS. We had data center folks on the bridge who were physically standing next to the racks. Legacy tech on its own rarely breaks things. The seams between old and new are where I’d look first.
In my view, humans don’t cause incidents. The systems that let the incident happen do. All four of these organizations are running complex systems built up over decades, and every one of them will have another bad day. The part they control is how they respond, and how they talk about it.
Microsoft told their own story
Of the four, Microsoft had the longest outage by far. Nearly a full day without email for a huge chunk of the working world is a big deal. And yet I’d argue they came out of the week in the best shape, because they’re the only ones who explained themselves.
Here’s what they did:
- They said they were looking into it, and kept saying so. Updates went out on their status page while the incident was live. They said early that they saw no sign of malicious activity, which shut down the cyberattack theory before it got going. And they promised a post-incident report before the outage was even over.
- They kept that promise. The 7-page preliminary report went out to admins the next day. It has a timeline, three contributing conditions and the fix (a patch that reverted the defective change, deployed early on Nov 13).
- A named person owned it in public. Charles Lamanna (EVP of Copilot, Agents, and Platform) posted an abridged version on X, with an apology and the reason Teams recovered sooner than email. A real face explaining what happened builds more trust than 27 status page updates.
It wasn’t perfect. Microsoft’s early explanation pointed at a network capacity upgrade, and the preliminary report tells a more detailed and somewhat different story. The full report also sits behind an admin login, so most people only saw the X thread and the trade press. But look at where the coverage went. Outlets like Office 365 for IT Pros and The Register spent the weekend explaining lock contention and failover, not asking whether AI is taking over the world. Microsoft wrote the first draft of their own story, so everyone else ended up quoting them.
For what it’s worth, the bar here isn’t new. Cloudflare’s CEO published its Nov 18, 2025 retrospective the same day as the outage. Fast, specific and signed by someone people know beats careful and anonymous every time.
Same week, different headlines
If you want to see what happens when you don’t control the narrative, put the headlines side by side. Microsoft’s coverage was about their bug. Everyone else’s coverage was mostly about somebody else’s outage.
| Organization | What they put out | Headlines that followed |
|---|---|---|
| Microsoft | A 7-page preliminary report and a named executive explaining it on X | The Register: “Microsoft blames monster email outage on software defect introduced in July fix” · Network World: “Microsoft outage: A cascading failure that started with an undetected bug” |
| FAA | A Q&A page and press briefings, no technical writeup | CNBC: “‘Root cause’ of nationwide ground stop is unknown, FAA says” |
| Chase | One sentence on X | The Daily Beast: “U.S. Hit by Another Massive Outage as Major Banks Go Down” · The Jerusalem Post: “JPMorgan, Wells Fargo among main banks reportedly down according to users” |
| United | “A technology issue” | CBS News: “United Airlines says it resolved a ‘widespread system outage,’ days after nationwide ground stop” · TIME: “United Airlines Experiences Temporary System Outage in the Wake of Nationwide Ground Stop” |
Microsoft had the longest outage of the four and got the most boring headlines, which is exactly what you want. (They still got pulled into the roundups, but their own stories were about the bug.) The FAA’s headlines were about what nobody knew yet. Chase’s turned one bank’s bug into “major banks” going down. United’s outage lasted 90 minutes, and both of their headlines are really about the FAA.
Everyone else got explained by somebody else
The other three left gaps, and the gaps got filled fast.
The FAA deserves some credit. Their Q&A page separated what they knew from what they didn’t, and the FAA Administrator told reporters, “We couldn’t explain the instability.” That’s an honest update. But with no technical writeup, the story moved to press conferences. The contractor was named and blamed within a day, the President described it as old equipment that had a heart attack, and questions about a new AI system had to be knocked down separately. All of that happened before anyone finished a post-incident analysis. Pointing at a vendor before the analysis is done is something I’d never put in an update, no matter who the vendor is.
Chase explained themselves in one sentence on X around 4pm, after customers had been seeing errors since that morning. Meanwhile Downdetector lit up for Wells Fargo, Bank of America, U.S. Bank and others, so several of those banks spent the afternoon telling reporters they were fine, and Zelle had to say their network wasn’t down. When I was at Chime we had great monitoring and we still checked Downdetector during big incidents, because the first question is always “is it just us, or is it everyone?” When the bank at the center goes quiet, everyone around it has to answer that question for it.
United said “technology issue” and not much else. So the most detailed public account came from a pilots’ union memo about weight and balance calculations. Operationally, United recovered in about 90 minutes and delays were back in line with competitors soon after. That’s not a bad day by airline standards. But with no story of their own, their outage got filed under “days after the FAA ground stop,” and became item four in every “what is going on?” roundup.
To be clear, I know why companies go quiet. It’s also only been a day for Chase and United, and Microsoft’s preliminary report took three. But the version of the story that sticks gets written in the first few hours. I’ve spent most of my career in regulated fintech. Banks and airlines have real legal, regulatory and security reasons to slow down (public companies face SEC disclosure rules, for starters). Got a legal or PR team pushing back? Write comms that are honest enough to push right up to the line. If they don’t tell you “we can’t share that” at least once, you didn’t push far enough. Find the line and toe it every time (but don’t go over it, because there is a fine line between good PR and asking for litigation).
Silence doesn’t keep your outage out of the news. It just means someone else writes the story.
Before your own bad week
None of this needs a crisis to set up, and most of it is impossible to set up during one. Here’s where you can start:
- Post “we’re aware” in minutes, not hours. Even “we’re aware of an issue and looking into it” tells people they’re not imagining things. If your status page is down too, use socials or a banner in the product.
- Separate what you know from what you don’t. “We haven’t identified the cause yet” says more than “we’re looking into it.” Before anything goes out, ask: if our leading theory turns out to be wrong, is this update still true?
- Say what it isn’t as soon as you know. All four organizations said there was no sign of a cyberattack, and that one sentence did more to calm things down than anything else they said.
- Update on a schedule. For a high-severity incident I’d go every 15 to 30 minutes, and end every update with when the next one is coming.
- Get templates approved by legal now. Pre-approved language is general enough to cover most scenarios and specific enough to say something real, and nobody has to wait on a review at 3am.
- Decide who signs the public retrospective. A CTO or a well-known engineer, picked before you need them. Promise a date for the writeup and hit it, the way Microsoft did.
- Leave blame out of it. No “human error,” no “our vendor,” no “unprecedented.” Customers don’t care whose fault it is while they’re locked out, and you’ll probably be wrong about it anyway this early.
The story is part of the response
The week wasn’t one crisis. It was four bad days at four different organizations, and the team that explained themselves got to keep their own story.
Nobody has 100% uptime, and everyone reading your status page this week has had a bad day at work of their own. In my experience they’ll forgive the outage a lot faster than the silence. Incident management doesn’t end when the service comes back. Telling people what happened is part of the job, and it’s the part you can start practicing before anything breaks.
At Rootly we build tooling for status pages and retrospectives, but you can do every item on that list with a shared doc and some discipline (and if you want help tightening up your incident comms, shoot me an email or DM me on LinkedIn).




