How One Outage Split a $16 Billion Bill Very Unevenly
5 Min Read

On Thursday, 12 November, Microsoft hit a bit of a road bump. Within two hours more than 80% of the networking gateway servers sitting in front of Microsoft 365 were unresponsive, and Outlook went dark for users in 74 countries. Microsoft only declared active customer impact over at around 18:10 on Friday, and the email backlog wasn't fully cleared until Saturday morning.
Locally, SARS put out a notice warning taxpayers that its own Outlook and Teams might be affected, the City of Cape Town said it had no timeline for restoration, UCT's ICT team warned staff and students about connection problems, and Richemont briefly delayed its interim results. Further afield, Euronext and Maersk told customers they were having communication problems, and some businesses reportedly went back to fax machines (in the big 2026) to confirm trades.
But what we really want to know, is what does an outage like this cost. Insurers expect to pay somewhere between $100 million and $700 million of it, and Microsoft's own liability is capped at a credit against next month's licence fees. More on that below.
P.S. We have a resource hub with articles programmes, bursaries, VAC work and other resources. Access it here 👇
🧠 Fun fact: The fax machine is older than the telephone. Scottish clockmaker Alexander Bain patented his "electric printing telegraph" in 1843, 33 years before Alexander Graham Bell patented the phone. Which means that on Friday, some of the world's most sophisticated trading desks were backed up by Victorian technology, and it worked.

Wall Street Oasis' Elite Modeling & Valuation Certification: six courses covering Excel, financial statements, DCF, valuation, M&A and LBO. 100+ lessons, self-paced → Start the certification
The supplier's SOPs failed in a very specific way
Microsoft's post-incident report is unusually candid.
The root cause was a bug introduced in a July update to the gateway servers. It ran for more than three months because it only triggered under a particular traffic pattern.
On the day, two things lined up to create that pattern: roughly 15 hours earlier, Teams had automatically failed over onto the infrastructure Exchange Online (Microsoft's cloud-based email and calendaring service) uses, adding load, and nearly 14% of gateway capacity was already offline for an unrelated reason. Microsoft says the remaining capacity should have coped on its own. But as we all know it didn't once the extra traffic arrived.
What turned a bad afternoon into a lost day though?
The safeguards meant to stop failures spreading between regions didn't contain it, and the automatic failover moved traffic onto healthy servers, which then failed too. When engineers restarted servers, the backed-up traffic flattened them again. Other workloads, Teams and SharePoint among them, could switch to alternative infrastructure and recover sooner, but Exchange Online had no such escape route. Microsoft also admitted its load testing ran on less powerful hardware than production, which is why the bug was never caught.
For anyone who has sat through a lecture on general IT controls, this is change management and capacity planning failing together. The fix list (a patch for the bug, more capacity, better monitoring, and safeguards so traffic redirection can't cascade) is essentially an admission that the SOP for "a region is unhealthy, move the load" had never been tested against "most regions are unhealthy at once".
The customer's SOPs are the ones that mattered
Microsoft can't promise you it won't go down. Its SLA commits to 99.9% monthly uptime, which in a 30-day month is about 43 minutes. Everything beyond that is your problem to manage.
The organisations that coped on Thursday were the ones with a fallback someone had written down in advance and descriptive continuity procedures. That doesn't need to be elaborate. It's knowing which clients and counterparties get a phone call instead of an email and which internal channel the team moves to, agreed before anyone needs it. It's also a short list of what can't wait (a payment run, a SARS submission, a results release, a client deadline) along with who owns each item if the usual tools vanish. Richemont delaying its results by a few hours is a mild version of what happens when a hard deadline meets a dead inbox.
Most companies just let everyone have an extended-extended-extended coffee break…
Auditing standards already assume this split of responsibility. ISA 402 calls them complementary user entity controls: the controls a service organisation designs its system around, on the assumption that the customer has implemented its own side. When a client points to Microsoft's SOC report as comfort over its cloud environment, that report is only half the picture. If the client has no continuity arrangements of its own, the design the service auditor tested was never complete in the first place.
So who pays for the day?
Let’s start with Microsoft. Its SLA credits run on three tiers: 25% of that service's monthly fee if uptime drops below 99.9%, 50% below 99%, and 100% below 95%. A tenant that lost Exchange Online for a full 24 hours in November lands at about 96.7% uptime (1,440 lost minutes out of 43,200), which earns the 50% tier. Credits apply only to the affected service, aren't paid automatically, come off future invoices rather than as cash, and have to be claimed by the end of the following month. The SLA also states plainly that they don't compensate for lost revenue or other indirect losses.
In accounting terms, that's tidy for Microsoft. A service credit is a reduction in the price of the service, so under revenue recognition it's variable consideration rather than an expense. Microsoft reports under US GAAP, but IFRS 15 work the same way here: credits and penalties come off revenue. The exposure is bounded by design, and Microsoft hasn't said whether any of its own costs are insured. At its scale, the more likely answer is that it simply absorbs them, though that's my read rather than anything it has disclosed.
Then the customer. If you're claiming a credit, it's a contingent asset until Microsoft accepts it, because IAS 37 doesn't let you recognise income that is only probable. Your lost billable hours are just gone.
Which leaves insurance, and this is where the premium either earns its keep or reveals what it never covered. Cyber policies were the obvious place to look, but that cover only responds where a policy extends business interruption to non-malicious system failure at an outside provider. Some policies only respond to attacks, and Microsoft has been clear this wasn't one. Even where the clause exists, contingent system failure cover typically carries a waiting period of eight to 12 hours before anything pays out. Severe disruption on Thursday lasted about 18 hours, so policyholders with the right wording cleared the bar, and those with the wrong wording got unlucky.
Insurance feels like dead money right up until the week you wish you had it, and for the businesses that had contingent BI on non-malicious failure, last week was that week. For everyone else, it was a reminder that the premium only protects you against the risks that your schedule actually names.
What downtime costs the company that had it
Microsoft's direct costs is the credit it must now give, but the more expensive costs don't show up tangibly on an invoice.
The market barely flinched: shares fell as much as 3% before the open on Friday and closed roughly flat, which tells you how locked-in the customer base is.
The slower cost is trust, and it compounds: this follows other Microsoft 365 disruptions earlier in the year, and every one of them is a slide in a competitor's sales deck. It also landed in the same week as Ignite, Microsoft's flagship conference, where the pitch is that businesses should run more of their work through Copilot and AI agents, which happen to depend on the same Exchange data that just disappeared for a day - plus, Copilot still doesn’t have the best rep in the AI space.
Then there's opportunity cost → Engineering time spent on recovery is time not spent shipping product, but the deeper trade-off is capacity itself. Idle servers held in reserve are an expense with no revenue attached, and every cloud provider is constantly deciding how much headroom to carry. Running with 14% of capacity already offline was a perfectly defensible position on an ordinary Thursday.However, the outage is the price of that decision on the one Thursday it wasn't ordinary, and it's the same calculation any business makes when it decides whether a backup system is worth paying for.
Insights From The Pocket CA
We've partnered with Pocket CA, an AI tool built for accountants, like your friend if they actually listened in lecture.
IAS 37 Provisions, Contingent Liabilities and Contingent Assets
An entity shall not recognise a contingent asset, because doing so could mean recognising income that may never be realised (IAS 37.31 and 37.33).
Once an inflow becomes virtually certain, the asset is no longer contingent and is recognised in the period the change occurs. If it's only probable, it's disclosed (IAS 37.35).
Where some of the cost of settling a provision will be reimbursed by another party, such as an insurer, the reimbursement is recognised only when virtually certain, as a separate asset that can't exceed the provision (IAS 37.53).
The expense and the reimbursement may be netted in profit or loss, but not on the statement of financial position (IAS 37.54).
IFRS 15 Revenue from Contracts with Customers
Consideration can vary because of discounts, refunds, credits, price concessions or penalties, among others (IFRS 15.51).
Where consideration includes a variable amount, the entity estimates what it will be entitled to (IFRS 15.50). Service credits under an SLA sit squarely here.
ISA 402 Audit Considerations Relating to an Entity Using a Service Organisation
Complementary user entity controls are the controls a service organisation assumes its customers will implement, and which may be necessary to achieve the control objectives (ISA 402.8(a)).
A type 1 or type 2 report helps the user auditor understand the controls at the service organisation that affect the client's transactions (ISA 402.A22).
A type 1 report gives no evidence that controls actually operated effectively. Only a type 2 does that (ISA 402.A22).
Check out the → 🔗Pocket CA
The Bottom Line
Microsoft's outage was a testing gap that turned into a recovery problem, and the company has been fairly open about both. What's less discussed is how cleanly the contract shifts the consequences. Microsoft's exposure is a revenue adjustment it can estimate in advance. The customer's exposure is a day of lost work, partly offset by a credit it has to remember to claim and an insurance policy whose wording it probably hasn't read since renewal.
None of that is a reason to leave the cloud, but it is a reason to treat continuity planning and insurance wording as one control, because the SLA has already decided which side of the contract carries the risk.
We’d love to hear what you thought of the post this week, reply to this email with one of the following:
= You suck.
= Pretty good.
= Awesome, more like this.
Until next week,
The Journal Entry Team
If you enjoy our content we’d love it if you showed your friends and peers!
Please copy and paste this link to others:


