TRENDING

Microsoft Went Down. Africa’s Banks Had No Real Plan B

By: indexprima

November 16, 2026

Image Source: Microsoft offices in Nairobi. Image: TechCabal

Share

When Microsoft’s email service failed on 12 November 2026, banks, airlines and state agencies across Africa lost work email and told customers to use the phone or WhatsApp. Payments mostly kept moving, but the notices showed how many African firms depend on one US company, and on a second one when the first fails.

This deep dive tracks what broke, how long it took banks to speak, what regulators say, why firms stay with one vendor, and what a real backup would look like. The outage was a warning more than a disaster. The next one may land on offices where AI agents do the work that staff did this time.

Here is the short version. The fault began around 16:00 UTC on Thursday 12 November and was fully fixed only in the early hours of Saturday 14 November, by Microsoft’s account. It reached 74 countries. Lenders in Kenya, Nigeria, Zambia, Namibia and South Africa lost email, and so did Eskom, FlySafair and the SABC. The first bank notices we found came about 15 hours after Microsoft’s own first post. Most banks pointed customers to WhatsApp, two of the three channels Wema Bank named belong to Meta, and no regulator we could find commented. Microsoft says it was not an attack and not linked to AI. Its full report is still to come.

To see why this matters, start with what actually failed.

Microsoft’s first public post, at 17:54 UTC on 12 November, said users in the Americas could not reach Exchange Online mailboxes. About 13 minutes later it warned that the problem could be global. One incident record puts the start at 16:17 UTC, and Al Jazeera puts it at about 16:00. Within an hour, the outage tracker Downdetector had reports from 74 countries and territories. Reports peaked at roughly 130,000 to 140,000, depending on the count, and tens of thousands of Outlook reports were still open on Friday morning, one source says. Those counts show people who reported a problem. They do not show how many people were hit. More than 430 million people use Microsoft 365, Al Jazeera notes.

The service hit hardest was Exchange Online, Microsoft’s hosted email. People could not open mailboxes on the web, on phones or on desktops, and mail that did move arrived late. Calendars failed, and so did Teams meeting invites and some Copilot answers that draw on email. Teams calls and chats, and SharePoint files, mostly kept working, according to Tech-ish. Azure showed no issues. Microsoft extended its mail retry window to 48 hours so that messages sent on Thursday night would keep trying until Saturday night.

Here is the timeline as reported, in UTC.

When What was reported
12 Nov, about 16:00 to 16:17. Problems begin.
12 Nov, 17:54. Microsoft’s first public post, covering the Americas.
12 Nov, about 18:07. Microsoft warns the problem could be global.
12 Nov, about 18:00. Downdetector reports peak at 130,000 to 140,000.
13 Nov, 04:36. Partial recovery in some regions.
13 Nov, 09:22. I&M Bank in Kenya posts its first notice we found.
13 Nov, 13:32. Broader gains in North America and Europe.
13 Nov, 14:17. Mail queues near normal.
14 Nov, early. Microsoft says service is fully restored.

The cause is still open. Microsoft’s preliminary account blames “an issue during an earlier network capacity upgrade that surfaced following a recent network change.” In plain terms, routine network work exposed a dormant bug. Requests queued up, the queue grew, and the servers that set up connections to Exchange Online ran out of capacity. They then failed automated health checks and dropped out. That pushed more load onto the rest. Microsoft says total traffic had not risen. Engineers restarted servers and rolled back a network change. Microsoft says the issue was “not related to cyber or AI / agents,” and its executive Charles Lamanna wrote that there was no sign of malicious activity. A full report, with root cause and what Microsoft calls “defense-in-depth actions,” is promised.

Outside experts offered a similar picture. Walter Pasquarelli told Al Jazeera that servers failing health checks can push traffic onto others until they overload. Rupert Goodwins also offered a view to Al Jazeera. Those are views, not findings. Only Microsoft’s final report will settle the cause.

Africa felt it from Friday morning. Reports rose between 8 am and 11 am local time, just as offices opened, TechCabal’s Kenn Abuya writes. The notices came in steadily.

Here is what we found, by country.

Country Who said what
Kenya. Equity Bank warned of a “temporary disruption to its email communication services.” I&M Bank said its email was unavailable “due to a global Microsoft email outage.” Loop said email and in-app support were down. Standard Chartered, Stanbic and UBA Kenya also warned customers, and NCBA and Absa Kenya reportedly did, though one report could not confirm that.
Nigeria. Wema Bank said it was “currently unable to access incoming emails.” Stanbic IBTC said email replies would be delayed and its digital banking stayed available.
Zambia and Namibia. Stanbic Bank Zambia and Standard Bank Namibia said they could not receive or answer email.
South Africa. Eskom lost email and pointed customers to phone, WhatsApp, chatbot and USSD. FlySafair could not send or receive email. Some Absa staff had trouble with Microsoft apps. SARS warned taxpayers that its services could be affected. The SABC was also hit.

Two points stand out. First, nearly every notice said the same thing about the money. Payments, apps and cards were not reported down. None of the Kenyan lenders reported a payments failure. Stanbic IBTC and Stanbic Zambia said banking services stayed available. That is good news, and it should be said clearly. Second, the gap sat in the support layer. Loan requests, statements, complaints, and instructions to relationship managers move by email. When that goes, a bank can still pay but cannot always talk.

One columnist described the human side. Abuya was at Avenue Hospital in Nairobi on Friday morning when the screens went blank, he writes. Insured patients needed approval from an insurer whose systems also run on Microsoft, so they waited. That is one account, and we cannot confirm which insurer or system failed. It shows how a bank’s email problem can spread to hospitals and insurers who rely on the same stack.

Now the time gap. By our sums, the first notice we found from a Kenyan lender came 15 hours and 28 minutes after Microsoft’s first post. I&M posted on X at 12:22 pm Kenyan time on Friday. Loop texted customers at 1:26 pm, which is 16 hours and 32 minutes after Microsoft’s post, and I&M followed with an SMS at 1:36 pm. Microsoft’s first post came at 8:54 pm Kenyan time on Thursday. Banks may have sent private messages to key clients earlier, and we could not check that. But the public notices we found came the next day. A rehearsed plan, Abuya argues, would have moved faster. Nothing in the reports we read shows when any bank’s own team first knew.

Notice also what the banks offered. Equity, Wema, I&M and others pointed to WhatsApp. I&M gave a WhatsApp number for its contact centre. Wema offered phone lines, WhatsApp and Instagram. Eskom, FlySafair and Equity named WhatsApp too. The channels are official and well run, and on Friday they worked. But WhatsApp and Instagram belong to Meta. A bank that moves from Microsoft email to Meta messaging has swapped one US platform for another. It has not removed the risk of a platform failing. If both were down on the same morning, many of these banks would have had only the phone.

There is a second layer. Absa’s contact centre uses Amazon Connect, TechCentral reports, so that part of its chain sits with a third US cloud firm. We do not know which tools Absa used on Friday, so this is only a reminder to map every link. A phone line is only a backup if its own systems are separate.

Why do firms run without a true plan B? The answer is mostly money, and two views help. Abuya argues that a second email system, tested and staffed, costs money every month and pays off perhaps one day a year. He adds that a shared outage protects reputations. When several banks fail together, customers do not leave, and no executive loses a job. A bank that builds its own backup and watches it fail takes the blame alone. These are his arguments, not data.

Industry experts voiced similar worries after the Amazon Web Services outage in October 2025. BMIT’s Geerdts told ITWeb that copying data across regions inside one provider is likely the cheapest option. A second cloud provider adds redundancy, but he said it could roughly double costs. He also noted that real-time data sync is hard, and he said firms should check the backup plans of the collaboration and payment tools they rely on. Check Point’s van Rensburg urged tested failovers. The market is concentrated. Amazon held about 30% of world cloud infrastructure in the second quarter of 2025, Azure about 20% and Google Cloud about 13%, by Synergy Research Group’s count.

Is this a new problem? No. In July 2024, the CrowdStrike fault hit African firms too. Capitec and Absa in South Africa lost services, and Kenya Airways reported a booking system outage, according to Reuters and Ahram Online. In October 2025, AWS failed, and IDC said banks, insurers, government departments and retailers using it saw interruptions. BMIT said Standard Bank was directly affected. Microsoft 365 itself has had a rough year. BusinessDay reports interruptions in January, July and late August, and one outage tracker lists a 16 March incident with hours of downtime. Al Jazeera adds a North American outage of almost 10 hours earlier in 2026. Abuya says Downdetector counts this one as the largest of 2026 across all monitored services, but we found no second source for that.

What do regulators say? In the reports we read, none of them commented on the outage. That includes the central banks of Kenya and Nigeria and South Africa’s Prudential Authority. Silence is not proof of inaction, but it leaves a question. Earlier statements show where each stands.

In Nigeria, the Central Bank’s Rakiya Opemi Yusuf told a banking conference in Abuja in September that cyber risk can no longer be handled bank by bank. She described a “one-fire” effect, where one weak institution, vendor or cloud operator can spread trouble through a connected system. She urged banks to test whether technology partners can withstand failures, to know where critical data sits and what can be taken out of it, and to measure resilience by how fast critical services recover. Those are the right questions for this week. The CBN also issued a data localisation circular on 15 June 2026, which takes full effect on 1 January 2027. It requires payment transaction data, including backups and disaster recovery copies, to be stored and managed in Nigeria. It expects geographic redundancy and written continuity plans. Its FAQ does not address email, and it does not mention exit plans. The rule is useful, but it would not have prevented this outage, which hit email.

In South Africa, a 2025 Prudential Authority report on third-party risk calls concentration risk “unavoidable and multidimensional” and says it may create systemic risk. It finds that most institutions have exit strategies but notes “persistent gaps” in contingency planning and incident management. Medium and smaller banks lean more heavily on a few key suppliers, it says. Separately, a July 2025 joint communication from two regulators set best practices for cloud and offshoring, such as due diligence and board-approved data plans. It is not mandatory, and a joint standard is still to come, with no date given. For Kenya, we did not find a recent central bank statement on cloud concentration.

Many people ask whether a local data centre would have helped. The Kenyan example suggests it would not on its own. In May 2024, Microsoft and the Abu Dhabi firm G42 announced a $1 billion geothermal data centre at Olkaria, meant to host Microsoft’s first Azure region in East Africa. Techeconomy, citing Bloomberg, says the companies asked Kenya to guarantee payments for part of the capacity, and the government could not meet the level they wanted. Power was another problem. The site would need about 1 gigawatt, a third of Kenya’s installed capacity of 3,000 to 3,200 megawatts. President William Ruto said, “To switch on that one data centre, we would need to shut off power for half the country.” By May 2026, an official said the project was not dead and was being restructured, with a 100-megawatt start under study. Local hosting has value for speed and rules on data. But Microsoft has not said where the failing servers sat, and the cause it describes sits in network and server logic. As Abuya puts it, where the servers sit was never the problem. Depending on one company is.

There is a bigger worry, and it is about the next outage. AI tools now sit inside the same accounts that failed. Copilot Chat failed to load during this outage, and Microsoft says the cause was not AI. Absa shows how far some African banks have gone. In its first-half technology update, as reported by TechCentral, more than 30,000 staff use Copilot monthly. About 4,400 staff with Copilot Pro licences build automation agents, and more than 3,800 such agents run in production. Its customer chatbot handles over 100,000 queries a month. The report does not say what these tools replace. Banks elsewhere are moving the same way. Abuya’s point stands. Africa got through Friday because people answered WhatsApp messages on their own phones. If an insurer moves claims onto an agent running in Microsoft 365, he warns, a similar outage would stop claims entirely, and the people who once did that work by hand may be gone. That is a forecast, not a fact. But the logic holds. The more work software does, the more a failure matters, and the fewer people remain to step in.

So what would a real plan B look like? Our reading, drawn from the experts above and from the gaps in Friday’s notices, has six parts.

First, name the people who can decide. A plan needs an owner who can say, within an hour, that the vendor is down and the fallback is on. On Friday we cannot see who made that call, or when.

Second, build channels that do not share a vendor. A fallback email route outside Microsoft, a phone tree, SMS to key clients and a staffed call line do not depend on the failed service. BusinessDay’s list of a secondary email route, offline copies of key documents and a second meeting tool is a fair start. A single social platform is not a backup if the plan only moves traffic from one US firm to another.

Third, pre-write the message. A notice that names the cause, says what still works and gives two ways to reach the bank can go out in minutes if it is ready. Only I&M named Microsoft among the notices we saw. Customers value plain words.

Fourth, protect customers from scams. Sophos has warned that criminals use big outages for phishing. Banks should tell customers to use only numbers and pages they already trust, and not to click links in messages about the outage.

Fifth, test it. A plan that has never run is a document. Banks should rehearse the loss of email for a full morning, at least once a year, and record what broke.

Sixth, count the cost honestly. A second email system adds monthly cost. Cross-region copies within one cloud are cheaper than a second cloud, BMIT says, and a second provider could roughly double cost. Boards should weigh that against what a long outage costs in delayed loans, missed trades and lost trust. Regulators can help by asking for evidence of testing, not only a filed plan.

Why it matters: Africa’s banks run more of their daily work through a few US platforms than most customers realise. This outage hurt little because payments held and staff improvised. That luck will not last as agents take over desks and more services move to the cloud. The lesson is not to leave Microsoft or Meta. It is to know what you depend on, to have a fallback that does not depend on the same company, and to practise using it. Regulators from Lagos to Pretoria have already said concentration risk is real. This week showed the cost of leaving it untested.

What to watch next: Microsoft’s full report on the root cause and what it will change. Watch whether the central banks of Kenya, Nigeria and South Africa ask banks for outage reports or for proof of tested fallbacks. Watch how many banks publish what they learned. Watch whether Nigeria’s 1 January 2027 localisation rule changes how banks set up email and backups. Watch the Kenya data centre talks. And watch whether any bank or insurer slows its move to AI agents until it has a plan for the day the software stops.


Sources: TechCabal, “The Next Wave: Microsoft’s outage showed Africa’s banks had no real plan B”, 15 November 2026; Al Jazeera on the outage, 13 November 2026; Webhosting Today on Microsoft’s preliminary cause, 16 November 2026; The World Data on outage figures; Tech-ish on I&M Bank and Loop, 13 November 2026; BusinessDay Nigeria on Nigerian banks; BusinessDay Nigeria on transaction risks in Africa; Nairobi Business Monthly on Kenyan banks; Geeky Nigeria on lessons for Nigerian banks, 13 November 2026; BusinessDay South Africa on the outage, 13 November 2026; Afropolitain on South Africa’s dependence on Big Tech; SARS notice on the Microsoft outage, 13 November 2026; ITWeb on cloud diversification after the AWS outage; Arise on the CBN’s cyber warning, 10 September 2026; CBN data localisation FAQ; Baker McKenzie on South Africa’s cloud communication; Prudential Authority third-party risk report; Techeconomy on the Microsoft and G42 Kenya data centre, May 2026; TechCentral on Absa’s AI use.

 

A note for startup founders: Don’t practice on investors. Practice on AI first. RaiseReady by Index Prima is an AI fundraising simulator for African founders: upload your pitch deck, get a readiness score and face a tough AI investor before the real meeting. Find this post, and every post we publish, at raiseready.indexprima.com.

🕶 Relax!

Put your feet up and let us do the hard work for you. Sign up to receive our latest Intel directly in your inbox.

We’ll never send you spam or share your email address.
Find out more in our Privacy Policy.