Most small business owners in India never think about which cloud their software runs on. They find out on the day that cloud has a bad few hours, and the billing screen, the customer portal or the delivery app simply stops loading.
2026 has supplied plenty of examples. This post is about designing small business software so that a cloud outage at one vendor slows you down instead of stopping you. It is the technical companion to our piece on the November Microsoft 365 outage, written for owners and for the developers who build for them.
A year of cloud outages, briefly
A few incidents from this year, kept to what was publicly reported:
- Microsoft 365, January.A problem with part of Microsoft's North American infrastructure caused a disruption lasting close to ten hours, hitting Outlook as well as Defender and Purview.
- Azure, 2 to 3 February. Customers in several regions could not reliably create, scale, start or stop virtual machines. Because so much sits on top of VMs, the effect spread to node provisioning in Azure Kubernetes Service and to Azure DevOps pipelines. Reporting linked it to a configuration change that cut off access to storage hosting VM extension packages.
- AWS CloudFront, 10 February. DNS lookups failed for some CloudFront distributions, so sites and APIs served through it could not be found even though the servers behind them were healthy. The acute phase was short, but follow-on delays took several more hours to clear.
- Microsoft 365, 12 to 14 November. The one most offices noticed, taking down Outlook, Teams and Copilot for many users.
Regulators see the same pattern. In November 2025 the EU's financial supervisors published their first list of critical ICT third-party providers under the Digital Operational Resilience Act (DORA), and AWS and Microsoft were on it. The reasoning: when that many banks and insurers lean on the same few providers, a failure at one of them is a risk to the whole sector. A trading company in Noida is not a bank. The logic still holds, just at a smaller scale.
Find the single points of failure first
A single point of failure is any one component that, when it breaks, takes the whole process down with it. In small business software it is rarely the obvious one. These are the ones we keep finding when we review an existing system:
- DNS for the domain sits with one provider, and nobody is sure who has the login.
- Every page pulls a script or font from a third-party CDN and hangs if that CDN does.
- Sign-in works only through “Sign in with Microsoft” or Google, with no fallback for admins.
- One database in one region, with its backups stored in the same cloud account.
- A payment gateway or SMS provider whose failure the app treats as a crash rather than a delay.
Write the list out for your own system. You can't remove every dependency, and trying to would make a modest project absurdly expensive. The point is to know which ones would stop work, then decide on purpose which of them deserve a fallback.
For a small business, multi-cloud is usually the wrong answer. Knowing exactly what breaks, and in what order, is almost always the right one.
Design patterns that survive a cloud outage
Graceful degradation.When a non-essential dependency fails, the app should carry on with its main job. If the SMS gateway is down, save the order and queue the message. If an analytics script won't load, the page shouldn't care. Most of this comes down to code that expects failure: a timeout on every external call, retries with backoff, and a queue for anything that can happen later.
Offline modes for field and counter work. A delivery app or a billing counter can store work on the device and sync when the connection returns. Progressive web apps and native mobile apps both support this. It adds design effort, mostly around what happens when two people edit the same record while offline, so we use it where staff genuinely work through patchy networks or must keep billing regardless.
Backups that live somewhere else.Daily database backups copied to a different provider, or at minimum a different account, with a restore you have actually tried. An untested backup is a hope. We also like a plain CSV export of the few tables you would need to trade by hand for a day: open orders, today's schedule, customer phone numbers.
Monitoring from the outside.Check your app from somewhere other than your hosting provider, so you hear about a failure before your customers do. Subscribe to your providers' status pages as well, but expect them to lag behind what users are already seeing. In several of this year's incidents, independent trackers picked up problems before the official pages did.
Accounts written down. Domain registrar, DNS, hosting, email, payment gateway: who owns each login, where the recovery codes are kept, and who to phone. This costs nothing and removes the worst hour of most incidents.
How much resilience a small business actually needs
Our working rule, for a client in Delhi NCR or anywhere else, is to match the effort to what an hour of downtime costs. A brochure website can be down for an afternoon and nobody loses money. A billing system for a pharmacy chain or a dispatch board for a transport company can't. Most businesses sit in between, where outside monitoring, tested backups, careful handling of third-party failures and a one-page runbook cover nearly all the realistic risk at modest cost. Running the same system live on two clouds is a different budget altogether, and we rarely recommend it.
If you already have a system and don't know where it stands, a dependency review is a sensible first step. That kind of upkeep is part of our software maintenance retainer, and planning for outside services to fail is built into our API integration services.
Want to know what your software would do during the next cloud outage? Tell us what it runs on and what it connects to, and we will map the weak points and say which are worth fixing. If you are reviewing costs at the same time, read what the Microsoft 365 price rise means for Indian businesses, or start a conversation on our contact page.