Table of Contents
It’s two o’clock on a Tuesday. A customer places an order on your website. Behind the scenes, an automation is meant to pick it up, log it in your accounting software and send a confirmation.
But the power at the office went off ten minutes ago, and the router went with it. The automation tries to reach the accounting software, which runs on a computer in the back room, and can’t. What happens to the order?
In a well-designed setup, it waits, tries again when the connection returns, completes, and tells you if it can’t. In many setups, nothing happens at all. The order sits unlogged and the customer waits. The first anyone hears of it is an angry message two days later.
This is the part of automation that most guides skip. They assume the power is on, the internet is stable and every system answers every time. If you run a business where that isn’t guaranteed, automation with unreliable internet and power needs to be designed differently. This article explains how.
The problem is quiet failure, not loud failure
A broken automation that shouts is easy to deal with. You see the error, you fix it. The dangerous ones fail quietly: they stop, skip or half-finish, and nothing tells you.
In unreliable conditions, almost every failure falls into one of three types.
| Type | What happens | Example |
| Lost | The work never arrives, or arrives and is dropped | An order is placed while your system is offline and never reaches it |
| Half-done | One step completes and the next doesn’t | An invoice is created but never sent, or stock is reduced but the order isn’t logged |
| Doubled | A retry repeats work that already succeeded | A customer gets two confirmations, or is invoiced twice |
Everything else in this article is a way of preventing one of these three.
Six design habits that make automation resilient
1. Know what has to be switched on
For every automation, trace the chain and list everything that must be working for it to run. That includes your website, the automation tool, your accounting software, the computer it runs on, the router and the power supply.
This often reveals the weak link. A cloud-based automation tool keeps running when your office goes dark, but only if everything it needs is also in the cloud. The moment it depends on a computer in your office, it depends on that computer’s power and connection. A staff member’s laptop that goes to sleep at night is a very common hidden dependency.
Where you can, move critical steps to systems that don’t depend on your premises. Where you can’t, protect those premises (more on that below).
2. Retry, but with limits
Networks fail briefly all the time. A step that gives up on the first failure turns a thirty-second blip into a lost order. A step that tries again a few times, with a pause between attempts, usually succeeds without anyone noticing.
Retries need two limits. They should stop after a sensible number of attempts, so a real fault doesn’t loop forever. And when they stop, they should raise an alert that a person will act on. Check how the tool you use behaves when a step fails, because defaults vary and some tools simply stop.
3. Record first, process second
The safest approach is to make sure the incoming work is saved somewhere durable before anything tries to process it. If an order is stored by your website or form tool the moment it’s placed, then a failure further down the chain doesn’t lose it. The order still exists, waiting to be processed.
In practice, this means your automation should work from a record that persists, not from a message that arrives once. When your systems come back online, it picks up what’s waiting. This single habit prevents most “lost” failures.
4. Make repeating harmless
Retries create a new risk: doing the same thing twice. The fix is to give every piece of work a unique reference, such as an order number, and to check before acting whether that reference has already been handled.
If the invoice for order 1042 already exists, the automation should recognize that and skip it, not create a second. Done properly, running a step twice has the same result as running it once. This is what lets you retry freely without fear of duplicates.
5. Make silence visible
This is the most important habit. An automation that stops working should be noticed within hours, not days.
There are simple ways to achieve this:
- A daily summary. A short message that says “14 orders processed today, 0 failed.” If it doesn’t arrive, that is your warning.
- An alert on zero. If you normally get orders every day and the count is zero, someone is told.
- An end-of-day count check. Compare the number of orders on your website with the number in your accounting system. They should match, and if they don’t, you find out the same day.
- Alerts to a person who acts. Failure notices that go to an inbox nobody reads are no use. Name the person responsible, and a backup.
6. Write down the manual fallback
Sometimes the automation will be down for hours. Staff need to know what to do: where to record orders, who to tell, and how to catch up afterwards. Write it on one page and name an owner.
The catch-up step is the one people forget. When systems return, someone must check that everything handled by hand was also entered, without duplicating anything the automation handled before it stopped.
Protect the basics
Software design only goes so far. Some practical hardware and setup choices make a large difference:
- Keep the router and any critical computer on backup power, such as a small UPS, so a short outage doesn’t interrupt a process halfway through.
- Have a second internet connection, such as mobile data, that takes over when the main one fails.
- Don’t run critical automations on a machine that sleeps. Check power and sleep settings on any computer that does background work.
- Prefer cloud-hosted systems for critical steps, so that the work continues even when your premises don’t.
When it must work offline
Some work happens where connectivity is poor by nature: shops, warehouses, vehicles, and field staff collecting data. For these, an app that needs a live connection to function will frustrate people.
If this is your situation, ask any provider whether the app can capture information without a connection and send it when connectivity returns, and how it handles conflicts if two people changed the same record in the meantime. This is a design requirement from the start, not a feature you can add easily later.
If you use AI in the process
AI steps, such as drafting a reply, reading a document or classifying a message, add another dependency: an outside service that can be slow or unavailable. Design the workflow so the essential work still happens without it. If the AI step fails, the item should drop into a person’s queue instead of blocking the whole process. Treat AI as something that improves the process when available, not something the process can’t live without.
A test you can run this week
You don’t need a technical team for this. Choose a quiet time and run a deliberate bad day:
- Pick one important automation, such as order handling.
- Switch off the internet at your router, or disconnect the system it depends on.
- Place a test order, or trigger the process once.
- Reconnect after a few minutes.
- Check three things: did the work go through exactly once, did you receive an alert or summary, and did anything get duplicated?
Whatever goes wrong is a gap you now know about on your own terms. Fix it, then test again.
If you’re building or buying automation, these are the questions to put to your provider:
- What happens to a task if the system it needs is unavailable?
- Does it retry, how many times, and what happens afterwards?
- Where does the work wait while it’s pending?
- How do I know when something has failed?
- How do you stop the same task being done twice?
- What should staff do when it’s down?
A good provider will have clear answers. A vague one is a warning.
Match the effort to the cost of failure
Not everything needs this treatment. A reminder that occasionally doesn’t arrive costs you little. Orders, payments, customer messages and anything that touches money deserve all six habits. A useful question for each automation is: what does it cost me when one gets missed? The answer tells you how much safety to build in.
This also fits with how you choose what to automate in the first place. Our guide to the 10 business processes every growing company should automate first can help you start with processes where failure is cheap, and learn before you automate the risky ones.
Frequently asked questions
Can I use cloud automation if my internet is poor?
Yes, if the critical steps run in the cloud and the incoming work is saved durably at the source. The cloud tool doesn’t need your office to be online, but anything connecting to your premises does.
Do I need a developer to make automation resilient?
Not for the basics. Daily summaries, alerts, count checks, backup power and a written fallback need no code. Retries, queues and duplicate protection are easier with the right tool or developer, and worth asking about.
How will I know if an automation has failed?
Only if you set that up. Many tools tell you about errors inside the tool but not about runs that never happened. A daily summary or a count check catches those.
What should I do first?
Run the bad-day test on your most important automation. It costs nothing and shows you your biggest gap.
Build for the bad day, not the good one
An automation that only works when everything goes right isn’t finished. The ones that earn trust are the ones that recover quietly, avoid doubling work and tell you the moment something didn’t run. If you’d like help making your existing automations more resilient, or building new ones with unreliable conditions in mind, our custom software development and AI Automation work covers this, and you can contact us to talk it through. For the process side of the same problem, see why automation projects fail when the work behind them isn’t understood.
