OpenAI Outage: What Happens to Your Business Process When AI Goes Down?

Outages should be considered when we design a process

Any technology can become unavailable. An API can fail, an integration can stop working, a third-party system can be temporarily inaccessible. AI is no different.

When I design a process, I want to understand these dependencies and what happens to the work if one of them becomes unavailable. Ideally, that question is answered during the design, not for the first time when an outage actually happens.

That doesn't mean creating a complete manual duplicate of every automated process. In many cases that would defeat the purpose of automation.

But for an important business process, I would want to know that there is still a practical way to continue the work.

A simple customer enquiry is a good example

Imagine a business uses AI to help process incoming customer enquiries.

A customer submits an enquiry through the website and explains what they need. AI reads the request, identifies what the customer is asking for, extracts the relevant information and helps move the enquiry through the next steps of the process.

When everything is available, most of this can happen automatically.

But I would still want the original customer enquiry to be stored before the process becomes dependent on AI.

The customer's name, contact details and original message should still be there. The lead should still be visible to the business. Any other information required to handle that enquiry should still be accessible.

Now imagine the AI service is unavailable.

The automated route can't continue, but the customer enquiry hasn't disappeared with it.

Someone can open the lead, read exactly what the customer submitted and continue processing the enquiry manually. If necessary, they can contact the customer directly and complete the work that AI would normally have helped with.

It will probably take longer.

But slower is very different from stopped.

The process can have more than one path

This is where I think the design of the process becomes important.

The normal route can still be heavily automated. We don't need employees manually checking every enquiry simply because an outage might happen one day.

Instead, the process can have a decision point.

Is AI available?

If yes, the enquiry continues through the normal automated process.

If no, the enquiry remains available and can be handled manually until the automated service is restored.

The manual route isn't an error in the design. It is part of the design.

And importantly, it doesn't mean employees have to work manually when everything is operating normally. The automated route remains the normal route.

The manual option is there when it is actually needed.

The same principle applies beyond AI

AI is a useful example because of the OpenAI outage, but this isn't really an AI-specific problem.

Businesses increasingly rely on APIs, integrations, payment providers, cloud platforms and other external services. Any of them can become temporarily unavailable.

The questions I would ask depend on how important that particular process is.

If an integration fails, is the information still stored somewhere?

Can the transaction be retried later?

Will someone know that something hasn't been processed?

If a customer is waiting, can an employee access the information and deal with the request manually?

Or does one unavailable component stop everything?

Not every process needs the same level of protection. If an internal report is delayed for an hour, perhaps the business can simply wait. If a customer request is time-sensitive, the impact can be very different.

This is why the fallback should reflect the importance and risk of the process rather than being added automatically to everything.

Automation should remove manual work, not the ability to intervene

A big part of process improvement is finding manual work that people shouldn't need to do in the first place.

If technology can reliably complete a repetitive task, I don't want an employee performing it manually every day just because we want a backup.

But there is an important difference between requiring someone to perform a task manually and allowing someone to perform it manually when necessary.

That distinction becomes more important as AI moves beyond being a tool people occasionally open and starts becoming part of operational workflows.

If AI is summarising a document for an employee, an outage may simply be inconvenient.

If AI sits in the middle of every customer enquiry and nothing can move without it, that's a different business dependency.

I would want to understand that before implementing the process.

One more question to ask when introducing AI

When deciding whether to introduce AI into a business process, there are already plenty of questions worth asking.

Does the process actually need AI? Could ordinary automation handle the task? What information will AI have access to? What happens if it interprets something incorrectly? Where should a person review or approve an action?

I think there is one more question that belongs in that conversation:

What happens if the AI isn't available at all?

Sometimes the answer may simply be: we wait until it comes back.

That's completely reasonable if the business impact is low.

But for an important process, if the answer is "the entire process stops and nobody can access or continue the work", I would want to look at the design again.

The OpenAI outage was resolved within a few hours. The next outage could involve another provider, another system or another integration.

That isn't a reason to avoid AI or automation.

It's a reason to design the process properly.