By Chloe Castle, National Head of Sales & Marketing at Backspace

On September 03 2026, ChatGPT, Claude and Grok stopped responding. Claude was down for nearly three hours and ChatGPT for just over two.

The incidents involved elevated errors and degraded service, and downtime was attributed to routing configuration errors, infrastructure issues and an outage, respectively. The incident raised a practical question for companies that rely on AI in customer service – if the automated channel becomes unavailable, what happens to the customer? 

When an AI customer service tool stops working, who picks up the phone? For a business that has handed customer conversations to automation, that question comes down to three key elements: who is accountable, who owns the unresolved query, and how does the customer reach the company?

While these tools are designed to help your business optimise systems and improve productivity and processes, they shouldn’t be used to replace people. When technology fails, the difference between a disaster and a brief hiccup is the human being on the other end of the line. 

The customer is the first to lose out when systems are comprehensively automated with little to no interaction with another human being. Then the business loses – customers, engagement, reputation. It also loses insight because most of the fixes that a leadership team debates in the boardroom can be traced back to the moment when a customer said a request took too long or nobody called back.

Those complaints are the fuel for change, helping companies to improve their processes. While an AI can read through customer interactions and surface trends at speed, a bot returns pre-populated responses instead of the raw feedback the business needs. 

People do not want to be a number or a line item. They want to speak to someone who can help them when they experience a problem that has an impact on their life, like a billing issue or an outage. However, they also want control and the ability to solve some of their problems on their own. And here lies the balance between the bots and the humans. Self-service has a place, especially when it is matched to the right tasks like buying movie tickets or reading product information about a firewall or downloading a monthly invoice. These are quick, self-contained jobs that a human is happy to do alone. However, a customer who has been invoiced double for a service and has tried repeatedly to speak to someone who can help has little patience for a scripted journey that opens with: have you read your invoice correctly? 

Customers are frustrated by the inability to reach human support, especially when it comes to downtime or outages. They don’t want a script asking them if they’ve rebooted their router; they want a capable engineer.  Many faults do need physical investigation.

For example, an upstream network provider problem cannot be resolved by one company’s bot talking to a mobile operator’s bot because privacy requirements alone prevent that exchange. Large operators employ people in specialist roles for a reason, and so, when parts of support are turned into do-it-yourself steps, the customer is the one who pays the price. 

Again, it comes down to balance. You need to decide where a bot stops and a person takes over within the user journey, and that means research done beyond paper. Staff need to take the journey themselves, from onboarding to complaints, and sit in the customer’s seat to experience where it slows down.

A bot works well as a first point of contact for logging a fault, particularly over WhatsApp when a customer’s internet is down and their email sits on an office laptop. It can also resend an invoice. It cannot show empathy or make a judgement call or help a customer with a unique issue. That is where you need a person who understands the situation and the relationship.

In the end, accountability starts at 2 am when the system detects a failure, opens a ticket and has clients back online before the business day begins. It is having a team of experts who own a problem so the customer doesn’t have to when a storm takes out a network site, and 2,000 clients go down at once. It is the communication channels your business has open to let people know that their service is down, how long it will take to repair and what caused it. It is following up with a post-mortem and a full and transparent reason for the outage. 

People can scale with AI, and your support team can be large enough to give your customers the right levels of attention. While yes, this is a significant operating expense, the counterweight is retention. A business that cuts salaries by automating customer contact has to ask if it has kept customers on the books.

Yes, AI has a role and can absolutely help employees work more efficiently, but it should never stand between a customer and a problem resolution.