AI in customer service: Why your newest employee is failing their probationary period

AI in customer service promises greater efficiency, lower costs and scalable service processes. Yet in practice, many projects fall short of these expectations. In this guest article, René Jacobi from Diabolocom explains why it is not the technology itself that is the problem, but often the way it is implemented. He explains the role played by well-defined processes, existing practical knowledge and integration with back-end systems – and why companies should treat their AI more like a new employee than just another software licence.

The industry is chasing the next big promise: AI in customer service is set to revolutionise operations, cut costs and increase scalability to immeasurable levels. Expensive licences are being purchased and chatbots or voicebots are being hastily rolled out.

What’s the result in practice? Customers get stuck in frustrating loops and shout exasperatedly into the phone, “I want to speak to a real person!” There is widespread disillusionment in call centres. Yet the failure is not down to the technology, but to fundamental management errors during implementation.

The first mistake starts right at the definition stage. In many strategy meetings, simple automation and artificial intelligence are blithely lumped together. This is a costly mistake.

Automation is a rigid, rule-based process (if A, then B). If a customer wants to reset their password or change their address, I don’t need a cognitive language model (LLM) that simulates human empathy. I need a clean, error-free workflow. Automation does not necessarily require AI.

Artificial intelligence, on the other hand, comes into play when the context is unclear, when intentions need to be filtered from unstructured free text, or when complex conversations are being conducted. Anyone who overcomplicates simple if-then processes with expensive AI – or, conversely, expects rigid automation to cognitively process complex customer enquiries – is simply wasting the budget.

Let’s stop viewing AI as some sort of magical, out-of-the-box solution. At its core, AI in customer service is nothing more than a new member of staff.

Imagine putting a human agent on the phone on their first day. You give them an outdated, 500-page PDF manual, but deny them any opportunity to consult with experienced colleagues. They would fail miserably. Yet this is precisely the standard approach in current AI projects. The message is simply: ‘We’ll just integrate the AI with your existing knowledge base.’

However, outstanding service staff excel thanks to years of accumulated, often undocumented, solution know-how. They know the workarounds and the unwritten rules. If you feed an AI exclusively with sterile FAQ documents, you’re simply creating a digital parrot. Proper onboarding means first extracting this hidden experiential knowledge from top agents and then teaching it to the machine.

Carry out a reality check within your service organisation: take exactly the same customer problem and ask five different staff members for the correct solution.

If you don’t get exactly the same answer five times, your process isn’t AI-ready.

The reality is often that every agent has devised their own workaround because the official process is too cumbersome. However, AI relies on processes running in a deterministic manner. If you give AI an inconsistent process, it won’t fix it – it will simply scale the chaos at record speed.

The biggest source of frustration for customers is when an employee is friendly but isn’t authorised to resolve the problem. What use is even the most motivated agent if they don’t have the rights to access the ERP system or the CRM? They’ll have to transfer the customer to someone else.

Exactly the same applies to your digital colleague. An isolated AI that can conduct natural conversations but lacks real-time interfaces to your core systems is toothless. A genuine customer experience can only be achieved through deep backend integration. The AI must be able to check a delivery status in real time, close a ticket independently or trigger a workflow. Without this API access, the AI degenerates into an expensive doorman who ultimately just passes the work on to human agents anyway.

It’s time to cut through the hype and focus on the practicalities of day-to-day operations. Invest in the onboarding and knowledge base of your digital colleague. Standardise your processes before handing them over to a machine. And above all: free AI from data silos by creating genuine backend integrations. AI isn’t just a software licence you buy – it’s a team member that needs to be orchestrated, managed and trained.