​How To Get AI Customer Service Right

Alex Nazha is head of Gatestone Digital.

When you call a company’s help line and hear an automated voice on the other end, what’s your first reaction? Chances are it’s searching for the option to speak with a real person. That predictable response is evidence that the technology has moved faster than customers’ appetite for it. AI is powerful—but do customers want to interact with it in the first place?​

We hear constantly from clients and companies who have invested in expensive AI customer service systems and are not seeing the returns they expected. While CFOs push for the financial outcomes, customer service and sales teams are asking for something different: more personalization and happier clients. The tension between those two sets of expectations is a big part of why so many AI rollouts fail.​

Where Companies Go Wrong

The biggest challenge I see companies underestimate is the sheer amount of work required to get an AI system ready for client-facing use. Most people’s first experience with AI was typing prompts into Claude or ChatGPT, which felt simple and immediate. Enterprise AI is a different beast. It requires a lot of unglamorous work behind the scenes, including getting data organized and making systems talk to each other. Processes that were built around humans need to be adapted so they work for bots and agents.​

There’s a tendency to overestimate how willing customers are to deal with a system that can’t fully meet their needs. Many companies are able to install AI at a basic level; it can give a customer their account balance, say, but not resolve the actual problem they called about without talking to a person. That half-measure creates more frustration than if AI hadn’t been there at all. And when companies pour money into client-facing AI only to discover the appetite simply isn’t there, they often end up scaling back and blaming their customers for their own miscalculation.​