Rethinking AI in Customer Service: A Cautious Approach

As the founder of a voice AI company, I find myself in a unique position—I often advise potential clients against utilizing our product. This may sound counterintuitive, and some of my sales team might even think I’m out of my mind, but my experience has shown me that incorporating AI into unsuitable scenarios frequently leads to more issues than it resolves.

Just last month, a law firm sought our assistance to streamline their client intake process using AI. However, upon reviewing their call recordings, I quickly determined they were not ready for such an implementation. Their intake involved sensitive legal inquiries, emotional clients recounting distressing experiences, and intricate eligibility assessments—all of which would have been disastrous in the hands of AI.

The truth is, the current buzz surrounding AI has led many businesses to believe they require it immediately. However, the reality is that AI excels in specific applications while faltering dramatically in others.

Before considering the integration of voice AI, businesses should evaluate their operations against three key criteria:

1. Consistency in Call Patterns

In my analysis of over 10,000 customer service calls across various sectors, I discovered that many industries see as much as 80% of calls stemming from just a handful of common issues—such as appointment scheduling, frequently asked questions, and basic troubleshooting. These situations are ideal for AI deployment.

Conversely, if your calls vary significantly, it’s time to reconsider. For instance, a mental health clinic I assessed faced completely unique calls where every patient required nuanced, empathetic responses. In such cases, relying on AI would not only be inappropriate but could also be damaging.

To aid in this evaluation, we developed a pattern analysis tool that reviews call transcripts. If fewer than 70% of calls exhibit discernable patterns, AI is likely not a suitable option. A home services company realized that 85% of their calls were for scheduling—making them perfect candidates for AI. In contrast, a B2B software firm found only 30% of their calls fit into recognizable patterns, pointing to a continued need for human representatives.

2. Establishing Clear Escalation Protocols

For AI to operate effectively, it must be paired with well-defined escalation strategies. I witnessed a company deploy a chatbot without any escalation logic, leading to mounting frustration among customers who were simply trying to speak with a manager.

Before deploying AI, it’s crucial to pinpoint when calls