The AI Receptionist Bubble

Vertical voice AI companies are the hottest startups in Silicon Valley these days. Companies like Assort Health, Elise AI, and countless other “replace your receptionist” startups are raising at sky-high valuations to answer your phone for you. Assort Health raised a $50m Series B at $750 million with only $3 million ARR. Elise AI recently raised $250m from a16z (although they have substantial revenue).

There can be no question: we’re in a golden era for voice AI companies. Answering the phone is sexy, not just to VCs, but also to enterprise buyers and even to administrative staff themselves. My mom works as a secretary at a lawyer’s office far from Silicon Valley, and even she wishes that “AI could answer the phone.”

The Promise

The buyers of voice AI see it as the solution to freeing up their staff from an endless stream of queries and problems that come in by phone. It makes sense: they see their staff spending 6 hours a day answering calls and would rather they spend their time doing something more “productive” (or alternatively, they would love to fire the staff altogether). Then the AI receptionist salespeople start calling, the managers get on a demo, and it’s amazing, it answers the phone and sounds like a human.

The Problem

The problem is, the AI receptionist does not actually solve any of the customers’ problems.

In most cases, people calling you is a symptom of a problem, not the actual problem itself. People are calling because something unusual has happened: they didn’t receive their order, they’re going to be late to their appointment, the online payment portal is broken, and so on.

Consider a shipping company fielding constant calls about late deliveries. The root cause isn’t the phone calls, it’s the inaccurate delivery estimates that need to be fixed upstream.

Sure, you get a few calls from people who would simply prefer to call in rather than do something online, even if nothing is wrong (e.g., the elderly scheduling an appointment). In this case, the AI receptionist kills it. But for solving actual, messy, one-off issues? No. The way to stop people calling you about those issues is to dig into the actual root causes of the problems about which they are calling. It’s a tricky process with no quick wins, which is why people are trying to paper over the cracks with voice AI.

The Real Question

To be clear, voice is a fantastic interface. It’s high bandwidth, low latency, and natural for humans. But we must distinguish between a good interface and a good solution to customer problems. Right now buyers are enamored by the former while ignoring the latter. Eventually, reality will catch up.