Let Me Connect You With Anjana – Role Continuity in Voice AI

You go to a upscale barbershop. There is a person sitting in reception. You ask them about the services, prices, products being used etc. They tell you everything about it. You also demand a small sample hair cut. And voila, that person sitting at the reception, gets up from their chair, puts on an apron and picks up a scissor.  

Did that feel unnatural?

I tried Bolna.ai‘s General Sales & Platform Inquiries voice agent today.

This isn’t a review of Bolna’s voice-AI platform as a whole. I’m looking at one specific interaction from a UX perspective: what happens when a voice agent has to demonstrate another voice agent. And once you read what follows, you will get what I meant by the barbershop example.

The call was picked up by a young male voice. He told me about the services, including pricing. He also demo-ed a sample call for a company. But what felt unnatural, was that he did it in his own voice. 

I even asked him, if they had other voice types and agents. He confirmed they had both male and female voices, in multiple Indian accents. When asked to sample a Female bengali accent, he switched his accent to Bengali, but kept his male voice. When asked why can’t I hear the female voice, he said it was not possible in this call. That he could setup a different call. Why do I have to setup another call for something so basic? 

The UX issues that I could identify in this call were:

1. Role Collapse – Breaks expected role separation, between receptionist vs call executive. 

2. Persona Collapse – when I also wanted gender switching – Creates an uncanny mimicry effect – like I suddenly switched on Kapil Sharma show. 

3. Interaction dead-end – Forced Redial – This is the biggest problem with this. Perhaps there is a technical reason why switching voice personas within the same call isn’t supported. But that implementation constraint shouldn’t necessarily dictate the user’s experience.  Making me call again and talking to someone else, resets trust. It assumes I ‘the user’ will bother twice. And the worst of all, it silently converts a hot lead into a lost one. Why not transfer the call to a different voice agent? Like actual humans do, “Let me connect you with Anjana”. At the very least he could have used a recorded sample of Anjana’s voice. His switching his accent to bengali, when I wanted to hear Anjana speaking was at least a little weird. Could it not redirect the current active call to whatever service that lets me speak to the Bengali lady? It is not like he has to check her google calender for a free slot. This creates high friction and forces me to either cold start the call again. Or worse, not call at all.