The new trend is Voice AI and I believe "voice first" will be how we build applications with a fall back to traditional methods. This is great however, I think the approach of building by relying on large single prompts, letting AI run free based on set instructions within prompts, is wrong. Not to mention current costs range from 12-25 cents per minute which means while it scales well, most AI Voice methods cost more than having an overseas representative.

Running Voice AI apps specifically for phone support is expensive, especially if you lack scale to reduce costs through volume, didn't optimize through engineering, or build from scratch without a thoughtful way to divide problems. Right now most building in this direction rely on AI to handle everything and provide the right outputs. The results have been subpar and over-engineered in the wrong direction.

One specific use case is AI versus IVR. Do we really need an entire AI system just to replace the IVR, or can we merge the two?

I've worked with IVRs for Autodial.com and noticed most inbound call apps don't need complex voice AI. While IVRs are "bland", they've cost-effectively solved most customer service issues for decades compared to expensive AI. The problem wasn't the IVR itself - it was wait times and reaching a real representative.

When building, we want to effectively solve customer issues, not just add sweet frosting to a bad cake.

For years IVRs have been used for:

• Smart call routing
• Callbacks
• Surveys
• Basic support like banks

But IVRs have faced issues:

• Limited interactions: presses or commands
• Limited responses
• Robotic, not lifelike voice
• No backward compatibility
• Time consuming prompts

An intermediate step before complex AI could be Smart IVRs combining the best of both to resolve limitations and keep costs low while allowing more complex capabilities if needed.

A Smart IVR should:

• Increase interactions through natural language
• Understand requests
• Provide dynamic, lifelike responses
• Enable dynamic navigation and backwards movement
• Save and reference data
• Enable real-time sentiment analysis

I'd argue Smart IVRs are the way to start effective AI voice systems for inbound calls. With the ability to jump around, micro models can be easily implemented and optimized over time, reducing costs and improving performance. Voice applications are here to stay and costs will fall, but we should consider the best way to blend IVR and AI to solve problems.