Most AI receptionist products stop at "answer the call and email you a transcript." That's not automation — that's delegation to a slower process.
We've been building integrations between our voice AI and the actual business tools Irish companies use daily: Dentally (dental practice management) and Flipdish (restaurant ordering platform).
The Technical Challenge
The interesting engineering problem isn't the voice AI itself — it's the real-time decision loop during a live call:
- Caller says "I need to book an appointment for Thursday"
- AI queries Dentally API for Thursday availability (< 500ms budget)
- AI presents options conversationally
- Caller confirms
- AI creates booking via API + sends SMS confirmation
All of this needs to happen within natural conversation flow. You can't have a 3-second pause while your API call resolves.
Flipdish Integration Pattern
Restaurant ordering has a different challenge: menu complexity. A restaurant might have 200+ items with modifiers, sizes, and combos. The AI needs to:
- Map spoken items to exact menu entries ("large pepperoni with extra cheese")
- Handle modifications ("no onions, add mushrooms")
- Calculate pricing in real-time
- Push the complete order to the kitchen display
We sync the full Flipdish menu on a schedule and keep a local cache for sub-100ms lookups during calls.
Key Takeaway
If you're building voice AI for business use cases, the voice model is maybe 30% of the work. The other 70% is integrations, edge cases, and making API calls fast enough that the conversation feels natural.
Would love to hear from others building real-time API integrations into conversational AI — what latency budgets are you working with?
![Defluffer - reduce token usage 📉 by 45% using this one simple trick! [Earthday challenge]](https://media2.dev.to/dynamic/image/width=1000,height=420,fit=cover,gravity=auto,format=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fiekbgepcutl4jse0sfs0.png)
