Ana — voice AI receptionist so you never miss a call

The invisible cost of the calls nobody answers

In a technical-services company, the phone is the first point of contact. But answering calls eats hours every day, most are repetitive (“do you do X?”, “how much does Y cost?”, “can you call me back later?”) and the ones that come in after hours simply get lost. That loss doesn’t show up in any report: you never know how many prospects never called again.

As a solo founder juggling sales, delivery and support, that friction was daily. And as a technologist following the field, we knew the tech was mature enough to fix it. So we built Ana at home before selling it outside.

What Ana actually does

• Answers incoming calls 24/7 in natural Spanish
• Introduces herself as a virtual assistant and informs callers the call is processed according to our privacy policy
• Identifies the reason for the call (quote, support, commercial, other)
• Holds a contextual conversation about VisibleSoft and our products (Power Time)
• Collects caller name, company, reason for contact, phone and email
• Summarises the conversation and sends an email to the team on hang-up with all the extracted data

How it’s built

Voice. Enterprise-grade telephony for inbound calls. Speech recognition tuned for Castilian Spanish, with a custom business vocabulary (proper names, products, sector-specific terminology) so it transcribes domain terms accurately. Speech synthesis with ElevenLabs in a neutral Castilian Spanish voice, tuned to sound natural over the phone.

Brain. OpenAI LLM through Telnyx’s native integration. The assistant’s instructions are versioned inside the portal itself: every change is stored as a version that can be diffed, reverted and deployed.

Integration with our systems. When the conversation ends, Ana invokes a function that sends the data to our backend through an authenticated webhook. The assistant has one extra turn specifically for this, just before hanging up.

Team notification. The backend receives the summary and lead data and sends an email to the corporate inbox with everything needed for follow-up. No intermediate database: the information lives where the team already manages commercial follow-up.

Compliance. Personal data handled per GDPR. Caller notified at the start of the call. Privacy policy updated naming Telnyx LLC as data processor. Legal basis: legitimate interest (Art. 6.1.f GDPR).

Three things we learned building Ana

1. The prompt is code and you version it like code. Every change to the main prompt is stored as a new version in Telnyx’s portal: you can diff two versions, roll back if the latest broke a behaviour and leave descriptive commit-like messages. Tweaking “just the text” without that discipline leads to invisible regressions.

2. The team stops watching the phone. Previously you had to log into the Telnyx portal to find out who called and why. With the email notification at the end of every call, all the context lands directly in the inbox: name, company, reason and an actionable summary written by the model. That small operational change completely shifts how commercial follow-up happens.

3. Summary quality matters more than length. The model produces a 2–3 sentence summary that captures real nuance: whether the caller was hesitant, whether the conversation got cut off before all the data was collected, what’s still pending. That removes the need to listen to the recording or read the full transcript to understand context.

What’s next

We’re iterating. Immediate roadmap:

• Native multilingual support: dedicated voices and speech recognition per language (Catalan, English, French, German, …) instead of the current Castilian voice adapting, so the experience feels natural also when the caller doesn’t speak Spanish.
• Automatic scheduling into Google Calendar for human follow-up calls when Ana detects it’s needed.
• WhatsApp notifications alongside the current email channel, for urgent or after-hours cases.
• Structured Knowledge Base with FAQ, pricing and use cases, replacing the inline content in the system prompt.
• Live human handoff for cases flagged as urgent during the call.

Want this in your company?

The same solution can be adapted to your company: your voice, your services, your integrations, your vocabulary. Tell us about your case and we’ll prepare a tailored proposal with no strings attached.