How to set up a cross-channel guest agent with automated grading

The context fragmentation problem

Guests reach out across phone lines, text messages, and chat apps. A customer calls your front desk in the morning. They follow up over WhatsApp at noon. They send an SMS in the evening. Most operations handle these as three separate conversations. Your team spends hours catching up on context, or worse, asking the guest to repeat themselves.

Centralizing communication requires one system that holds conversation history across voice and chat. Voicetta operates as a unified brain for these interactions. Here is how to set up an agent across phone, messaging, and operational tools without fragmenting context or locking into proprietary models.

Step 1: Select your model architecture and pricing path

First, decide how you want to route underlying model calls. You have two main choices: Bring Your Own Key (BYOK) or Managed AI.

• Bring Your Own Key (BYOK): This route starts around 1.6 cents per minute, or roughly $0.96 per hour for inbound US traffic. You plug in your own API keys from providers like OpenAI, Anthropic, Gemini, Deepgram, and ElevenLabs. This approach gives you direct control over your model accounts and lowers software overhead.
• Managed AI: This setup starts around 12.4 cents per minute, or roughly $7.44 per hour. Voicetta provides the underlying model stack out of the box. It gets your deployment moving faster if you do not want to manage separate developer accounts.

Neither path locks you into a single provider. If you want to use Deepgram for transcription, Anthropic for reasoning, and ElevenLabs for voice synthesis, you can stitch those specific engines together.

Step 2: Connect your operational channels and systems

An agent is useless if it cannot read or update your tools. Connect your external channels and background systems to the platform before going live.

Attach your incoming communication channels first. That includes phone lines, SMS routes, and WhatsApp threads. Once channels are active, connect your database and scheduling software. Supported tools include HubSpot, Google Calendar, Cal.com, property management systems (PMS), and general CRMs.

When a guest calls to book or ask a question, the agent checks live availability in Google Calendar or Cal.com. If the conversation moves from phone to WhatsApp, the transcript and context stay in the same timeline. Your CRM or PMS gets updated automatically, keeping staff up to speed without manual triage.

Step 3: Test interaction behavior inside the browser

Do not deploy voice agents directly to live phone numbers without testing. Voice latency, interruption behavior, and tool calls require dry runs.

Use the browser testing sandbox to talk directly to the agent. Test boundary cases in real time:

1. Speak over the agent while it answers a routine question to test interruption handling.
2. Ask complex questions about policies to ensure response accuracy.
3. Trigger a scheduling request to verify that calendar entries populate correctly.

Iterate on response style, listening parameters, and prompt rules inside the sandbox environment. Push the agent live only when performance matches your operational standards.

Step 4: Configure automated conversation grading

Manual quality assurance does not scale when call volumes grow. You cannot listen to every recording or scan every chat transcript by hand.

Set up automated performance grading for phone calls and WhatsApp threads. Define the specific behaviors that mark a successful conversation:

• Did the agent identify the guest and retrieve their booking details?
• Did it adhere to your compliance and policy requirements?
• Did it successfully schedule a callback or calendar appointment?

The system evaluates every completed phone call and WhatsApp message thread against your rules. It assigns a clear Pass or Fail status, accompanied by explicit reasoning for the rating. Operations managers can filter by failed grade to spot broken workflows or policy gaps immediately.

Step 5: Review unified timelines and performance metrics

Once live, track guest interactions inside a single dashboard. Rather than checking phone logs in one tab and WhatsApp Web in another, view everything on a searchable timeline.

Each record includes the complete transcript, audio call recording, performance analytics, and clear status pipeline updates. If a specialist needs to step in on a complex issue, they read the consolidated history across all channels first. Context stays intact, call triage drops, and guests get immediate answers.