Voicemail Detection Is a Classifier. Plan for Both Ways It Is Wrong.

Описание и применение

В статье описывается, как правильно обрабатывать звонки в голосовых агентах, чтобы избежать ошибок классификации голосовых сообщений и повысить эффективность кампаний.

Задача

Outbound voice agents often misclassify voicemail greetings as human or vice‑versa, leading to wasted minutes, lost leads, and misleading metrics.

Результат

By treating voicemail detection as a classifier with two error types, agents can choose cost‑effective strategies, reduce wasted minutes, improve conversion metrics, and provide accurate reporting to clients.

Рабочий процесс

• 1. Configure outbound agent to set start_speaker to agent and wait for the other side to speak.
• 2. Keep the first line short (e.g., "Hi, is this Mike?").
• 3. Decide voicemail action per campaign (hangup, leave short message, or none).
• 4. Script fixed short voicemail messages (≤15s, business name first, one reason, callback number twice, no sensitive info).
• 5. Pair voicemail with SMS via n8n or GoHighLevel if consented.
• 6. Define a closed list of call outcomes and store one outcome per call.
• 7. Cap retries (2–3 attempts, different days/times, one voicemail per contact).
• 8. Test against a variety of real greetings before launch.
• 9. Review recordings and outcome buckets to ensure correct classification and messaging.

Доказательства

If the agent treats a greeting like a person, three things happen.

Источник

The single most useful change is the first few seconds.

Источник

Every one of my exported agent configs uses the same voicemail action: "hangup".

Источник

I define a closed list of outcomes and make every call land in exactly one.

Источник

If a call hits voicemail, something has to decide whether and when to try again.

Источник