Talkif separates interruptions, caller pauses and AI-decision billing, with speech now generated roughly one sentence ahead of playback. After an interruption, voice billing covers only the sentence or two already prepared, not the remaining text that has not been generated. AI decisions carry a separate token-based charge per call. These are vendor behavior and billing claims, not measured savings or latency results. The context is telephone-conversation, July 2025; the archive photo is not a Talkif call or test.