We're currently using the warm-transfer-experimental mode with voicemailDetectionType: "transcript" to transfer calls to a specialist through the AI agent. I’ve also tried switching to the "audio" type to compare behavior. The main issue is that when the specialist picks up the call, there’s a delay of around 10 to 15 seconds before the summary is provided, and only then is the caller connected. This delay is quite noticeable and negatively impacts the experience. I also tried using just a simple message instead of a full summaryPlan, but the delay still persists. Here’s the configuration we’re currently using:

typescript
destination: {
  type: "number",
  message,
  number,
  numberE164CheckEnabled: true,
  callerId,
  transferPlan: {
    mode: "warm-transfer-experimental",
    voicemailDetectionType: "transcript",
    fallbackPlan: {
      message: `It seems ${fullName || "the person"} is currently busy or on another call. Can I take a message?`,
      endCallEnabled: false,
    },
    summaryPlan: {
      enabled: true,
      messages: [
        {
          role: "system",
          content: `Start with greeting: 'Hello ${fullName || ""}, this is ${assistantName}, the digital assistant. I have {{caller_name}} on the line.' Replace the caller's name with the actual caller's name. Then summarize the call in 1-2 short sentences. Include the caller's main reason for the call and any key details needed for the agent to take over. Be concise and professional. After your summary, say: 'You are now connected to the caller. Please go ahead and speak.'`,
        },
        {
          role: "user",
          content: "Here is the transcript:\n\n{{transcript}}\n\n",
        },
      ],
    },
  },
}

Let me know if there’s any way to reduce this delay or if there's an alternative approach you'd recommend.

The 10–15 second delay is caused by the summary being generated only after the specialist answers to reduce this, switch to a “say-first” warm transfer mode or pre-generate the summary before the call is transferred..