Revisit

The “Revisit” tracker detects instances where the representative or customer mentions the need to circle back to a topic at a future time—whether due to insufficient information, pending decisions, or time constraints. Understanding these moments helps teams keep track of unresolved subjects that warrant follow-up, ensuring nothing falls through the cracks.

Who Benefits:

• Project Managers & Team Leads: Identify deferred items that require future action, guiding agenda planning, resource allocation, or scheduling follow-up calls.
• Sales & CX Teams: Recognize when customers or agents agree to revisit certain product features, pricing discussions, or service issues, informing proactive follow-ups.
• Business Managers & Strategists: Spot patterns in delayed decisions or pending topics, refining processes or policies to reduce recurring deferrals.
• Compliance Officers: Ensure that required callbacks, verifications, or compliance checks set for future revisit are tracked and fulfilled.

Value Proposition:

• Actionable Follow-Ups: Highlighting revisit points aids in ensuring that deferred topics are not forgotten, improving service completeness.
• Efficiency & Scalability: Structured JSON output integrates easily into dashboards or task managers, streamlining tracking of outstanding issues.
• Continuous Improvement: Understanding why certain topics need revisiting guides process enhancements to reduce deferrals over time.

Data Dictionary

Objective:

Detect phrases in the conversation that indicate a topic will be revisited or addressed later. The tracker returns occurrences of matched phrases, along with timestamps and confidence. This data assists in ensuring that no pending subjects remain unresolved.

Proposed JSON Schema:

{
  "trackers": [
    {
      "name": "Revisit",
      "matches": [
        {
          "text": "<string>",
          "timestamp": "<ISO 8601 or relative time>",
          "confidence": <number> // 0.0 to 1.0
        }
      ]
    }
  ]
}

Data Dictionary:

If No Matches Found:
If no revisit-related phrases are detected, return an empty matches array for "Revisit" or omit the tracker. If no trackers match at all, return:

{
  "trackers": []
}

Determining Confidence & Thresholds

How the Revisit Tracker Works:

• Seed the tracker with phrases that suggest revisiting a topic later (e.g., "let’s circle back," "we can address this next time," "we'll discuss this issue later," "we’ll follow up on this").
• The system interprets these seed phrases and identifies semantically related statements in the transcript.
• Confidence indicates how closely the detected phrase aligns with the concept of revisiting a topic in the future.

Confidence & Calibration:

• Initial Testing: Validate on known calls to ensure detected phrases accurately represent future revisit intentions.
• Refine Phrases: Add or remove seed phrases to capture evolving language patterns.
• Thresholding: Downstream consumers can filter out low-confidence matches if needed.
• Stakeholder Feedback: Include input from team leads or compliance officers to refine what constitutes a meaningful revisit mention.

Prompt Construction & Instructions

Role Specification & Reiteration:

• Present the system as a “highly experienced assistant” identifying revisit mentions.
• Reiterate instructions: return only JSON, follow schema, no extra commentary.

No Hallucination:

• Only detect revisit mentions if supported by transcript evidence.
• If uncertain, assign a lower confidence or omit matches.

Strict Formatting:

• Return only JSON with trackers.
• If none found, return empty trackers.

Example Prompt for Implementation

System Message (Role: System):
"You are a highly experienced assistant that identifies 'Revisit' mentions in a conversation. You have:

• A transcript
• A 'Revisit' tracker defined by seed phrases indicating a topic should be discussed again later.

Your task:

1. Detect phrases suggesting the need to revisit or follow up on a topic in the future.
2. Return text, timestamp, and confidence.
3. Return only JSON as defined. No extra commentary.

Instructions & Variables (example):

{
  "tracker": {
    "name": "Revisit",
    "phrases": [
      "circle back to this",
      "address this next call",
      "follow up on this topic",
      "discuss this again later"
    ]
  }
}

User Message (Role: User):
"Analyze the following transcript and detect 'Revisit' mentions:

[TRANSCRIPT_JSON][TRANSCRIPT_JSON]"

Expected Output Format:

{
  "trackers": [
    {
      "name": "Revisit",
      "matches": [
        {
          "text": "Let’s circle back to this issue after we get more data.",
          "timestamp": "2024-07-10T11:50:00Z",
          "confidence": 0.9
        }
      ]
    }
  ]
}