Alternatives

The “Alternatives” tracker detects when participants in a conversation suggest different approaches, workarounds, or solutions to a problem. By recognizing these mentions, teams gain insights into how participants navigate obstacles, brainstorm options, and foster innovation or problem-solving.

Who Benefits:

• Project Managers & Team Leads: Understand which team members propose alternate solutions, encouraging creativity and adaptability.
• Sales & CX Teams: Identify when agents provide multiple product options or alternate features, informing training on solution-focused selling.
• Business Managers & Strategists: Spot trends in alternative-solution offering, guiding product improvements, or refining problem-solving approaches.
• Compliance Officers: Confirm that representatives only propose approved solutions, ensuring adherence to guidelines and avoiding unauthorized suggestions.

Value Proposition:

• Actionable Insights: Detection of alternative solutions helps teams understand how conversations steer towards options and flexibility.
• Efficiency & Scalability: Structured JSON output integrates into dashboards or analytics tools, streamlining monitoring and decision-making.
• Continuous Improvement: By tracking how often and effectively alternatives are offered, organizations can refine playbooks, training, and support materials.

Data Dictionary

Objective:

Detect phrases that indicate “alternative solutions” during a conversation. Rather than a numeric score, trackers return occurrences of matched phrases (or semantically similar phrases). This data helps teams understand the frequency, timing, and context of offering alternatives.

Proposed JSON Schema for Tracker Output:

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

Data Dictionary:

If No Matches Found:
Return an empty matches array for the "Alternatives" tracker or omit the tracker from the trackers array entirely. If no trackers are matched at all, return:

{
  "trackers": []
}

Determining Confidence & Thresholds

How Trackers Work:

• A tracker is defined by a set of example phrases or keywords that represent its intent (e.g., "we could try something else," "another approach might be," "have you considered a different option?").
• The system uses these phrases as a seed to understand the semantic intent of offering alternatives.
• During analysis, the system detects semantically similar phrases, assigning confidence based on how closely they align with the tracker’s intent.

Confidence & Calibration:

• Initial Testing: Evaluate known calls to see if detected phrases truly represent alternatives.
• Refine Phrases: Add or remove seed phrases to better capture the intended meaning of “alternatives.”
• Thresholding: Downstream consumers can filter out low-confidence matches (e.g., below 0.5) if false positives occur.
• Stakeholder Feedback: Incorporate insights from sales leads, PMs, or compliance to refine the tracker’s vocabulary over time.

Prompt Construction & Instructions

Role Specification & Reiteration:

• Cast the system as a “highly experienced assistant” specializing in identifying tracker matches.
• Reiterate instructions: no extraneous commentary, return only JSON.

No Hallucination:

• Only return matches if clearly indicated by the transcript.
• If uncertain, assign lower confidence or omit.

Strict Formatting:

• Return only JSON with trackers array, following the schema.
• If none found, return an empty trackers array.

Example Prompt for Implementation

System Message (Role: System):
"You are a highly experienced assistant that identifies instances of 'Alternatives' intent in a conversation. You have:

• A transcript
• A predefined tracker named 'Alternatives' with seed phrases representing offering different solutions.

Your task:

1. Identify phrases in the transcript that suggest offering alternative solutions.
2. For each detected instance, return text, a timestamp, and a confidence.
3. Return these matches in a JSON structure under the trackers array.
4. No extra commentary.

Instructions & Variables (example):

{
  "tracker": {
    "name": "Alternatives",
    "phrases": [
      "another option could be",
      "have you considered another approach",
      "we might try a different solution",
      "an alternative method might be"
    ]
  }
}

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

[TRANSCRIPT_JSON][TRANSCRIPT_JSON]"

Expected Output Format:

{
  "trackers": [
    {
      "name": "Alternatives",
      "matches": [
        {
          "text": "We could also try outsourcing this task",
          "timestamp": "2024-07-10T11:17:00Z",
          "confidence": 0.92
        }
      ]
    }
  ]
}