Assurance

The “Assurance” tracker detects when a representative expresses commitment to solving a customer’s issue, offering reassurance or guaranteeing a resolution. Understanding these moments helps teams gauge the representative’s confidence, empathy, and problem-solving stance.

Who Benefits:

• Project Managers & Team Leads: Identify whether agents consistently provide reassurance, guiding training on instilling confidence and trust.
• Sales & CX Teams: Recognize patterns where agents reassure customers about product fixes or service improvements, influencing satisfaction and loyalty.
• Business Managers & Strategists: Spot trends in how often and effectively representatives promise solutions, refining scripts and support materials.
• Compliance Officers: Ensure no misleading or unauthorized assurances are given, maintaining integrity and adherence to policy.

Value Proposition:

• Building Trust & Confidence: Highlighting reassuring moments helps measure how representatives instill confidence in customers.
• Continuous Improvement: By analyzing where assurance is provided or lacking, teams can refine coaching efforts.
• Operational Efficiency: Structured JSON output integrates into dashboards, enabling trend analysis and strategic adjustments.

Data Dictionary

Objective:

Detect phrases in the conversation where an agent assures the customer that their problem will be solved. Rather than a numeric score, this tracker returns instances of matched phrases along with timestamps and confidence.

Proposed JSON Schema:

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

Data Dictionary:

If No Matches Found:
If no assurances are detected, return an empty matches array for the "Assurance" tracker or omit the tracker entirely. If no trackers match at all, return:

{
  "trackers": []
}

Determining Confidence & Thresholds

How the Assurance Tracker Works:

• Define a set of seed phrases that represent reassurance or commitment to solving the problem (e.g., "We will solve this for you", "I assure you we’ll fix it", "You can rely on us to handle this").
• The system interprets the semantic intent of these phrases and identifies similar statements in the transcript.
• Assign confidence based on how closely the detected phrases match the assurance intent.

Confidence & Calibration:

• Initial Testing: Use known calls to confirm that detected phrases truly signify reassurance.
• Refine Phrases: Add or remove seed phrases to improve accuracy.
• Thresholding: Downstream consumers can ignore low-confidence matches if false positives occur.
• Stakeholder Feedback: Incorporate input from supervisors or compliance officers to refine what constitutes a valid assurance.

Prompt Construction & Instructions

Role Specification & Reiteration:

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

No Hallucination:

• Only identify assurances present in the transcript.
• If uncertain, assign a lower confidence or omit the match.

Strict Formatting:

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

Example Prompt for Implementation

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

• A transcript
• A tracker named 'Assurance' with seed phrases indicating promises or reassurances of resolving issues.

Your task:

1. Detect phrases suggesting the agent assures the customer that their problem will be solved.
2. For each detected instance, return text, timestamp, and confidence.
3. Return only JSON as per the defined schema. No extra commentary.

Instructions & Variables (example):

{
  "tracker": {
    "name": "Assurance",
    "phrases": [
      "we will solve this for you",
      "you can rely on us to handle this",
      "rest assured we are working on it",
      "I assure you we’ll fix it"
    ]
  }
}

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

[TRANSCRIPT_JSON][TRANSCRIPT_JSON]"

Expected Output Format:

{
  "trackers": [
    {
      "name": "Assurance",
      "matches": [
        {
          "text": "I assure you we’ll fix it by tomorrow",
          "timestamp": "2024-07-10T11:22:00Z",
          "confidence": 0.93
        }
      ]
    }
  ]
}