Implementing Real-Time Agent Coaching Triggers based on Speech Analytics Events

How-To-Articles
1

Implementing Real-Time Agent Coaching Triggers based on Speech Analytics Events

What This Guide Covers

This masterclass details the implementation of Real-Time Coaching Triggers in Genesys Cloud. By the end of this guide, you will be able to architect a system that monitors live interactions for specific speech patterns (e.g., mention of a competitor, failure to say a required disclosure, or high-stress keywords) and automatically triggers an intervention. You will learn how to send real-time alerts to supervisors, push context-aware scripts to the agent’s screen, and automatically schedule coaching sessions in WFM when a critical performance gap is detected.

Prerequisites, Roles & Licensing

Real-time coaching requires the Genesys Cloud AI Experience and speech analytics capabilities.

  • Licensing: Genesys Cloud CX 3 OR CX 1/2 with AI Experience.
  • Permissions:
    • Speech Analytics > Program > View/Add
    • Quality > Evaluation > View/Add
    • WFM > Coaching > Add
  • OAuth Scopes: speech_analytics, quality, workforce_management.
  • Infrastructure: Genesys Cloud Native Transcription must be enabled for the target queues.

The Implementation Deep-Dive

1. Defining “High-Value” Speech Analytics Programs

A “Program” in speech analytics is a collection of topics (keywords and phrases) that the system listens for.

Architectural Reasoning:
Do not alert on every keyword. You must group them into Coaching Categories.

  • Compliance: “I must inform you…”, “recorded for quality…”.
  • Competitive Intelligence: “Competitor X”, “lower price elsewhere”.
  • Empathy/Stress: “Frustrated”, “Manager”, “Unacceptable”.

2. Implementing the Real-Time Notification Bridge

Speech analytics events occur in the cloud. To trigger a coaching action, you must bridge these events to your operational workflow.

Implementation Pattern:

  1. The Topic: Configure a topic in Admin > Quality > Speech and Text Analytics.
  2. The Event: Use Genesys Cloud EventBridge to subscribe to the v2.analytics.speech.topic.matches event.
  3. The Action: Route the event to an AWS Lambda function.

3. Automated “Coaching Appointment” Scheduling

Instead of just sending an email, use the WFM Coaching API to put time on the agent’s schedule automatically.

Logic Pattern (Lambda):

// Detect a "Compliance Failure" topic match
if (topicName === 'Missing_Mandatory_Disclosure') {
  // 1. Log the failure in the Agent's performance record
  await qualityClient.postEvaluation(conversationId, ...);
  
  // 2. Schedule a 15-minute Coaching Session in WFM
  await wfmClient.postCoachingAppointment({
    name: "Immediate Compliance Review",
    attendeeIds: [agentId],
    facilitatorId: supervisorId,
    lengthMinutes: 15,
    startAfterMinutes: 60 // Schedule it for 1 hour from now
  });
}

4. Agent-Facing “Live Assist” Prompts

The most effective coaching happens during the call.

Implementation Step:
Use the Agent Scripting API to push a notification to the agent’s workspace.

  • Trigger: Topic Match = “Competitor Mentioned”.
  • Action: Dynamically update the agent’s script to show the “Competitor X Battlecard” with current pricing and rebuttal points. This ensures the agent has the exact knowledge they need to save the sale in the moment.

Validation, Edge Cases & Troubleshooting

Edge Case 1: Transcription Accuracy and “Ghost” Alerts

  • The failure condition: The supervisor gets an alert that the agent was rude, but the agent actually said “I’m sorry you feel that way,” which the AI transcribed as “I feel that way.”
  • The root cause: Low confidence scores in the STT (Speech-to-Text) engine.
  • The solution: Implement Confidence Score Filtering. Only trigger coaching actions if the speech analytics topic match has a confidence score of > 85%.

Edge Case 2: Supervisor Fatigue (The “Alarm” Flood)

  • The failure condition: A supervisor has 50 agents, and the system sends them a Slack message for every minor “Empathy” slip, leading to them ignoring all alerts.
  • The root cause: No thresholding or prioritization of alerts.
  • The solution: Implement Cumulative Scoring. Instead of alerting on a single event, alert only if an agent misses a required disclosure in more than 3 calls within a single shift.

Official References