Implementing Voicebot Call Containment Rate Tracking and Optimization Strategies

How-To-Articles
1

Implementing Voicebot Call Containment Rate Tracking and Optimization Strategies

What This Guide Covers

  • You will architect a robust measurement framework to track “Call Containment”-the percentage of interactions successfully resolved by a voicebot without escalating to a live agent.
  • You will implement custom milestones and participant data tagging in Genesys Cloud Architect to capture why and where customers exit the bot flow.
  • When complete, you will have the data-driven insights needed to identify “intent friction” and optimize your conversational AI to improve containment rates and reduce operational costs.

Prerequisites, Roles & Licensing

  • Licensing: Genesys Cloud CX 1, 2, or 3 with a Voicebot integration (Dialog Engine, Lex, or Dialogflow).
  • Permissions:
    • Architect > Bot Flow > View/Edit
    • Analytics > Conversation Detail > View
    • Routing > Queue > View
  • OAuth Scopes: architect, analytics.
  • Reporting Tools: Access to Genesys Cloud Analytics Dashboards or the Conversation Details API.

The Implementation Deep-Dive

1. Defining “Containment” in an Omnichannel World

The most critical architectural step is defining what “Success” looks like for your bot.

The Definition:

  • Contained: The customer completed their intent (e.g., checked a balance, changed a password) and disconnected without ever requesting an agent or being transferred to a queue.
  • Deflected: The customer was offered a digital channel (e.g., SMS) and moved away from the voice channel.
  • Escalated: The customer requested an agent, or the bot failed to understand them and triggered a fallback transfer.

Architectural Reasoning: You must distinguish between “Technical Containment” (the customer hung up) and “Functional Containment” (the customer’s problem was solved).

2. Implementing Milestone Tagging in Architect

To track where customers drop out, you must use Architect Milestones.

The Step:

  1. In Architect, go to Admin > Architect > Milestones.
  2. Create milestones such as Bot_Started, Intent_Identified, Self_Service_Success, and Escalation_Requested.
  3. In your Bot Flow, place the Register Milestone action at each critical junction.
  4. The Trap: Avoid “Milestone Overload.” If you have 50 milestones in a 2-minute flow, your analytics will be too noisy to identify real trends. Focus on the 4-5 major “Success/Failure” gates.

3. Tagging Outcomes via Participant Data

While milestones are great for flow analysis, Participant Data is better for high-level reporting and API-based extraction.

The Step:

  1. At every “Successful Disconnect” point in the bot flow, use the Set Participant Data action.
  2. Key: ATTR_BOT_OUTCOME, Value: CONTAINED.
  3. At the escalation point, set ATTR_BOT_OUTCOME to ESCALATED and add ATTR_BOT_EXIT_REASON (e.g., ConfidenceThresholdFail).
  4. Architectural Reasoning: By tagging the outcome on the conversation, you can run a single Analytics Query to find all “Contained” calls without parsing the entire flow path.

4. Analyzing the “Why” (Intent Friction)

A low containment rate is usually caused by “Intent Friction”-customers using phrases the bot doesn’t understand.

The Step:

  1. Use the Genesys Cloud Analytics API to fetch conversations where ATTR_BOT_OUTCOME == ESCALATED.
  2. Cross-reference the Utterance History for those interactions.
  3. Identify the “Dead Intents” (intents that frequently lead to escalation).
  4. Optimization: Re-train the NLU model with the utterances that caused the failures.

[THE TRAP]
Many organizations celebrate a “90% Containment Rate” only to find their CSAT is plummeting. This is the “Containment Trap”-where customers are so frustrated by the bot that they hang up in anger. This looks like “Containment” in the logs but is actually a failure. Always cross-reference Containment with NPS/CSAT and Repeat Caller Rate (within 24 hours).

Validation, Edge Cases & Troubleshooting

Edge Case 1: The “Hanging Up in the Menu” Problem

  • The Failure: 20% of callers hang up in the first 10 seconds of the bot.
  • The Root Cause: The introduction is too long or the “Menu” is too complex.
  • The Solution: Implement a Menu_Abandon milestone. If this is high, use A/B Testing to try a shorter greeting or a “Natural Language” open-ended prompt (“How can I help?”) instead of a list of options.

Edge Case 2: Intent “Bounce”

  • The Failure: The customer identifies an intent (e.g., “Billing”), but then escalates immediately.
  • The Root Cause: The bot is asking for information the customer doesn’t have ready (e.g., “Please enter your 16-digit meter number”).
  • The Solution: Use Data Actions to look up customer info via ANI before the bot starts. If you know who they are, the bot can say “I see you’re calling about your billing for account 1234, is that right?”

Edge Case 3: False Containment via Disconnect

  • The Failure: A system error causes the call to drop, which is counted as “Contained” because it never reached an agent.
  • The Root Cause: Inbound flows don’t distinguish between a “User Hangup” and a “System Error” disconnect by default.
  • The Solution: Use a Set Participant Data action as the very last step before a successful Disconnect action. If the call ends without this flag, it should be categorized as an Error_Drop.

Official References