Implementing Conversational Analytics to Identify High-Value Automation Opportunities

How-To-Articles
1

Implementing Conversational Analytics to Identify High-Value Automation Opportunities

What This Guide Covers

  • Utilizing Genesys Cloud Speech and Text Analytics to systematically discover intent clusters across thousands of unstructured agent-customer interactions.
  • Architecting a data-driven pipeline to calculate the ROI of building new Bot flows by cross-referencing intent volume with Average Handle Time (AHT) and Agent transfer rates.
  • The end result is an objective, mathematical framework for prioritizing conversational AI development (e.g., “We should build a Password Reset bot first because it occurs 5,000 times a month with an AHT of 8 minutes”).

Prerequisites, Roles & Licensing

  • Licensing: Genesys Cloud CX 3 (WEM/Speech Analytics).
  • Permissions: Analytics > Speech and Text Analytics > View, Analytics > Topic > Edit.
  • Infrastructure: A baseline of at least 30 days of recorded voice and digital interactions with Speech Analytics enabled.

The Implementation Deep-Dive

1. The Flawed “Guessing Game” of Bot Development

When organizations purchase an AI bot platform, business leaders usually hold a brainstorming session to decide what the bot should do. They typically guess: “Customers probably want to check their account balance.”

The Trap:
Building a bot based on assumptions leads to low utilization. You might spend three weeks building a complex “Account Balance” integration, only to discover that 80% of customers are actually calling because they don’t understand the complex wording on page 3 of their paper invoice. If you don’t use conversational data to drive your roadmap, you are building in the dark.

2. Configuring Topic Miner for Discovery

Genesys Cloud Topic Miner uses unsupervised machine learning to group phrases from transcripts into clusters without requiring you to define keywords upfront.

Implementation Steps:

  1. Navigate to Admin > Quality > Topic Miner.
  2. Create a new mining session. Select a target queue (e.g., Customer_Service_Tier1) and a date range of the last 30 days.
  3. Crucial Filter: Filter the session to only analyze interactions where the Agent transferred the call or the AHT was extremely short (under 2 minutes) or extremely long (over 10 minutes). Short calls are usually highly transactional (easy to automate). Long calls are complex (hard to automate).
  4. Run the miner. The system will analyze the transcripts and group them into natural clusters.

3. Creating Targeted Speech Topics

Once Topic Miner identifies a cluster (e.g., “Reset Password”, “Can’t Login”, “Account Locked”), you must formalize it into a trackable Speech Topic.

Implementation Steps:

  1. In the Topic Miner results, highlight the relevant cluster and click “Create Topic.”
  2. Name the topic Intent_Password_Reset.
  3. Refining the Dialect: Add strict phrases. Use proximity rules. For example, [password] near [reset] or [account] near [locked].
  4. Exclude irrelevant phrases (e.g., NOT [password] near [wifi], if you are a SaaS company and don’t troubleshoot home networks).
  5. Publish the Topic. Genesys Cloud will now tag all future interactions containing these phrases with the Intent_Password_Reset tag.

4. Calculating the Automation ROI

Now that you are actively tagging interactions, you must extract the data to prove the value of automating this intent.

Implementation Steps:

  1. Navigate to Performance > Workspace > Interactions.
  2. Filter by the Intent_Password_Reset topic.
  3. Export the data via the Analytics API POST /api/v2/analytics/conversations/details/query.
  4. In your data visualization tool (PowerBI/Tableau), calculate the ROI:
    • Volume: 4,500 calls per month.
    • Average Handle Time (AHT): 6.5 minutes.
    • Total Hours: (4,500 * 6.5) / 60 = 487 hours per month.
    • Agent Cost: 487 hours * $25/hour = $12,187 per month.
  5. If you build a Voicebot that successfully contains 50% of these calls, the projected savings is ~$6,000 per month. This data completely justifies the development time required to build the bot.

Validation, Edge Cases & Troubleshooting

Edge Case 1: The “Agent Parroting” Skew

  • The Failure Condition: Your Speech Topic for “Cancel Service” shows massive volume. You build a cancellation bot, but nobody uses it. Upon reviewing the transcripts, you realize customers weren’t asking to cancel; the agents were asking: “Are you sure you don’t want to cancel your service today?” as part of an up-sell script.
  • The Root Cause: Standard Topic searches look at both the Customer and the Internal (Agent) channel.
  • The Solution: When configuring your Speech Topic phrases, strictly enforce the channel. Set the rule to: Phrase: "cancel service" -> Participant: External. This ensures the topic only fires when the customer speaks the phrase.

Edge Case 2: IVR Abandonment Misclassification

  • The Failure Condition: Topic Miner struggles to identify intents for interactions that abandon in the IVR before reaching an agent, because there is no human conversation to transcribe.
  • The Root Cause: Speech Analytics requires a connected media path to generate a transcript.
  • The Solution: To identify automation opportunities for abandoned calls, you must rely on IVR Milestones, not Speech Analytics. Ensure your legacy IVR menus use Set Flow Milestone actions at every node. Export the milestone data via the Analytics API to identify where callers are dropping (e.g., 40% of callers abandon after hearing the “Press 1 for Billing” prompt, indicating a high-friction area ripe for conversational AI replacement).

Official References