[object Object]

Freshdesk’s Conversation Intelligence (CI) suite uses Freddy to score sentiment, detect intents, and flag escalation risk in every ticket. The dashboards are pretty. The actions are what matter.

The four signals worth wiring

  1. Sentiment trajectory (not last-message sentiment)
  2. Escalation language detection (“speak to your manager”)
  3. Topic clustering for emerging issues
  4. Agent script adherence (Enterprise tier)

Sentiment trajectory triggers

Single-message sentiment is noisy. Trajectory across the last three messages predicts churn. Configure a workflow rule that fires when the trajectory drops two levels in a thread.

Trigger: sentiment_score(now) - sentiment_score(thread_start) <= -2
Action:
  → Add tag "at_risk"
  → Notify CSM via Slack
  → Increase priority to High

Escalation language as intent

Freddy detects phrases like “this is unacceptable” or “I want a refund.” Treat these as routing signals, not just alerts. Auto-assign to a senior agent group when detected.

Topic clustering: weekly cadence

The Topic Discovery report groups recent tickets into emergent clusters. Run it every Monday. New clusters beyond your known taxonomy point to product issues, outage symptoms, or marketing-driven confusion.

Agent coaching from CI data

Filter CI by individual agent over 30 days. Look for negative-sentiment ratio higher than team median. Pair the metric with three sample tickets for coaching, not raw shame.

API access for downstream use

Pull CI scores into your data warehouse:

GET /api/v2/conversations/{id}/intelligence
// Returns: sentiment_score, intent_labels, risk_score

Join this with Salesforce or your billing system to compute churn correlation, not just CSAT.

What not to do

  • Do not assign tickets to “the most positive agent” — survivor bias
  • Do not auto-discount based on sentiment alone — gaming risk
  • Do not surface raw scores to customers in a self-service portal

What to do this week

Build the sentiment trajectory workflow, schedule the Monday topic clustering review, and pull last quarter’s CI data into your warehouse to validate the churn correlation.

[object Object]
Share