Intent Drift Index

Customers often begin a conversation with one goal and finish with another. Maybe a “Where’s my order?” turns into a “Please cancel it.” The Intent Drift Index (IDI) quantifies that shift so operations leaders can see how often calls veer off course—and fix the root causes before they snowball.

Why it matters

Routing accuracy. If intents drift frequently, your IVR or chatbot is funneling people to the wrong queue.
Agent performance. High drift can mean agents aren’t clarifying the customer’s real need early enough.
Process gaps. Repeating drift patterns flag upstream issues (e.g., unclear billing emails) that drive avoidable call volume. Monitoring how intent evolves is key to proving that a fix actually reduced those repeat contacts (cxtoday.com).

How to calculate it

For each conversation:

IDI =  ( Σ drift_t ) / (N – 1)

where:
N          = total customer turns in the conversation
drift_t    = 1 if intent_t ≠ intent_0, else 0
intent_t   = predicted intent label for turn t
intent_0   = intent label for the customer’s opening turn

The result ranges from 0 (no drift) to 1 (intent changes every turn). At a higher level, aggregate the average IDI across all interactions for the day, queue, or agent to spot trends.

Example

A call with five customer turns produces these intent labels: Billing > Billing > Billing > Cancellation > Cancellation

Drifts occur at turn 3 (Billing→Cancellation). IDI = 1 drift / (5 – 1) = 0.25

Data requirements

Turn-level transcripts segmented by speaker.
An intent classifier that labels each customer utterance (many CCaaS platforms already surface intent per message) (support.zendesk.com).
A taxonomy of intents tight enough to be actionable but broad enough to capture real shifts.

Interpreting the score

IDI Range	Interpretation	Common Fixes
0.00 – 0.10	Calls stay on topic.	Keep monitoring.
0.11 – 0.30	Some mis-routing or unclear self-service content.	Review IVR menus, update FAQs.
> 0.30	Frequent pivots; customers or agents lack clarity.	Tighten discovery questions, coach agents on first-turn probing, audit marketing/ops comms.

Best practices

Pair with First Contact Resolution. A stable intent that still ends unresolved signals a different issue than a drifting one.
Drill into drift direction. From “How do I?” to “I want out” is a warning sign. From “Account help” to “Upgrade” may signal an opportunity.
Automate alerts. Flag live calls where the IDI crosses a threshold so supervisors can intervene before churn risk spikes. Predictive intent models already guide real-time pairing and escalation decisions (behavioralsignals.com).

References

How to Capture and Analyse Customer Intent, CX Today. (cxtoday.com)
Predicting Intent and Its Impact on Call Centers, BehavioralSignals. (behavioralsignals.com)
Automatically Detecting Customer Intent, Sentiment, and Language, Zendesk Support. (support.zendesk.com)