Intent inference
Intent inference is deriving what a user was trying to accomplish from what they observably did: classifying sessions by goal (compare plans, find an invoice, cancel) from navigation, interaction, and timing patterns. It sits one interpretive level above frustration signals, which detect difficulty without knowing the goal.
How it works
Behavioral data underdetermines intent, so inference is probabilistic classification, not reading minds: models (rule-based, statistical, or LLM-over-digest) assign likely goals from path shape, dwell patterns, and the objects interacted with, calibrated against sessions where the intent became explicit (a completed task labels its own past). The output layer changes what analytics can say: instead of "high exit rate on the pricing page," intent-segmented analysis says "comparers complete, but users arriving to find enterprise terms leave," which names a fix. The failure modes deserve equal billing, since misclassified intent poisons every downstream number, and honest deployments report confidence and keep the underlying sessions one click away for verification, the same evidence discipline the rest of session analytics runs on.
Where you encounter it
AI analytics features, support deflection analysis, and personalization systems that condition on inferred goals.