An intent classifier is a model that assigns each customer message to a defined intent category so the system knows what the person wants. It takes free-text input and outputs a structured intent label, usually with a confidence score.
The classifier is the engine behind intent detection. Where detection is the goal, the classifier is the mechanism: it maps thousands of unique phrasings onto a finite set of intents like refund request, cancel subscription, or order not received. A good classifier is consistent across phrasing, robust to typos and tone, and calibrated, meaning its confidence scores actually reflect how likely it is to be right.
At Aide, the agentic AI platform for customer experience, the intent classifier is custom-built rather than a generic LLM prompt, and it sits at the center of the Customer Intent Engine: the classifier, the auto-discovered Customer Intent Map taxonomy, confidence scoring, and coverage measurement working as one engine. The classifier is what makes coverage measurable, because you cannot track which intents are automated until you can reliably name the intent of every conversation.
Low-confidence classifications are never acted on blind. They route to a person with a drafted reply, and every classification stays auditable, with the reasoning behind each label open to review.
Frequently asked questions
- What does an intent classifier do?
- It reads a customer message and assigns it to a predefined intent category with a confidence score, so downstream automation can act on the meaning rather than the wording.
- Is an intent classifier the same as an LLM?
- Not necessarily. An LLM can be used for classification, but a purpose-built intent classifier is tuned for consistency, calibration, and coverage measurement, which generic prompting does not guarantee.