Short answer
The intent is what the customer wants; entities are the specifics the request carries. In "cancel order 4817 and refund it to my original card", the intent is cancel my order. The entities are the order number and the refund destination. Intent and entity together turn a message into something a system can act on: one names the goal, the other supplies the details.Resolution needs both, because they do different jobs. Intent selects the procedure: cancel-order triggers a different resolution path than change-address or where-is-my-order. Entities fill that procedure's parameters: which order, which address, what date, what amount. Entity extraction is the step that pulls those specifics out of natural language, reading "next Tuesday" as a date and "my work address" as a location, so the procedure runs on real values instead of guesses. In conversational AI, entities are the bridge between free text and structured action.
Systems that have one without the other fail in predictable ways. Detect intent vs entity extraction alone, and the system knows what to do but does it to the wrong thing: the right cancellation procedure aimed at the wrong order, the right refund for the wrong amount. Extract entities without intent, and the system holds perfect data with no idea what to do: it has the order number, the date, and the amount, and still cannot choose an action. The first failure damages customers; the second just describes them.
Intent vs entity at a glance
| Dimension | Intent | Entity |
|---|---|---|
| Question it answers | what does the customer want? | what specifics does the request carry? |
| Example | cancel my order | order 4817, next Tuesday, the work address, the amount |
| What it drives | which procedure runs | the parameters the procedure runs with |
Aide, the AI platform for customer service, is built on this pairing. Every inbound conversation is classified by intent against the Customer Intent Map, and the entities it carries are read against live systems before anything runs: the order number is checked against the real order, the address against the account on file, the amount against the actual invoice. The procedure that executes is the one the intent selected, filled with verified values rather than extracted guesses.