Glossary
›
Intent

Intent

Glossary
Also known as:
User intent, customer intent

What is an intent?

An intent is the goal behind what a person says or types, expressed as a named category such as apply for a loan, change a delivery address or dispute a charge, which software uses to decide how to respond and which process, team or system should handle the request. The same intent can be phrased in thousands of ways, and one message can carry more than one intent.

In marketing, intent has a related meaning: signals such as searches, page visits or questions that suggest someone is likely to buy. Both senses describe the same idea, inferring what someone wants from what they do. In customer conversations, the list of intents a system recognizes is, in effect, a model of the business seen from the customer's side: every reason someone gets in touch, written down and named.

How intents are defined

An intent set is designed, not discovered once. Five questions decide whether it works.

Design questionWhy it mattersExample
How broad is each intent?Too broad loses meaning; too narrow splits identical needsPayment problem, versus card declined, versus card declined abroad
What details does it need?An intent is incomplete without its parametersChange due date needs the account and the new date
What if several apply?Customers often ask for two things at once"Cancel my card and send a new one"
What is out of scope?Some requests should never be attempted automaticallyLegal advice, complaints about a named employee
Where does it lead?Each intent must map to an answer, an action or a personDispute a charge leads to the dispute process

There is also a gap between the stated intent and the underlying one. A customer who asks "what's the rate?" may really be asking whether the payment is affordable, and the most effective employees answer the second question as well as the first. Intent can also shift during a conversation: a call that starts as a cancellation can end as a renewal on better terms.

Why intents matter

Even a single industry produces a long list. When researchers built a public benchmark for detecting intents in banking, they labeled 13,083 customer service queries with 77 separate intents, many divided by fine distinctions, such as a card that has not arrived versus a card that is not working.

Getting intent right is the first decision in every automated conversation. A misread intent sends the customer down the wrong path, and everything that follows, however fluent, is wasted. In revenue conversations, intent is also a signal of readiness: questions about timing, documents or next steps usually mean a customer is closer to deciding than a general question about features.

Where intents are used

  • Routing. Sending calls, chats and emails to the team or process that handles that need.
  • Phone menus that accept speech. Callers say why they are calling instead of pressing keys.
  • Self-service flows. Each recognized intent starts a defined set of steps.
  • Contact-reason analytics. Measuring why customers get in touch, and how that changes over time.
  • Lead prioritization. Using buying signals to decide which prospects to contact first.

Language models can now interpret requests without a predefined list, which reduces the need to anticipate every phrasing. Businesses still benefit from named intents, because reporting, routing rules and compliance controls all need stable categories to attach to.

What a good intent set looks like

  • Built from real conversations. Intents reflect how customers actually describe their needs, not how the organization is structured.
  • Clearly separated. Each intent has a definition and examples that distinguish it from its neighbors.
  • An explicit out-of-scope category. Requests the system should not handle have somewhere safe to go.
  • Multi-intent handling. The system can recognize and sequence two requests in one message.
  • Mapped to outcomes. Every intent leads somewhere, and that destination is reviewed.
  • Maintained. New products, policies and seasons produce new intents, and the set is updated to match.

How to measure intents

  • Coverage. The share of real requests that match a defined intent.
  • Misroute rate. Conversations that reached the wrong process and had to be redirected.
  • Fallback rate. Requests the system could not place, by channel and over time.
  • Intent-to-outcome rate. For revenue intents such as apply or renew, the share that end in the completed outcome.
  • New-intent discovery. Recurring requests found in unmatched traffic that deserve their own category.

Intents that trigger legal obligations

Some intents carry legal weight the moment a customer expresses them, whatever words are used. Three US examples show the pattern; sector rules and other countries add many more.

  • Stop means stop. Under FCC rules adopted in 2024, consumers may revoke consent to automated calls and texts by any reasonable means. A reply to a text such as "stop", "quit", "end", "revoke", "opt out", "cancel" or "unsubscribe" counts as a revocation, and it must be honored within 10 business days.
  • Cancellation requests. California's automatic renewal law requires businesses to let consumers cancel through the same medium they used to sign up, and to offer online cancellation, through a prominent direct link or button, to those who signed up online.
  • Inconvenient times in collections. Under Regulation F, a debt collector may not contact a consumer at a time or place it knows or should know is inconvenient, and its official interpretation treats a consumer saying something like "I cannot talk now" as enough to put the collector on notice.

Frequently asked questions

What is the difference between an intent and an entity?

The intent is what the person wants, such as transfer money. Entities are the details needed to do it, such as the amount, the account and the date.

Do AI agents still need predefined intents?

Less than older systems did, because language models can interpret unfamiliar phrasing. Named intents remain useful for routing, reporting and controls that must apply consistently.

What is a fallback intent?

The category a system uses when it cannot match a request to any defined intent. A good fallback asks a clarifying question or hands the conversation to a person.

How many intents does a business need?

Enough to separate requests that need different handling, and no more. Most organizations find a small number of intents covers most volume, with a long tail of rarer ones.

What is buyer intent?

In marketing, signals that a person is likely to purchase, such as searching for pricing, comparing products or requesting a quote. It applies the same idea to buying behavior rather than to a single message.

Read the intent. Then act on it.

Encore AI's agents recognize what each customer is trying to do and carry it through to the completed application or arrangement, drawing on how your best people handle the same request.

Scale your best to be the best.

Book a Demo