Named Entity Recognition (NER) Defined
Named entity recognition (NER) is a natural language processing technique that finds and labels specific pieces of information in text, such as names, order numbers, dates, product names, and locations. In customer service, NER lets AI pull the details it needs from a message so it can look up records or take action.
A customer writes, "My order 88213 was supposed to arrive in Denver on Friday but it's still in Chicago." Before any AI can help, it has to notice the order number, the two cities, and the date. Picking out those details is named entity recognition.
What is named entity recognition?
Named entity recognition is the task of finding entities in text and assigning each one a type. An entity is a specific, real-world thing that can be named or identified. Common types include:
- People: "Maria Lopez"
- Organizations: "Acme Corp"
- Locations: "Denver," "Chicago"
- Dates and times: "Friday," "last March"
- Money and quantities: "$49.99," "three items"
- Domain-specific identifiers: order numbers, account IDs, tracking numbers, product SKUs, plan names
NER is one part of natural language understanding. If intent recognition answers "what does the customer want?", NER answers "which specific things are they talking about?"
Why named entity recognition matters
- Enables actions: an AI cannot look up an order or change a booking without the order or booking number
- Fewer questions for the customer: if the details are already in the message, the AI does not need to ask again
- Better routing: entities like a product line or country can send the conversation to the right team
- Structured data from unstructured text: extracted entities can populate ticket fields and power reporting, such as which products generate the most issues
- Privacy controls: NER can detect personal data, such as card numbers or email addresses, so systems can redact or protect it
How named entity recognition works
- Tokenize the text: split the message into words and sub-words.
- Analyze context: a model looks at each word and its neighbors. "Paris" could be a city or a person's name; context decides.
- Label spans: the model marks which words form each entity and assigns a type (for example, "88213" is an order number).
- Normalize values: convert entities into standard formats, such as turning "next Friday" into a specific date.
- Use the result: pass the entities to the next step, such as a database lookup or an API call.
Older NER systems used hand-written rules and patterns, like "a five-digit number after the word order." Later systems used machine learning models trained on labeled examples. Today, large language models can extract entities from almost any phrasing with no task-specific training, and can be instructed to return them in a structured format.
For the example above, NER output might look like this:
| Text | Entity type |
|---|---|
| 88213 | Order number |
| Denver | Location (destination) |
| Friday | Date (expected delivery) |
| Chicago | Location (current) |
NER vs. intent recognition
| Named entity recognition | Intent recognition | |
|---|---|---|
| Question it answers | Which specific things are mentioned? | What does the customer want to do? |
| Example output | Order number: 88213 | Intent: track a delayed order |
| Used for | Lookups, form filling, redaction | Routing, choosing a workflow |
Most AI agents use both at once: intent decides what to do, and entities supply the details needed to do it.
How Fin uses entity extraction
Fin reads the details customers include in their messages, such as order numbers, email addresses, and dates, and uses them when calling connected systems. When a Procedure needs a specific piece of information that the customer has not provided, Fin asks for it, then continues the workflow.
Frequently asked questions
Is NER still needed with large language models?
The task is still needed, but it is usually no longer a separate model. LLMs perform entity extraction as part of understanding a message, though some systems still use dedicated NER for speed, cost, or redaction.
What is a custom entity?
A custom entity is a type specific to one business, such as a plan tier, a product code, or an internal ticket ID. Custom entities often matter more in customer service than general types like people or places.