Multi Entity Identification allows agent designers to attach more than one entity for recognition onto a question node in Copilot Studio, providing additional flexibility for customer self-service and leading to higher containment rates. This capability improves customer self-service by recognizing one of several valid inputs provided by a user, such as an account number or partial social security number. It also enables AI agents to offer alternatives when users do not have specific information, avoiding immediate escalation and enabling fallback options like additional questions or logic branches.

During runtime, if multiple entities are attached to a question node, the agent attempts to match the user's input to one of them, whether by voice or text. If multiple entities are detected, the agent matches only the first entity in the list. Designers should ensure prompt wording and follow-up logic account for this behavior. If no match is found, the agent follows the configured re-prompt treatment in the question node.

On the Copilot Studio authoring canvas, designers must perform the following steps:

Define and create the entities to capture.
Create a question node and craft the prompt to the user.
On the question node:
1. Under Identify, select One of Multiple Entities.
2. Select + New Entity and then select from the Entity picker.
3. Repeat for all applicable entities.

All responses are saved in a variable of type “record” that can be renamed by the author.

This recognition method identifies only one entity per user utterance. If a user input contains more than one valid entity, only the first entity listed in the configuration is recognized.

For example, a question node uses Multi Entity Identification, where the entities are Tracking Number and Delivery Address. A user says, “My package to be delivered to One Microsoft Way with the tracking number 7658641290753147”. The result is 7658641290753147, as the tracking number is the first entity in the list.

To access specific information from the record object, subvariables are automatically created based on the possible entities captured by the question node and displayed in dot notation, {recordName}.{entityName}. This is used to create conditional branches following the Multi Entity Identification step to route the information captured.

We recommend using the Is (not) Blank triage option in the conditional nodes to proceed logically based on what information is received by the user at question time.

The following is a list of known limitations for Multi Entity Identification:

Dual-tone Multi Frequency (DTMF) input for entity identification is not supported in this modality. However, it may exist for the entity on non-Multi Entity Identification-based question nodes based on the type of entity created.
Currently, a maximum of five entities can be within the recognition pool on any given question.
No external entity support.