Verified AI summary ◀▼
Auto assist is an AI-powered tool that helps agents handle tickets by suggesting replies and actions based on procedures, help center articles, or similar solved tickets. It reduces repetitive tasks and ensures consistent responses. Agents review and approve suggestions, which are tailored to the conversation's tone. Auto assist doesn't work on AI agent tickets and requires agent approval for all actions.
Auto assist is an AI-powered assistant for your agents. Using large language model (LLM) technology, auto assist understands the contents of submitted tickets and makes suggestions to your agents on how to solve them.
With auto assist, agents spend less time on repetitive tickets, solve tickets in a more consistent way, and ultimately close more tickets.
Watch the demo video below to see auto assist in action, or read on for more information about auto assist. To get started, see Turning on and configuring auto assist.
Zendesk agent copilot featuring auto assist (4:43)
Understanding auto assist and how it works
Auto assist makes suggestions to your agents based on procedures that you've created or that have been automatically generated, generic information from its LLM training, your public help center articles, or similar solved tickets. These suggestions can be replies that the agent should send to the customer or actions that the agent should take.
Auto assist suggestions appear in tickets in the Agent Workspace interface in place of the composer. Any replies or actions suggested by auto assist require an agent’s approval before being sent to the customer or performed, and are sent or performed under the agent’s name. The customer isn’t aware of auto assist during their interactions with the human agent.
When using auto assist, agents don’t have to compose replies or execute actions themselves, except when following instructions you've defined. Instead, agents review and approve the replies or actions suggested by auto assist. Replies suggested by auto assist are automatically adapted to the conversation's tone and writing style so that agents can reduce the time spent editing responses. When making suggestions, auto assist takes into account all public ticket comments in the conversation, including previously generated comments approved by the agent. Suggestions are made in response to end-user comments only, though, not agent comments.
If your procedures include direct instructions for agents, then auto assist suggests the instructions to agents when they need them based on the flow of their conversation. Instructions are steps that agents should perform manually. For example, your procedures might include instructions for checking your internal stock management system if auto assist can't access it.
To learn more about the agent experience with auto assist, see Using auto assist to solve tickets.
Understanding procedures for auto assist
A procedure is a set of written steps that tells auto assist how to advise an agent in solving a user's request. Auto assist can suggest replies on any topic, but to achieve best performance, creating procedures is essential. You can write your own procedures, or review, edit, and publish AI-generated procedures.
You can link the following directly in procedures to ensure that auto assist makes the right suggestion or performs the correct action for a user's request:
- Standard actions
-
Custom actions
Custom actions are currently limited to changing the ticket status and ticket assignee, standard and custom ticket field updates, tag updates, custom API actions, action flows, and certain Shopify actions.
- Macros, if they're available to all agents
- Agent instructions, which tell agents how to manually perform steps that auto assist can't
- Public help center articles or other procedures
You can also reference certain ticket fields in your procedures.
Auto assist knows which procedures to follow for a given customer request as a result of its LLM training. You don’t need to map specific procedures to specific intents or other ticket metadata in order for auto assist to suggest replies or actions. Instead, auto assist understands the content of a customer’s request out of the box and can map it to an existing procedure, if one has been created that addresses the topic the customer is asking about.
About ticket fields in procedures
Auto assist has access to read the following standard ticket fields:
- Assignee email
- Assignee name
- Brand
- Priority
- Requester email
- Requester name
- Status
- Subject
- Type
- Tags (can read and update but not remove)
The Priority, Type, and Tags standard fields can be updated by auto assist.
- Checkbox
- Date
- Decimal
- Dropdown
- Number
- Regex
You can write procedures in plain language that references these ticket fields and auto assist will use this information to suggest the best responses.
For example, you can leverage ticket fields in auto assist procedures in the following ways:
- Brand-aware procedures
- Select the correct refund duration from an existing procedure based on the ticket brand (for example, Brand X is 20 days, but Brand Y is 30 days).
- Use the account verification procedure for Brand X because the ticket brand is X.
- Email confirmation
- If the requester’s email address is example@acme.com, proceed with the refund process.
About help center content in procedures
You can insert the following help center content in a procedure:
-
Articles: You can insert public, published articles in a procedure.
When you insert an article, you can choose which locale will be used.
By inserting an article in a procedure, you're telling auto assist that the content of the article is important in the context of the procedure.
For example, say that a customer's order must be verified before approving a refund. You can enter a step to gather information from an article with the following wording:
"Before approving the refund, verify if the customer's order is meeting the requirements listed in <inserted article>"
Keep in mind that auto assist can't comprehend images, URLs, or videos in articles. If an article relies on this type of content, then auto assist won't be able to understand the full picture.
It's recommended to keep your help center articles up to date so that auto assist can make the most accurate suggestions.
-
Procedures: You can insert published procedures in a procedure. When
you insert a procedure within a procedure, you're telling auto assist to
follow the steps in the linked procedure.
For example, say that a customer indicated they want to upgrade their subscription plan. You can enter a step that links to a specific procedure based on some conditions with the following wording:
"If the customer indicated they want to upgrade their subscription plan, follow step X of <inserted procedure>"
You can insert up to four pieces of help center content into a procedure. For example, add a mix of articles and procedures, such as three articles and one procedure.
Understanding how auto assist makes suggestions
Auto assist can suggest replies and actions based on the information provided in procedures and its generic LLM training information. It can also suggest replies based on your public help center articles, or similar solved tickets, when a relevant procedure or article isn't available. It can't suggest replies or actions based on other sources.
Auto assist can suggest replies on any topic based on the following sources:
- AI-generated procedures or procedures that you've created. Procedures are
essential to achieving the best performance.
Auto assist uses this procedure content, along with its built-in LLM training, to generate suggested replies or actions, which it then presents to agents as they work on tickets in the Agent Workspace.
Procedure content can also include direct agent instructions. These types of instructions are steps that agents should follow and perform manually.
- If a relevant procedure isn't found, then auto assist uses your public help center articles to generate suggested replies. Only articles with a brand and language matching the ticket brand and language are used for reply generation. For example, if the brand on a ticket is Obscura, then only articles with the Obscura brand are considered.
- If a relevant procedure or help center article isn’t available, then auto assist uses similar solved tickets to generate suggested replies.
Auto assist suggestions appear in tickets in the Agent Workspace interface in place of the composer. If you have permission, you can view the source auto assist used to generate the suggestion. See Viewing sources for auto assist suggestions to learn more.
If you don't have permission to view an auto assist suggestion's source, then refer to the list above to understand what source may have been used to generate the suggestion.
Examples where auto assist can suggest replies include:
- Advising the customer on buying a product or service sold by your company
- Explaining a business policy to the customer
- Helping the customer troubleshoot a problem with a product
Examples of procedures where auto assist can suggest replies and actions include:
- Checking the status of a current order in Shopify, or canceling and refunding specific items or whole orders in Shopify
- Querying and modifying your own internal business systems or performing third-party actions via API
- Telling an agent to check your internal stock management system and adding a refunded item as 'back in stock'
- Instructing an agent on how to create and validate a new user account in your company's human resources system