Add-on | AI agents - Advanced |
Conversation flows for email are our newest addition to improve the performance of AI agents for email. The conversations flows for email feature is in early access, if you would like to try it out, please contact your Customer Success Manager.
What are conversation flows for email?
Conversation flows for email open up higher automation potential for your AI agents by enabling multiple interactions between the AI agent and the customer. Let’s go through an example, a customer emails in a query, but some key details are missing, like an order ID. Now, the AI agent is able to follow up and request the missing information, before then going on to fully resolve the customer’s query - enabling true, end-to-end automation.
Conversation flows for email are built on a new integration with Zendesk, which notifies the AI agent about new customer reply, as well as recognizing when a human agent needs to get involved. By using all these data points, you can keep track of reopen rates and measure the AI agent’s automation performance.
Based on your feedback, our capabilities are constantly being improved, and new features are being added to improve the experience of conversing with AI agents. If you would like to be part of the early access program please contact your Customer Success Manager.
Conversation design for conversation flows for email
The early access program focuses on automating the first triggered intent in each conversation. When building dialogues with multiple AI agent replies in mind, you need to consider additional replies from your customer, as well as decision points in the conversation, when escalating to a human agent would be the best next step.
With conversation flows for email, the following new node types have been added to the Dialogue Builder for ticket:
- Visitor message, with the following children node configurations:
- Intent Predicted - use if you want to allow the customer to trigger another intent, following it with a Link to reply node.
- Entity Recognized - use when expecting a Content Entity in the response.
- Free Text Written - use as a fallback, or follow with a Conditional node to analyze the content of the reply.
- Link to reply
- Multiple Ticket Replies
Conversation flows for email scenarios
As the early access program supports automating the first triggered intent every conversation, after the dialogue of the triggered intent is completed, the AI agent will add a tag to your ticket, indicating that the automation is complete. Every ticket that will not receive more responses from the AI agent is tagged with escalated_by_ultimate. The tag is applied in the following scenarios:
Single-reply dialogues
Single-reply dialogues work the same way as without conversation flows. If nothing follows a Ticket Reply node in the Dialogue Builder, the automation is considered to be completed.
Multiple-reply dialogues
Multiple-reply dialogues can utilize the new block options available for Ticket Automation: more than one Ticket Reply, Visitor Message, and the Link to reply blocks.
These dialogues allow the AI agent to send follow-up questions to confirm the customer’s intent or to ask for missing information. Additional messages from the visitor can be processed and used during the automation.
The number of blocks is not limited, but as a rule of thumb, you should aim for no more than 3 messages to gather the required information to automate a case.
Using Link to reply
While Conversation flows for email is limited to one intent per conversation, linking allows you to continue the conversation in another dialogue. The Link to reply node therefore is not considered to be the end of the conversation, even though it is the final node in the dialogue. You can link to other single- or multi-reply dialogues.
Escalating from a multiple-reply dialogue
In case an escalation is required in a multiple-reply scenario, following the Visitor message or Conditional node, you can add an empty Ticket Reply node. The empty reply adds the escalation tag to your ticket in Zendesk, can trigger additional actions, and will not send a reply to your customer, so a human agent can take over.
Events and Actions
For AI Agents for email, four events are supported today, which are also triggered in multiple-reply scenarios:
- Ticket Received by Ultimate - This event is only triggered once per conversation, when the first customer message is received
- Ticket Processed by Ultimate - This event and related actions are triggered with each Ticket Reply.
- Ticker reply delay events - Delay events are triggered for each reply. When using delays with Email Conversation Flows, it is not recommended to set too long reply delays (recommended 10-30 minutes).
Conversation Logs
In the Conversation Logs, you can seethe full message list of the conversation, including messages after escalation and agents' messages.
Analytics
New statuses for AI agents for email are being developed. With additional data being available on the tickets, reopen rates can now be calculated in the the AI agents - Advanced dashboard.
Limitations
As conversation flows for email are being developed, we would love to hear your feedback on improvement opportunities.
We are launching the early access program with the following limitations:
- Automation is limited to one intent per conversation.
- No rule-based escalations.
- No AI agent replies to tickets older than 3 days.
- New analytics metrics are in development.
- Email conversations don't support tickets created through the Zendesk API.
Setting up conversation flows for email
To enter the early access program, contact your Customer Support Manager, so they can enable the feature for you.
AI agents - Advanced relies on the escalated_by_ultimate tag in Zendesk Support, to know when it should provide an answer. Once the feature is enabled, AI agents - Advanced will start adding the tag to all tickets that have been processed by the AI agent. Without multiple-reply flows, this means all the tickets that have been sent to AI agents - Advanced, including not answered tickets, like Processed or Not Understood cases.
To enable the AI Agent to send multiple replies, you need to do two things:
- Update your AI agents - Advanced automation trigger in Zendesk
- Build a dialogue with multiple Ticket Reply nodes
If your AI agent is already live, you need to wait at least 3 days from enabling the feature to changing the automation trigger in Zendesk. This allows the AI Agent to mark all conversations completed, so in case a ticket is reopened, it won’t be accidentally answered by the AI Agent (as stated in the limitations and improvements section, the AI agent will not provide any replies to tickets older than 3 days).
If you are still building your AI agent, for the first 72 hours, you should modify your trigger in Zendesk. In the ANY section, make sure only the Ticket is Created criteria is present. After 72 hours, you can re-add the Ticket is Updated criteria.
What to change in the AI agents - Advanced automation trigger?
In Zendesk’s Admin Center, navigate to the Objects and rules > Business rules > Triggers menu item and find your AI agents - Advanced Ticket Automation trigger.
By default, you should see the following conditions:
To allow the AI Agent to react to additional customer messages, you need to delete the Ticket > Ticket is Created condition and replace it with:
- Ticket > Comment is Public - This ensures that the AI agent is notified of each customer, agent, or AI agent reply.
- Ticket details > Current user is not YOUR AI AGENT - As the AI agent now listens to all replies, we need to exclude it to avoid loops.
If your setup is more complicated, just make sure nothing interferes with these conditions.
Moving to the JSON body of the trigger, you need to update the object’s properties:
You can copy and paste the following object (make sure to replace the placeholder with your AI agent ID):
{
"botId": "YOUR_BOT_ID",
"message": "**{{ticket.title}}** \n\n {{ticket.latest_comment_rich}}",
"conversationId": "{{ticket.id}}",
"type": "message_multitouch",
"tags": "{{ticket.tags}}",
"requester": "{{ticket.requester.id}}",
"currentUser": "{{current_user.id}}",
"createdAt": "{{ticket.created_at_with_timestamp}}"
}
Changes to the JSON body include the new requester, currentUser, and createdAt properties. These enable AI agents - Advanced to differentiate between agent and customer replies and to not reply to tickets older than three days. The type has also been changed to “message_multitouch” for analytical purposes.
If you have a new AI agent or have not modified the automation trigger, you can simply remove and re-add it in the Settings > CRM Integration > Integration menu once conversations flows for email have been enabled for you.
0 comments