Intelligent triage is an AI-powered feature that automatically detects what a ticket is about (its intent), what language it's written in, and whether the customer's message is positive or negative (its sentiment). You can use this information to route tickets to the right groups automatically, create views to group similar types of requests, and report on trends in the types of tickets your customers are submitting.
Because intelligent triage can affect different areas of your ticket workflows, you might not know exactly where to start at first. This article discusses some best practices for getting started with intelligent triage.
For more information about intelligent triage, see Intelligent triage resources.
This article contains the following topics:
Understanding how it works and turning it on
Intelligent triage can have a powerful effect on your agents' workflows, saving them anywhere from 30 to 60 seconds per ticket by automatically identifying and routing a ticket based on its intent, language, or sentiment.
However, before you make any changes to your triage or routing workflows, it's helpful to understand exactly how intelligent triage categorizes the tickets in your account. Getting to know the specific intent values and trends in your account will help you decide which workflow changes will best improve the agent and customer experience.
In general, we recommend getting started with these four steps:
- Understand how intelligent triage works from ticket submission to resolution. You should also understand how the system populates intent, language, and sentiment values on tickets.
- Enable intelligent triage to start allowing tickets in your account to be categorized with an intent, language, sentiment, or all three.
- Build reports to analyze intelligent triage results to see trends in your tickets. As you get started, consider building separate reports for intent, language, and sentiment to allow you to focus on one prediction type at a time.
- Wait for approximately two weeks to allow for a sufficient sample size of tickets to be enriched by intelligent triage.
Analyzing and fine-tuning the results
After a couple of weeks, intelligent triage should have enriched enough tickets for you to be able to decide which actions to take. The following sections present some additional points to consider as you perform this analysis.
Identify trends in the predicted intents, languages, and sentiments
First, take a look at the reports you built above and review the High and Medium confidence tickets. Look for trends, and decide whether you want to take action to improve them.
|Trend||Actions to consider|
|What are the most prevalent intents and languages?||
|Are there any intents that would make sense for you to group together?||
|Are the predicted intents and languages consistent with the initial message on each ticket?||
|What trends are there in customer sentiment? Are negative-sentiment tickets especially prevalent for a specific product or category?||
Decide which metrics you want to improve
Next, decide which metrics matter the most to your team. Do you want to raise CSAT ratings, meet SLAs more consistently, improve first response time, reduce group assignments, or something else?
Start by targeting one or two metrics, or perhaps a subset of intents, and consider how workflow changes can improve the overall experience. Target those areas first to get the maximum impact from intelligent triage.
|Trend||Action to consider|
|Low first reply times on urgent issues||
|CSAT is low for tickets in a particular language||
|Tickets about a certain topic always require more information from an agent before they can be solved||
Design, implement, and report in an iterative process
Regardless of the changes you decide to make, remember that this is an iterative process. You will identify trends, make changes accordingly, track the success of those changes, and repeat.
Here are some questions to consider as you design, implement, and report on your workflow changes:
- What is the highest level of confidence needed for the workflow to be effective? For example, is it acceptable to send all tickets with a certain intent to a designated group and ask that they manually reroute if the intent was wrong, or should only tickets with a High confidence level be routed to that group?
- Should the workflow apply a tag or update some other ticket attribute to allow for easier reporting in the future?
Establish two-way communication with your agents
Inform your agents of any changes you make so that they’re equipped to provide feedback on them, both good and bad.
For example, consider setting up a macro to tag tickets where the agent has feedback, and include an internal note where they can record their feedback about the workflow.
Ask your agents about particular pain points they have with tickets. If there is a particular group of intents where they see complications, brainstorm ways to adapt your workflows to improve the agent and customer experience.
Can you clarify what is different about detecting the language in this feature, vs people not using this? How is this different from the email-based language detection that Zendesk uses for all accounts currently?
Hi Jake Bantz, will this work for live chat tickets?
CJ Johnson - Intelligent triage will detect the language for all new tickets for any channels enabled for intelligent triage. The alternate methods of language detection are focused particularly on the user creation event. For example, intelligent triage can be used for detecting the language for tickets which belong to users who had the wrong language assigned (via one of the methods/exceptions in the above article) or are multilingual and may contact support in their most comfortable language. This allows for the proper language considerations to be made on every ticket - given that it was created via a supported and intelligent triage enabled channel.
Riah Lao - Intelligent triage does not work for live chat or messaging. It is only available for asynchronous channels such as email, web form, and API. You can read more about the configuration here.
Is there a way in which i can set the triggers and automations to exclude the triage based on presence of a tag added on a previous trigger/automation?
Ian Marston do you mean you want intelligent triage to ignore specific tickets altogether based on certain conditions?
If that's the case - at this time the only way to include/exclude tickets from being enriched is based on ticket channel and the option to exclude agent created tickets (in the admin settings for intelligent triage).
If that's too broad of a change, my recommendation would be leaning on business rule order and using the presence of tags to override what actions may be implemented with conditions based on the intelligent triage conditions in later triggers/automations.
With that in mind, what sort of exclusion conditions would you like to see? Could this be something that doesn't have to reference a trigger applied action, but instead some other parameter(s)? If so, what sorts of parameters?
The intelligent triage enrichment is kicked off at virtually the same time as the ticket is created, and at that point only the subject and message body are consumed for the predictions. If additional checks are added for tags or other values added after creation, that would mean added delay in adding the enrichment.
Please sign in to leave a comment.