Analyzing intelligent triage results and taking action

Attention: You must have the Advanced AI add-on to use the features described in this article.

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 gathering, analyzing, and acting on intelligent triage information.

For more information about intelligent triage, see Intelligent triage resources.

This article contains the following topics:

• Gathering intelligent triage data
• Analyzing and fine-tuning the results

Related articles:

• Making sense of unexpected intelligent triage predictions
• Using intelligent triage to identify and act on ticket escalations

Gathering intelligent triage data

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:

1. 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.
2. Enable intelligent triage to start allowing tickets in your account to be categorized with an intent, language, sentiment, or all three.
   Tip: If you don't want your agents to see the Intelligence section of the context panel at this point, you can hide it. See Configuring Intelligence.
3. 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.
4. 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?	Make sure your agents are trained to handle the most common types of requests. Create knowledge base articles and other self-serve resources to deal with easy requests (like password resets), and automate responses to these types of tickets using macros. Make sure you have agents who speak the same languages as your customers.
Are there any intents that would make sense for you to group together?	Create views that group similar intents together, and route them to the agents who are best suited to deal with them.
Are the predicted intents and languages consistent with the initial message on each ticket?	If not, give feedback to Zendesk so we can improve the prediction model. You can also report on tickets where the intent was manually changed by an agent, signaling that the initial prediction didn't hit the mark.
What trends are there in customer sentiment? Are negative-sentiment tickets especially prevalent for a specific product or category?	If you receive many negative-sentiment tickets, it might be worth giving agents additional training on diffusing tough situations. If a specific product or category receives many negative-sentiment tickets, investigate whether there are improvements to the product that need to be made.

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	Create a trigger that raises the priority for tickets with certain intents so that agents get to the ticket faster.
CSAT is low for tickets in a particular language	Route tickets of a particular language directly to the agents who are fluent in that language.
Tickets about a certain topic always require more information from an agent before they can be solved	Exclude these tickets from the normal “Ticket is created” trigger and instead create a trigger that sends the customer a message which prompts them to provide the needed information. Consider updating your ticket forms to collect the required information as part of the original ticket.

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.