Triage analyzes ticket content to predict, categorize, and label, helping you route tickets to the right team or facilitate escalation. These predictions are generated by models trained on historical data specific to a field or by general models such as sentiment or spam detection. Now, you can create, configure, and deploy your own Triage LLM model to meet your business needs. This guide provides information on how you can create your own Triage Large Language Model and how you can use it to enhance the efficiency of ticket routing within your help desk.
Why does Triage LLM matter?
The Triage Large Language Model (LLM) makes customer support more efficient and effective. Using AI, it accurately predicts and categorizes support tickets, ensuring that they are routed to the most appropriate teams. This reduces the time it takes to resolve issues which ultimately improves customer satisfaction.
Pre-requisite
Before creating a model, ensure that your help desk is connected to Forethought by navigating to Settings > Integrations.
Note: Currently, the Triage LLM is only available to customers using Zendesk and Salesforce. Customers using other helpdesk systems will not see the "Create model" button.
Setup
1. Go to Triage > Classifiers > Create Model.
2. Enter your model name and a short model description.
3. In the Labels tab, click Create Label. When the model makes a prediction, this label will be the predicted field value in the ticket.
4. To help the AI better classify tickets, add at least 5-10 training phrases. Training phrases should typically be a few words to a sentence long. It must also encompass what a user may typically say to associate that ticket with a specific field.
For example, if you want to create a label about “Login issue”, the training phrases for this can be:
-
- I forgot my password
- I can’t access my account
- Unable to sign in
- Help me reset my password
- Why can't I log in with my email?
5. Click Create. You can add as many labels as you want to your model. However, when a model has multiple labels, it will select the best-matching label to assign to the output field you configure in the next step. For best practices, see Best Practices in Triage LLM.
6. Besides the Labels tab, click the Configuration tab to select an output field. The output field is the helpdesk field, to which we write a label.
7. Optional: Select a default label. If the AI is unable to identify the label, the default label will be used as opposed to leaving the field blank.
8. To ensure your model is working correctly, click Test. The confidence threshold is set at 50%. If the confidence falls below 50%, the AI will not produce any output unless you set a default label in the Configuration tab.
9. Map labels from the model to your help desk.
10. Once you are done, click Publish.
Important: Once you publish your model, it can't be edited. However, you can create a duplicate and make changes to the duplicated version.
Duplicate a model
1. To duplicate your model, click Duplicate to edit.
2. Enter a name for your duplicated model.
3. Click the icon next to the Test button to view the different versions of your model. You can publish, duplicate, and delete the selected version. However, the live version cannot be deleted; it can only be duplicated.
Value mapping
Accurate mapping is essential for your triage model to work effectively. When the user uses a dropdown field, it's important to ensure that the output label matches the tag. Otherwise, the information won't be written to the helpdesk correctly. This happens to both Zendesk and Salesforce. Thus, configuring mapping enables us to ensure that we can match the label value to the corresponding options on the helpdesk.
To map values, follow these steps:
1. Click the mapping icon.
2. This opens the map drawer, allowing you to select which labels correspond to the output values for the output field you selected.
3. Once done, click Save.
Use case
Scenario
A customer contacts the support team of an online store because one of the items she received is missing. She ordered four cat blankets but only received three.
Set up
To address these types of issues, you have set up your Triage LLM named “Order Issues” to categorize problems with customers' orders. In this model, you created three labels: “Order Incorrect,” “Missing Item,” and “Damaged Product”. You set the label descriptions as follows:
-
- Order incorrect - Customer receives items that are not what they ordered.
- Missing item - Items listed in the order are not present in the received package.
- Damaged product - Items arrived in a damaged or defective condition.
How your Triage Large Language Model (LLM) works
When the customer reports that her order is missing a cat blanket, your Triage LLM accurately predicts, classifies, and labels the ticket as "Missing Item." This directs the support ticket to the appropriate queue or agent group within your helpdesk. This process streamlines response times, ultimately enhancing your support team's efficiency and effectiveness.
Conclusion
Creating and implementing the Triage Large Language Model is a transformative step for optimizing your customer support operations. By harnessing the power of AI to predict and categorize tickets, you can ensure efficient routing to the right teams, reduce resolution times, and ultimately enhance customer satisfaction.