Intent models in Talkdesk CXA Operations Center™ (formerly AI Trainer) are classification models used to automatically uncover the customer’s intentions during calls. They represent the basis for conversational agents, such as Talkdesk Autopilot, but, moreover, they are the main triggers for any automated task or analytics in a contact center.
Today, Talkdesk Generative AI capabilities support the training of intents more quickly and easily, through the use of the Generate Phrases button when editing or adding new training phrases.
Table of Contents
- Why must Intent Models be Trained?
- Creating an Intent
- Creating and Managing Custom Entities
- Filling all the Slots
- Managing the Agreement System
Why must Intent Models be Trained?
To improve the automation rate, resolve cases faster, promote customer satisfaction, and know what your end customers are talking about. CXA Operations Center can be used to:
- Observe the intent model performance in production.
- Train models with real-world data from conversations where intent classification was not satisfactory.
- Publish tweaks and improvements to production.
The Intents page shows the following:
- The “Observe” section [1] provides information to your contact center staff about model performance in the production environment.
- The “Train” section [2] is a main work area where it is possible to manage intents, training phrases, and entities. The “Inbox” section contains trainable data, collected from live conversations, that could be used to improve the intent model.
- The” Deploy” section [3] has two functions: to review suggestions for model improvements, and to manage versions of the model in production.
In CXA Operations Center, you can manage intents for intent models: create, update, or delete intents, as well as add training phrases to intents. Training phrases can be added directly, as made-up or synthetic data, or by labeling real-world data through the Agreement system (explained in the subsection).
Creating an Intent
To create an Intent, please follow these steps:
On the Intents page, click on New Intent [1].
Add the “Intent name” [2], “Intent purpose”, and “Confidence Threshold” [3].
Note: Be aware that filling in “Intent purpose” will support better intent training phrase generation with the help of Generative AI. For these scenarios, ensure you provide a concise intent name and a descriptive intent purpose of the business need behind the set of training phrases you expect to auto-generate.
By clicking New phrase [4], add or accept at least a few training phrases, which will be either auto-generated [5] or self-typed [6].
Add “Training phrases” [5] and click Save[6].
Note: The suggested minimum number of training phrases per intent is 10.
When you’re done, publish your changes [7].
Creating and Managing Custom Entities
Entities are metadata associated with an intent. For example, if a speaker mentions specific products, places, or qualities, that information can be extracted in the context of the intent that is detected.
With CXA Operations Center, you can manage custom entities and also annotate training phrases, so they are extracted within the context of an intent.
To create a Custom entity, please do the following:
On the Entities page, select New entity [1].
Give the new entity a “Name” [2]. Add entity values and synonyms [3]. Click Create [4] to publish the changes.
For more detailed information on the best practices for training an intent model, please consult the full article here.
Filling all the Slots
In Autopilot, in a conversation, it is useful, and sometimes necessary, to extract annotated entities (parameters) when an agent is detecting a contact’s intent. If the parameter is required but not detected, then Autopilot will ask for this information and fill in all the slots in the conversation.
In CXA Operations Center, it is possible to configure parameters, automate prompts for when a parameter is not detected, as well as determine automatic responses for a detected intent.
To configure a parameter, please do the following:
- Open the training phrase [1] containing words to be annotated.
- Highlight words [2] and a list of entities will appear [3].
- Save[4] the training phrase and the created parameters.
- Open the list of parameters [5] and select one to edit [6].
- Tick the option “If enabled, allows configuring prompt responses” [7] to add an automatic prompt, and to fill in the slot defined by the parameter.
- At the end, select Save [8].
Managing the Agreement System
CXA Operations Center ensures that agents, as domain experts, can label data without decreasing the model’s performance by accidentally deploying the changes to production.
This is possible through the Agreement process, which happens in stages:
- Multiple agents can independently make suggestions for a model change.
- Supervisors can review these suggestions and decide to move on with the change or decline it.
- Post-review, only supervisors can deploy models to production.
In the Inbox section, Agents can see all the data that can be labeled. Each Agent has their own Inbox. “Agent A” may believe a phrase should be labeled as "student loan", while “Agent B” may believe otherwise. This is what defines the overall agreement percentage (%) when the phrase moves to quality check.
Agents can select one phrase to label [1], and either ignore it [2], flag it for later [3] (“To-do later”), or suggest this phrase as a training phrase for intents [4] (Train intent). Click Save[5] to apply the changes.
Note: Agents see only their own intent suggestions. However, users with access to the Quality Check page can view suggestions from all contributors.
On the “Quality check” page, supervisors can see all agents’ suggestions. To review them, please follow these steps:
Click on the suggestion [1].
Click on the Start review button.
Approve each suggestion based on the level of agreement between agents [2], and go to the Launch section [3].
Click on the reviewed suggestions [4] and on the “More actions” button (...) [5] to choose whether to launch the changes right away or schedule the deployment in the future [6].
Once the model is trained, check the “Intents” page to confirm if the new training phrases are properly added.
How is the data on the Trends dashboard calculated?
The Observe Trends page in Talkdesk CXA Operations Center™ (formerly AI Trainer) → Observe shows Volume as the number of phrases that were Matched or Detected across intents for the selected period.
Volume, Matched, and Detected in this context mean:
- Volume: The total count of phrases represented in the chart for the selected time period.
- Matched: Phrases that were correctly identified as belonging to a trained intent.
- Detected: Phrases the model flagged as expressing an intent, whether or not that match was confirmed as correct.
Important Information:
The Trends graphic only shows your top 10 intents
- The chart displays the 10 intents with the highest volume for the selected period — not every intent in your account.
- If you sum the Matched/Detected counts from a full per-intent export and compare that total to the number shown on the Trends graphic, they will not match. The export includes every intent; the chart (and any total read off it) reflects only the top 10.
- A difference between the “dashboard graphic total” and the “full export total” is expected behavior, not a data integrity issue.
If you need a total across all intents, use the full export rather than summing values read off the chart.