CoPilot - Intent Actions Control
The CoPilot - Intent Actions Control allows designers to assign intents to matched in the transcribed interaction, with these intents either being displayed to the agent or hidden for internal purposes, and with the possibility of triggering other Fields when an intent is completed.
The Field isn't visible to the agent at all, and neither requires nor allows specific triggering other than its Page being loaded.
This Control is part of the CoPilot Control pack, and will only be available if your System has the appropriate licence.
Add a CoPilot - Intent Actions Field to the Page.
Clicking the Advanced Options tab will allow the configuration of the Integration - AI - GPT Connector to be used and the intents.
An Integration - AI - GPT Connector must be selected from the list at the top to specify the AI model and configuration to be used for all subsequent processing within the Workflow (unless another, later, CoPilot - Intent Actions Field overrides this).
Below this, the intents themselves can be created. Each intent comprises a few elements:
-
Name: if the intent is visible to the agent, then this will be displayed in the Intents tab of the
CoPilot widget,
-
Prompt: the prompt that the AI model will be assessing to determine whether it has been matched. This can be entered manually, or the wand button used to automatically generate a suggested prompt from the provided Name,
-
Speaker: which speaker's utterances will be checked for whether the intent has been matched. This can specify the agent, the customer, or either, but please note that the intent still needs to be matched within a single utterance even if "Either" is selected,
-
Visibility: whether the intent is visible in the Intents tab of the CoPilot widget. Items which are set to a Speaker of "Customer" are never displayed, as the agent cannot reliably control the caller's utterances, and trying to do so can create awkward interactions,
-
Action: an optional configuration point that allows specifying a Field that will be triggered in a manner similar to
Update Links or
.click().
The intents can be reordered by dragging the icon at the left-hand side, but this is just for usability and manageability as the ordering of the intents has no impact on functionality.
Multiple CoPilot - Intent Actions Fields can be added to a Page, or to a Workflow, without any problems. However, it should be noted that if trying to add an intent with a name that matches an existing intent in the agent's CoPilot widget that nothing will happen; the existing intent won't be overwritten or reset in any manner.
The actions specified for an intent won't trigger if the agent has left the Page where the intent was originally added, even if they simply reloaded the current Page or navigated away and back. The intent can still be marked as completed in the CoPilot widget, but the action itself won't trigger.
It may be desirable to change the utilised Integration - AI - GPT Connector at different times in the Workflow. For example, it may be desirable to change to a Connector with a lower "temperature" for sections where a very brief and consistent analysis or response in the
AI tab of the CoPilot widget are desirable, or to a higher temperature when desiring a more creative or verbose output.
All intents are measured against individual utterances, rather than across multiple utterances or the entire transcription; this is so as to avoid consuming excessive numbers of tokens with your generative AI provider every time the conversation advances, but also means that the prompts need to be carefully designed so as to avoid relying upon multi-utterance understanding.
Specifying an effective prompt is key to ensuring proper matching of intents, and therefore of operation of the CoPilot widget. The wand button should automatically generate a reasonable prompt based on the provided name, but this may need some adjustment if the generated prompt isn't suitable or misunderstands the intent.
