| Question you are asking | Dataset to use |
|---|---|
| How many of my conversations involved AI Expert Assist? | Engagement Performance |
| Which AI Expert Assist agents are being activated most, and in which queues? | Agentic AI Expert Assist Agent |
| Which skills are doing the work? Which are configured but never firing? | Agentic AI Expert Assist Skill |
| Which tools are slow, failing to engage agents, or driving load? | Agentic AI Expert Assist Tool |
| Which knowledge base articles are being surfaced, and does anyone read them? | Agentic AI Expert Assist KB Article |
| Are the suggestions any good? What do agents think of them? | Agentic AI Expert Assist Suggestion |
AI Expert Assist Usage statuses
AI Expert Assist Usage is the headline adoption attribute available on the engagement grain. Use this field for adoption reporting and for splitting handle time between assisted and unassisted engagements.
| Status | Meaning |
|---|---|
| Not used | AI Expert Assist was either unavailable for this engagement or was available but nothing triggered. |
| Used, no interaction | AI Expert Assist triggered, but the human agent did not give any feedback (passive usage). The engagement qualifies if either the agent took an action (asking a question in Ask AI Expert Assist, running a manual KB retrieval, or clicking any tab inside the AI Expert Assist panel) or AI Expert Assist served at least one suggestion. |
| Used and interacted | Everything above, plus at least one qualifying interaction with what AI Expert Assist produced. |
Note: Enabling a skill does not make an engagement used. The skill has to actually produce a suggestion.
What counts as an interaction
The following seven actions define an interaction:
- Copy
- Insert
- Thumbs up or thumbs down
- View full article (knowledge base suggestions only)
- Switch language
- Pin
- Tool call Click to Action (CTA)
AI Expert Assist Flagged values
AI Expert Assist Flagged records whether AI Expert Assist flagged the engagement for supervisor attention.
Values: Flagged or Not flagged.
Agentic AI Expert Assist Agent dataset
One row represents one AI Expert Assist agent activation, in one queue, for one engagement.
Use this dataset to see which AI Expert Assist agents are being put to work, where they are deployed, and whether their output is landing with human agents. If you have configured several agents for different queues or lines of business, this is where you compare them.
Metrics
| Metric | Definition | Type | Calculation | Aggregation |
|---|---|---|---|---|
| AI Expert Assist Agent Usage | The number of AI Expert Assist agent invocations | Number | Counts each activation of an AI Expert Assist agent. An agent activated across three queues within one engagement counts as three | SUM |
| AI Expert Assist Agent Interacted Usage | The share of invocations in which the human agent interacted with any AI Expert Assist feature | Number | Invocations with at least one of the seven qualifying interactions | SUM |
| AI Expert Assist Suggestion Count | The total number of suggestions the AI Expert Assist agent served | Number | Counts unique suggestions served by the AI Expert Assist agent | SUM |
Dimensions
| Dimension | Definition | Filter | Group by |
|---|---|---|---|
| AI Expert Assist Agent | The name of the AI Expert Assist agent | Yes, multi-select | Yes |
| Queue | The queue the engagement is associated with | Yes, multi-select | Yes |
| Agent | The human agent who handled the engagement | Yes, multi-select | Yes |
| Direction | Inbound or Outbound | Yes | Yes |
| Engagement ID | The unique identifier of the engagement | Yes, multi-select | No |
Time dimensions
Year, Quarter, Month, Day, Hour, 30 minutes, and 15 minutes.
Agentic AI Expert Assist Skill dataset
One row represents one execution of a skill.
Use this dataset to find out which skills carry the load and which are configured but effectively dormant. Pair skill invocation counts with tool counts to see which skills are lightweight and which drive heavy downstream execution.
Metrics
| Metric | Definition | Type | Calculation | Aggregation |
|---|---|---|---|---|
| AI Expert Assist Skill Invocation Count | The number of skill invocations produced per skill | Number | Counts the suggestions in which the skill was invoked | SUM |
| AI Expert Assist Tool Count | The number of tool executions triggered by the skill | Number | Counts tool invocations attributable to the skill | SUM |
Dimensions
| Dimension | Definition | Filter | Group by |
|---|---|---|---|
| AI Expert Assist Skill Type | Whether the skill is built-in or custom | Yes, multi-select | Yes |
| AI Expert Assist Skill | Name of the skill. Only skills actually used in engagements are populated. | Yes, multi-select | Yes |
| AI Expert Assist Agent | The AI Expert Assist agent used in the engagement | Yes | Yes |
| Queue | The queue the engagement is associated with. An engagement can span multiple queues. | Yes | Yes |
| Agent | The human agent who handled the engagement | Yes | Yes |
| Direction | Inbound or Outbound | Yes | Yes |
| Engagement ID | The unique identifier of the engagement | Yes, multi-select | No |
| Channel | The channel that the engagement was in when it entered the queue | Yes | No |
| Source | The source that the engagement was in when it entered the queue | Yes | No |
Time dimensions
Year, Quarter, Month, Day, Hour, 30 minutes, and 15 minutes.
Agentic AI Expert Assist Tool dataset
One row represents one tool execution.
Use this dataset to assess tool reliability and responsiveness. High latency shows up directly as dead air in the agent's workflow, and a tool with high execution volume but near-zero agent interaction signals that the output is not useful.
Metrics
| Metric | Definition | Type | Calculation | Aggregation |
|---|---|---|---|---|
| AI Expert Assist Tool Count | The number of tool executions | Number | Counts tool invocations by the tool | SUM |
| AI Expert Assist Tool Latency | The latency of tool calls | Number | Time from tool execution start to tool response, aggregated by the method you select | SUM, AVG, MIN, MAX |
| Agent Interaction Count | The number of tool calls the human agent acted on | Number | Counts tool calls in which the human agent engaged with the tool call CTA (if configured) | SUM |
Dimensions
| Dimension | Definition | Filter | Group by |
|---|---|---|---|
| AI Expert Assist Tool | Name of the tool. Only tools actually used in engagements are populated. | Yes, multi-select | Yes |
| AI Expert Assist Skill | Name of the skill that invoked the tool | Yes, multi-select | Yes |
| AI Expert Assist Skill Type | Whether the skill is built-in or custom | Yes, multi-select | Yes |
| AI Expert Assist Agent | The AI Expert Assist agent used in the engagement | Yes | Yes |
| Queue | The queue the engagement is associated with | Yes | Yes |
| Agent | The human agent who handled the engagement | Yes | Yes |
| Direction | Inbound or Outbound | Yes | Yes |
| Engagement ID | The unique identifier of the engagement | Yes, multi-select | No |
| Channel | The channel that the engagement was in when it entered the queue | Yes | No |
| Source | The source that the engagement was in when it entered the queue | Yes | No |
Time dimensions
Year, Quarter, Month, Day, Hour, 30 minutes, and 15 minutes.
Agentic AI Expert Assist Knowledge Base Article dataset
One row represents one knowledge base article retrieval.
Use this dataset to evaluate knowledge base coverage and content quality. Articles served with no agent interaction usually mean the retrieval is off-target or the article content is not useful enough to open; both are actionable.
Metrics
| Metric | Definition | Type | Calculation | Aggregation |
|---|---|---|---|---|
| AI Expert Assist Articles Count | The number of articles surfaced by the AI Expert Assist agent | Number | Counts unique articles shown | SUM |
| Agent Interaction Count | The number of actions in which the human agent interacted with a knowledge base suggestion | Number | Counts interactions with KB suggestions, such as View full article, copy, insert, or pin | SUM |
Dimensions
| Dimension | Definition | Filter | Group by |
|---|---|---|---|
| AI Expert Assist Knowledge Base | The name of the knowledge base the article came from | Yes, multi-select | Yes |
| AI Expert Assist Article | The name of the article surfaced. Only articles actually used in engagements are populated. | Yes, multi-select | Yes |
| AI Expert Assist Agent | The AI Expert Assist agent used in the engagement |
Yes
| Yes |
| Queue | The queue the engagement is associated with |
Yes
| Yes |
| Agent | The human agent who handled the engagement |
Yes
| Yes |
| Direction | Inbound (including callback) or Outbound |
Yes
| Yes |
| Engagement ID | The unique identifier of the engagement | Yes, multi-select | No |
| Channel | The channel that the engagement was in when it entered the queue | Yes | No |
| Source | The source that the engagement was in when it entered the queue | Yes | No |
Time dimensions
Year, Quarter, Month, Day, Hour, 30 minutes, and 15 minutes.
Agentic AI Expert Assist Suggestion dataset
One row represents one suggestion served by AI Expert Assist.
Use this dataset to judge suggestion quality. Interaction rate tells you whether suggestions are useful enough to act on; the feedback split tells you what agents think of them explicitly. Read the two together, a suggestion type with high interaction and negative feedback is a different problem from one that is simply ignored.
Metrics
| Metric | Definition | Type | Calculation | Aggregation |
|---|---|---|---|---|
| AI Expert Assist Suggestions Count | The number of suggestions served by the AI Expert Assist agent | Number | Counts suggestions generated | SUM |
| AI Expert Assist Interacted Suggestions Count | The number of suggestions the human agent interacted with | Number | Counts suggestions with at least one of the seven qualifying interactions: copy, insert, thumbs up or down, view full article (KB only), switch language, pin, tool call CTA |
SUM
|
| AI Expert Assist Feedback Count | The number of suggestions that received any feedback | Number | Counts every instance of a human agent giving feedback on a suggestion |
SUM
|
| AI Expert Assist Positive Feedback Count | The number of suggestions that received positive feedback | Number | Counts thumbs-up feedback on suggestions |
SUM
|
| AI Expert Assist Negative Feedback Count | The number of suggestions that received negative feedback | Number | Counts thumbs-down feedback on suggestions |
SUM
|
Note: Feedback is voluntary, so feedback rate is usually a small fraction of interaction rate. Treat it as directional, not representative.
Dimensions
| Dimension | Definition | Filter | Group by |
|---|---|---|---|
| AI Expert Assist Agent | The AI Expert Assist agent used in the engagement |
Yes
| Yes |
| Queue | The queue the engagement is associated with |
Yes
| Yes |
| Agent | The human agent who handled the engagement |
Yes
| Yes |
| Direction | Inbound (including callback) or Outbound |
Yes
| Yes |
| Engagement ID | The unique identifier of the engagement | Yes, multi-select | No |
| Channel | The channel that the engagement was in when it entered the queue |
Yes
| No |
| Source | The source that the engagement was in when it entered the queue |
Yes
| No |
Time dimensions
Year, Quarter, Month, Day, Hour, 30 minutes, and 15 minutes.
