Help Agent Insights
The Help Agent Insights view inside the aprity managed app gives tenant admins a read-only dashboard of how the Help Agent is being used and where the underlying documentation could be strengthened.
The Help Agent answers questions using a deterministic knowledge-tools orchestrator — it queries structured facts about your org (objects, fields, access, dependencies, ownership, execution order, and the generated narrative) rather than searching free-text document fragments. Insights describe how that agent is used and where coverage is thin.
All telemetry surfaced here is aggregated and PII-scrubbed before persistence. The view never shows the verbatim question a user typed or the user's identity.
Where to find it
Open the aprity app in Salesforce and navigate to the Help Agent Insights surface (visible only to users with the Aprity_Admin permission set). When telemetry has not yet accumulated -- typically right after enabling the Help Agent -- empty-state messages explain what is missing.
What it shows
Adoption
Adoption metrics describe how widely the Help Agent is used:
- DAU -- daily active users (users who interacted with the agent in the last 24 hours).
- MAU -- monthly active users (users who interacted in the last 30 days).
- Agentforce (30d) -- number of conversations through the Agentforce channel in the last 30 days.
- Slack (30d) -- number of conversations through the Slack channel in the last 30 days (visible when the Slack adapter is enabled).
Coverage gaps -- most asked objects
A ranked list of the Salesforce objects that user questions cited most often over the last 30 days. This is a usage-driven coverage signal: if a frequently-asked object has thin documentation, it is a candidate for re-scan with deeper analysis settings or for targeted on-demand documentation.
Quality signals
Two heuristics flag documentation topics that may need attention:
- Flagged topics -- documentation areas with at least 5 user votes and a thumbs-up ratio below 30%. Users are voting these answers down consistently, indicating the underlying documentation is wrong, ambiguous, or missing context.
- Low-confidence topics -- documentation areas the agent draws on at least 10 times but with an average LLM confidence below 40%. The agent is using this documentation but is not confident it answers the question, often because the documentation is shallow or off-topic.
When no quality issues are detected, a positive empty-state message indicates the documentation is performing well.
How telemetry is collected
- Each Help Agent answer carries a thumbs up / thumbs down control. Each vote is recorded against the documentation topic the answer drew on (not the user, not the question).
- Usage counts and LLM confidence scores are recorded per documentation topic, in aggregate.
- The verbatim question, the user identity, and the answer text are not persisted to the telemetry store.
See Audit Logging for the retention policy that applies to telemetry data, and Data Residency for where it is stored.
Acting on the signals
| Signal | Recommended action |
|---|---|
| High DAU/MAU on a single object | Schedule deeper analysis on that object (Technical Analysis, Integration Inventory). |
| Coverage gap on an object the docs do not cover | Add the object to the next scan scope; consider running On-Demand Docs for a targeted document. |
| Flagged topic with low up-ratio | Review the source documentation page. Likely candidates: stale rule descriptions, ambiguous step explanations, or LLM-generated narratives that need a re-scan. |
| Low-confidence topic used often | The agent draws on this documentation frequently but it does not actually answer questions well. Re-scan with deeper analysis settings on the underlying object. |
Toggling telemetry
Telemetry is enabled by default when the Help Agent is enabled. To disable telemetry for your tenant, contact support@aprity.ai. When telemetry is disabled, the Insights view shows empty-state messages and no aggregate data is collected.
Plan availability
Help Agent Insights are available wherever the Help Agent itself is enabled (Intelligence and Trial). See Plan Comparison for the full feature matrix.