web
You’re offline. This is a read only version of the page.
close
Skip to main content
Community site session details

Community site session details

Session Id :
Dynamics 365 Community / Blogs / Community Help & Announcements / Product Insights Agent – RA...

Product Insights Agent – RAI FAQs

Views (1)
Bret Fraser Profile Picture Bret Fraser Community Manager
This FAQ provides answers to frequently asked questions about the AI technology that's used in the Product Insights Agent. It includes key considerations and details about how the AI is used, how it was tested and evaluated, and any specific limitations.

What is Product Insights Agent?
The Product Insights Agent is a Microsoft internal-facing AI system that helps Microsoft Dynamics 365 and Power Platform engineering teams analyze large volumes of community discussions. It transforms raw posts into themes, sentiment trends, verbatim excerpts, and exploratory opportunities, enabling product teams to more easily understand customer pain points and feedback.

What are the capabilities of the agent?
  • Data harvesting & filtering: Collects threads and replies from BIC Community forums while excluding spam, irrelevant, or harmful content.
  • Clustering & theming: Groups related discussions into themes such as feature requests, documentation gaps, or adoption blockers.
  • Sentiment analysis: Assigns directional sentiment (positive, negative, neutral) to each theme.
  • Traceability: Provides verbatim excerpts from community posts for grounding and transparency.
  • Exploratory opportunities: Highlights potential areas for teams to investigate (framed as opportunities, not prescriptive recommendations).
  • Delivery:
    • Insights Page with nightly 30-day view of top themes.
    • On-demand tailored reports filtered by product, forum, category, and timeframe, delivered via email.
 
How is the agent evaluated? What metrics are used to measure performance?
This feature underwent substantial testing before it was released. If you encounter inappropriate generated content, report it to the Community team by using this feedback form: Report abuse. Your feedback helps improve the functionality.

What are the limitations of the agent? How can users minimize the impact?
Known limitations:
  • Summaries may occasionally omit nuance or mis-cluster related insights.
  • Sentiment classification can be imperfect (e.g., sarcasm, mixed tone).
  • Works best for English-language posts; performance may degrade on non-English or code-mixed text.
  • Only reflects conversations in BIC Community forums, not comprehensive of all customer feedback sources.
  • “Opportunities” are AI-generated highlights, not authoritative guidance.
Mitigations for users:
  • Always validate AI outputs against the verbatim excerpts provided.
  • Use insights as exploratory input, not prescriptive direction.
  • Combine with other data sources (telemetry, surveys, support cases) for fuller context.
What is the data use policy for the agent?
  • Training: Community data is not used to train or fine-tune the models. The agent uses pre-trained Microsoft AOAI models (GPT-4.1-mini and GPT-O3).
  • Inference only: Community posts are passed to models at runtime to generate insights, then discarded.
  • Filtering: Spam, PII/PHI, and harmful content are excluded before processing.
  • Storage: Outputs (themes, summaries, verbatims, opportunities) are stored securely in Microsoft systems for internal engineering consumption only.
  • Access control: Insights are available only to authenticated Microsoft engineering teams; no external users or customers access this data.
 
 

Comments

*This post is locked for comments