Overview

The PM Agent is Hissuno's automated product manager. It reviews every customer feedback session, identifies actionable product insights, and maintains a living backlog of customer-reported issues. When enough customers report the same problem, the PM Agent generates a product specification to help your team move from feedback to implementation.

How the PM Agent Analyzes Feedback

Analysis Pipeline

Every feedback session -- whether from the support widget, Slack, Intercom, or Gong -- is automatically queued for PM Agent review. The analysis follows this pipeline:

Feedback session
    |
    v
Classification --> Tag with category from standard tags
    |
    v
Issue extraction --> Identify distinct product concerns
    |
    v
Deduplication --> Match against existing issues
    |
    v
Action --> Create new issue OR upvote existing issue

Session Classification

The PM Agent first classifies each session by assigning one or more tags from the standard set:

general_feedback -- General product observations
wins -- Positive feedback and praise
losses -- Churn signals or negative outcomes
bug -- Bug reports and defects
feature_request -- Requests for new capabilities
change_request -- Requests to modify existing behavior

Classification results appear on the session detail page and can be used to filter feedback in the dashboard.

Feedback Extraction

Within each session, the PM Agent identifies individual feedback items. A single conversation may contain multiple distinct concerns. For example, a customer might report a bug and also request a new feature in the same session. The PM Agent separates these into discrete items for proper tracking.

Each extracted item includes:

A concise summary of the customer's concern
The relevant quote from the conversation
The inferred category and urgency
The customer and company context

Issue Creation and Upvoting

How Issues Are Created

When the PM Agent identifies a concern that does not match any existing issue, it creates a new one. New issues include:

Title -- A clear, descriptive summary of the problem or request
Description -- Synthesized from the customer's words with relevant context
Category -- Bug, feature request, or change request
Priority -- Based on urgency, customer impact, and frequency signals
Customer link -- The session and customer that first reported the issue
Initial upvote count -- Starts at 1

How Upvoting Works

When extracted feedback matches an existing issue, the PM Agent upvotes it instead of creating a duplicate. The matching process considers:

Semantic similarity -- Is the customer describing the same problem, even in different words?
Scope -- Does the feedback relate to the same product area or initiative?
Category alignment -- Is this the same type of concern (bug vs. feature request)?

When an issue is upvoted:

The upvote count increments
The new session is linked to the issue
The customer is added to the issue's affected customers list
The issue description may be refined to incorporate new context

Issue Priority Recalculation

As upvotes accumulate, the PM Agent periodically recalculates issue priority. Factors include:

Customer count -- More affected customers increases priority
Customer segment -- Issues affecting enterprise or high-value customers are weighted higher
Recency -- Recent feedback is weighted more than old feedback
Sentiment intensity -- Frustrated or angry feedback raises priority

Spec Generation

What Triggers Spec Generation

Spec generation is triggered manually from the issue detail page. When you want a specification for an issue, click the Generate Spec button on the issue. This gives your team control over which issues receive detailed specifications.

What the Spec Contains

Generated specifications include:

Problem statement -- What customers are experiencing, grounded in real feedback
Customer evidence -- Quotes and data from the sessions that reported this issue
Affected customers -- List of customers and companies impacted
Suggested solution -- A proposed approach based on the product context and codebase knowledge
Acceptance criteria -- Concrete conditions for the issue to be considered resolved
Related issues -- Links to similar or dependent issues

Spec Review

Specs are generated as drafts and appear in the issue detail page under the Specification tab. Your team should review and refine the spec before moving it to engineering. The PM Agent provides a starting point, not a final product decision.

Configuring the PM Agent

You can customize the PM Agent by navigating to the Agents page, then clicking Configure on the Product Specialist card. The configuration dialog provides the following settings:

Classification Guidelines

Provide guidelines that help the PM Agent classify feedback more accurately. If the agent consistently miscategorizes certain types of feedback, add instructions here. For example:

Any mention of "slow" or "loading" in the context of page rendering
should be categorized as a bug, not a feature_request.

Analysis Guidelines

Control how the PM Agent analyzes sessions and extracts feedback. These guidelines influence how the agent identifies distinct concerns, determines severity, and decides whether feedback warrants a new issue.

Spec Guidelines

When a spec is generated for an issue, these guidelines shape the output. You can specify your preferred format, level of detail, and what sections to include.

Deduplication

The PM Agent uses fixed similarity thresholds when matching feedback against existing issues. Feedback that is semantically similar to an existing issue is upvoted; otherwise, a new issue is created. Sessions with fewer than 3 messages are automatically excluded from analysis.

Manual Corrections

You can always manually override the PM Agent's decisions:

Re-categorize a session from the session detail page
Merge duplicate issues from the issues list
Change issue priority manually (this prevents future automatic recalculation)
Unlink a session from an issue if it was incorrectly matched

Manual corrections help the PM Agent learn your team's preferences over time.

Monitoring the PM Agent

Track PM Agent performance from the Dashboard page. Key metrics include:

Issues created -- New issues identified per week
Upvote accuracy -- How often upvotes match the correct existing issue
Override rate -- How often your team manually corrects the agent's decisions

PM Copilot