Clusters

Clusters are groups of similar insights identified automatically by AI using vector embeddings. They help you spot recurring patterns — the same feature request from dozens of customers, a widespread bug report, or a common UX complaint.

How clustering works

ProductSights generates vector embeddings for each insight. When multiple insights have similar content (above a configurable similarity threshold), they're grouped into a cluster. Each cluster gets:

• An AI-generated title and summary (you can refine them)
• A count of how many insights it contains
• Links to all individual insights within the group

Viewing clusters

Navigate to Dashboard → Clusters to see all clusters. A summary strip at the top shows total cluster count, how many are trending up this week, and average priority.

Filtering and sorting

The toolbar provides several ways to narrow down clusters:

• Search — full-text search across cluster titles and summaries
• Theme filter — dropdown showing all themes with cluster count per theme
• Trend filter — show only rising, falling, or stable clusters
• Sort by — Most insights (default), Highest priority, Lowest sentiment, Trending, or Newest
• Clear all — reset all active filters at once

From Analytics → Top themes, you can open View clusters for a theme; that adds ?theme=<themeId> to the clusters URL and filters the list to clusters linked to that theme. Use Clear filter to show all clusters again.

Cluster cards

Each cluster is displayed as a card with visual signals:

• Insight count — prominent count in the top-left
• Trend accent — a left border color (green for rising, amber for falling)
• Priority highlight — cards with average priority ≥ 4.0 get an amber background tint
• Sparkline — a 7-day mini bar chart showing daily insight volume
• Status badges — "Discovery" (blue, in-progress), "Discovered" (green, complete), "Review" (amber, high variance), theme badge, priority, and sentiment

Click any cluster to open the cluster detail page — the product opportunity view.

Cluster detail: opportunity page

The cluster detail page is designed around three questions:

1. What are customers saying? — A Customer evidence section shows the top representative quotes in a quotes-only format (breadth across companies first). The header shows "Top quotes from N companies" and each quote displays an excerpt with attribution. Initially two quotes are visible, with an expandable "+N more" button.
2. How strong is the signal? — A signal overview sidebar shows report volume, week-over-week volume trend, companies affected, average sentiment/priority, and how many insights are synced to external tools.
3. What should we do next? — A Discovery workspace (structured fields) and Planning & execution (theme, owner, roadmap) support product discovery and delivery.

Tabs

• Reports — Each insight row shows a sentiment-colored left border (green = positive, red = negative, amber = mixed). Expand a row to see category, sentiment, priority, triage status, sync state, and the raw content.
• Discovery — Structured workspace with progress tracking (see below).
• Files — Upload research PDFs, screenshots, notes, and other artifacts tied to the cluster (not per-insight attachments).

Discovery workspace progress

The Discovery tab tracks completion across five structured fields:

1. Problem statement — What user problem does this cluster reveal?
2. Hypotheses — Why might this be happening? What opportunities exist?
3. Open questions — What still needs validation or research?
4. Research notes — Interview findings, patterns, internal synthesis.
5. Recommended next step — Dropdown: Not set, Investigate further, Prototype/validate, Ship a fix or change, Monitor/watch.

A progress ring in the top-right shows how many of the five fields are filled (e.g., 3/5). The ring turns green when all fields are complete. Fields auto-save on blur, and filled fields show a green checkmark.

Analytics (beta)

If your organization has connected an analytics MCP integration in Settings → Analytics, and event mappings exist for the product areas tied to the cluster, a beta panel may show 30-day usage context (trend-oriented signals) for mapped events. Customer evidence remains primary; analytics is supporting context.

High variance detection

Clusters where insights vary significantly from the cluster center are automatically flagged with a "Review" badge. This indicates the cluster may contain sub-topics worth splitting.

Workflow and execution tracking

On each cluster detail page, you can track execution state:

• Proposed — candidate problem to discuss
• Investigating — active product discovery
• Accepted — committed for planning/work
• In progress — implementation started
• Shipped — solution released

You can also assign a theme, execution owner, and roadmap link for traceability.

Impact tracking

Once a cluster reaches accepted or shipped status, ProductSights shows a before/after impact view (14 days before vs 14 days after) with:

• Volume delta
• Sentiment delta
• Before/after counts

Use this to verify whether actions reduced pain and improved customer outcomes.

Themes

Themes are user-defined categories you can assign to clusters for higher-level organization. For example, you might create themes like "Onboarding", "Performance", or "Mobile Experience".

Managing themes

Go to Settings → Themes to:

• Create a theme — give it a name, optional description, and color
• Edit a theme — update its name, description, or color
• Delete a theme — only possible if no clusters are assigned to it

Assigning themes

From the cluster detail page (Planning & execution), assign a theme to organize your feedback taxonomy. This makes it easier to track broad product areas over time.

Why clusters matter

Without clusters, you'd need to manually read hundreds of feedback items to spot patterns. Clusters do this automatically:

• Quantify demand — see exactly how many customers are asking for the same thing
• Reduce noise — instead of 50 individual feature requests, you see one cluster with a count
• Track trends — watch cluster sizes grow or shrink over time
• Prioritize effectively — larger clusters represent broader customer needs