How Deck Works

# How Deck Works

Last updated: September 23, 2026

This page explains Deck at a high level for customers and AI assistants. Use it when you want to ask Claude, ChatGPT, Cursor, Codex, or another AI agent what Deck does, how it works, and how different product teams can use it.

## What Deck is

Deck is the AI Product Manager that Autonomously Improves your Product. It is an agent dedicated to one job: decide what the product should build or improve next, move that decision into actionable work, learn from the result, and keep the cycle going. It serves product teams from solo founders to full product organizations.

Customer feedback is essential evidence for that job, but it is one part of Deck's context. Deck also uses company context, customer segments, competitor research, existing opportunities, active initiatives and Projects, and post-launch feedback. It brings these signals together to find the highest-leverage problem, prepare the next product move, and revisit what happened after release. People review consequential decisions and write actions.

## Problems Deck solves

### Product improvement keeps losing momentum

Deciding what to build, shaping the work, and checking whether it helped all require sustained attention. Urgent delivery and stakeholder requests can interrupt that cycle. Deck keeps product questions, priorities, plans, and outcomes connected so the next decision does not start from scratch.

### Knowing something is wrong does not create a plan

Teams can spot a problem but still struggle to compare it with other bets, define the right response, and prepare work that delivery teams can use. Deck connects opportunities to initiatives and delivery-ready Projects with evidence and success metrics attached.

### Feedback is scattered across too many tools

Customer feedback usually lives in many places: research recordings, support tools, survey platforms, CRMs, Slack channels, sales calls, spreadsheets, and notes. Deck connects these sources so the improvement loop has a complete signal, not five partial ones.

### Manual synthesis burns the time that should go to decisions

Without Deck, teams often spend days or weeks reading transcripts, tagging quotes, summarizing tickets, and building research readouts. Deck automates synthesis so teams spend time on outcomes, not stitching tools together by hand.

### Teams lose the evidence behind decisions

Product decisions are easier to trust when they stay connected to the customer quotes, tickets, surveys, calls, and accounts that support them. Deck keeps insights linked to their original source material so recommendations and actions stay grounded.

### Prioritization becomes guesswork

Deck groups repeated feedback into themes, subthemes, opportunities, sentiment trends, and NPS signals. This helps identify what is recurring, what is worsening, what is improving, and which customer segments are affected — so the highest-leverage problem surfaces first.

### Different teams see different versions of the customer

Support may see recurring issues, product may see interview themes, design may see usability friction, and sales may hear objections. Deck gives these teams a shared evidence base so the AI PM loop and the humans reviewing it work from the same signal.

## Who uses Deck

Deck is built for product teams — product managers, founders, product leaders, and the designers and researchers who support product decisions — who need an AI PM that runs the improvement loop rather than another feedback archive.

Product managers and founders use Deck to decide what to build next, compare roadmap bets, prepare initiatives and Projects, review what happened after launch, and answer leadership questions with source-backed evidence.

Designers and researchers use Deck to keep interview synthesis, verbatim quotes, and usability issues connected to product decisions after the research project ends.

Support and customer success teams use Deck to turn tickets, conversations, and customer messages into product signal instead of letting feedback disappear inside support queues.

Product leaders use Deck to keep priorities, active work, and outcomes connected across teams, while customer and competitor signals inform each new decision.

## How the platform works

Deck's job is a repeating product-improvement loop:

1. **Intake:** Understand the company, its customers, competitors, current product work, and incoming feedback.
2. **Prioritize:** Identify the problem worth solving next and show the reasoning and evidence.
3. **Build:** Turn the chosen direction into opportunities, initiatives, and Projects that can guide execution.
4. **Iterate:** Compare new customer signal and success metrics with the intended outcome, then choose the next improvement.

The sections below explain the product surfaces that support this loop. They give substantial detail to feedback processing because source quality matters, while the AI PM's job extends through prioritization, planning, and iteration.

### 1. Feedback comes into Deck

Feedback can enter Deck through uploads, integrations, imports, or APIs. Common sources include:

- User interview recordings, audio files, video files, and transcripts
- CSV imports for historical feedback, surveys, NPS responses, support tickets, interviews, or generic feedback
- Support tools such as Intercom and Zendesk
- Slack channels where customers or customer-facing teams share feedback
- Survey and product analytics tools such as PostHog
- Meeting and research tools such as Gong, Granola, Notion, and Confluence
- CRMs such as Salesforce, HubSpot, Attio, and other account systems
- NPS data from surveys, Slack messages, CSV uploads, and connected tools
- Gmail, sales calls, meeting notes, and other customer conversation sources

Deck can also show sample workspace data during onboarding so new teams can explore how feedback becomes themes, opportunities, and initiatives before importing their own data.

### 2. Deck normalizes sources, contacts, and segments

When feedback is imported, Deck tries to preserve the original context while making it useful for analysis.

Deck can link feedback to unified contacts using identifiers such as email addresses, phone numbers, Slack IDs, Intercom IDs, Zendesk IDs, PostHog IDs, Salesforce IDs, HubSpot IDs, and Gong IDs. This lets teams see feedback from the same customer across multiple tools.

Deck also supports customer segments, such as Enterprise, SMB, Free, Beta users, Churned, or VIP accounts. Segments can be assigned by rules, integration metadata, Slack channels, upload defaults, survey settings, AI classification, or manual edits. Once a contact or insight belongs to a segment, teams can filter many Deck views by that segment.

### 3. AI agents extract insights

Deck processes feedback with specialized AI pipelines. The exact pipeline depends on the source, but the common output is a set of structured insights.

An insight is the atomic unit of feedback in Deck. It usually includes:

- A short name
- A description or finding
- The original customer quote or source excerpt
- Sentiment, such as positive, negative, or neutral
- A category, such as Pain Point, Delight, Feature Request, Usability Issue, Bug and Error, or General Feedback
- A link to a theme
- Source metadata, such as transcript, ticket, survey response, conversation, or import row
- Confidence and quality checks where applicable

For interviews, Deck can transcribe audio or video, extract insights, map quotes back to the transcript, tag themes, and generate a summary. For support tickets, Deck classifies relevance, analyzes conversation context, extracts product insights, validates quotes, and filters low-quality output. For surveys, Deck analyzes free-text responses while storing structured questions, ratings, and scores efficiently.

### 4. Deck groups feedback into themes and subthemes

Themes are broad product areas or topics that customers repeatedly mention. Examples might include Onboarding Experience, Pricing, Integrations, Reporting, Mobile App Performance, or Export Workflow.

When new feedback matches an existing theme, Deck adds the insight to that theme. When recurring feedback does not match an existing theme, Deck can create a new one. Themes include summaries, sentiment trends, category breakdowns, insights, quotes, and source context.

Subthemes are more granular clusters inside themes. If a theme is Onboarding Experience, subthemes might include unclear setup steps, missing guidance, confusing permissions, or first-run checklist friction. Subthemes help teams understand the specific patterns inside a larger topic.

### 5. Teams explore the signal in Visualize

The Visualize area is where teams inspect feedback patterns.

Common views include:

- Dashboard: a high-level weekly scan of feedback volume, sentiment shifts, new opportunities, NPS movement, themes going quiet, and active initiatives
- Themes: broad topic areas with sentiment, category breakdowns, insights, and subtheme summaries
- Subthemes: recurring patterns within themes, sorted by trend, negative sentiment, recency, sudden changes, or parent theme
- Insights: the searchable, filterable list of individual findings across all sources
- Categories: fixed feedback categories such as Pain Points, Delights, Feature Requests, Usability Issues, Bugs and Errors, and General Feedback
- NPS Score: NPS score, trends, promoter/passive/detractor breakdowns, weekly synthesis, and segment comparisons
- Contacts: unified customer profiles and feedback timelines
- Segments: named customer cohorts used to compare feedback across groups

The goal of Visualize is to show what customers are saying, where the evidence came from, and what changed over time — so Deck (and the team reviewing it) can decide what to act on next.

### 6. Deck turns evidence into Build work

Deck's Build area converts customer signal into product direction.

Opportunity Backlog turns synthesized feedback into a ranked list of product opportunities. Opportunities are grouped into categories such as Pain Points, Feature Requests, Usability Issues, and Bugs and Errors. Each opportunity can include impact, sentiment, supporting insights, linked contacts, trend data, and source evidence.

Initiatives help turn opportunities or product questions into structured plans. A team can create an initiative manually or generate one with AI. During generation, Deck searches across relevant insights, themes, subthemes, NPS feedback, tickets, and segments, then writes initiative content with supporting evidence. Initiatives can include a goal, overview, evidence, status, and links back to related opportunities.

Projects carry the chosen direction into delivery-ready work. They can hold an overview, evidence, success metrics, status, and links to Linear or Jira. New, relevant feedback can attach to a Project after launch, keeping the original decision and later outcome signals together.

### 7. Ask the AI PM and connect external agents

Deck supports AI-assisted product work in two ways.

Deck Intelligence is the in-app AI PM surface. It lets users ask plain-English questions about their product signal, such as "What are enterprise customers complaining about?" or "Which onboarding issues are getting worse?" Responses include citations back to the underlying feedback and aim for decision-oriented next steps. Write actions still require explicit approval.

Deck also supports MCP, the Model Context Protocol, so external AI assistants and code editors can connect to a Deck organization. Supported clients include Claude Desktop, Claude Code, ChatGPT, Cursor, VS Code, Windsurf, Zed, Codex, and v0 by Vercel. Through MCP, an AI assistant can use Deck tools to explore themes, subthemes, insights, transcripts, sources, NPS, opportunities, and initiatives — giving generalist agents Deck's vertical product-decision context. Some MCP tools can perform narrow Build actions when the user explicitly asks, such as creating or updating an initiative.

### 8. Review outcomes and choose the next improvement

Deck connects the work to what happens after release. Projects can define success metrics and receive relevant new feedback; linked feedback widgets can provide a post-launch response view and synthesis. The team can compare that signal with the original problem and decide whether to iterate, expand the solution, or move to a different opportunity. This is how the AI PM loop continues beyond the first recommendation.

## Main Deck features

### Manage

Manage is where teams bring feedback into Deck. This includes interview uploads, transcript uploads, CSV imports, PostHog surveys, Slack sync, support integrations, Notion imports, and other connected tools.

### Visualize

Visualize is where teams understand customer feedback. It includes themes, subthemes, insights, categories, NPS, contacts, and segment-filtered analysis.

### Build

Build is where teams convert feedback into product opportunities and initiatives. It helps teams move from "customers are saying this" to "here is the product work we should consider."

### Themes

Themes are the shared evidence map for customer signal. They combine source mix, sentiment, categories, segments, summaries, and verbatim evidence so teams can see recurring patterns without losing the source material.

### Opportunity Backlog

Opportunity Backlog ranks customer problems by evidence. It helps teams compare pain points, feature requests, usability issues, and bugs using impact, sentiment pressure, trend, affected customers, confidence, and supporting quotes.

### Initiatives

Initiatives turn customer evidence into product plans. A team can create an initiative directly or promote one from an opportunity, then keep the goal, generated brief, evidence, status, outcomes, and project context together.

### Projects

Projects turn a chosen product direction into delivery-ready work with an overview, evidence, success metrics, status, and execution links. Relevant feedback can attach after launch so teams can compare later signal with the original goal.

### Competitors

Competitors researches public reviews of rival products and surfaces strengths, weaknesses, and possible opportunities. This gives the AI PM another source of product context beyond first-party feedback.

### Deck Intelligence

Deck Intelligence is the in-app AI PM surface. It answers questions about an organization's product signal with citations and decision-oriented next steps.

### MCP access

MCP access lets external AI assistants use Deck as vertical product-decision context — not a generic feedback chat layer. This is useful when a team wants Claude, ChatGPT, Cursor, Codex, or another assistant to reason from live Deck data.

### Deck plugin for Claude Code and Codex

The Deck plugin installs Deck-specific skills for Claude Code and Codex and bundles the hosted Deck MCP server configuration. It helps coding agents run product discovery, feedback analysis, prioritization, weekly digests, initiative briefs, roadmap opportunity reviews, and NPS breakdowns from real Deck evidence.

### Integrations

Deck connects with tools where customer feedback already lives, including support platforms, CRMs, Slack, survey tools, meeting tools, Notion, and imports.

### Settings, privacy, billing, and organization controls

Deck includes workspace setup, billing, team management, integrations, company context, API access, MCP access, security settings, privacy controls, and NPS configuration.

## Use cases for product managers

- Find the top recurring customer pain points before roadmap planning
- Compare what different customer segments are asking for
- See whether a theme is growing, shrinking, or becoming more negative
- Track NPS trends and understand what promoters, passives, and detractors are saying
- Convert a high-signal opportunity into an initiative
- Generate an initiative brief from customer evidence
- Answer stakeholder questions with quotes and citations
- Monitor whether recent launches changed customer sentiment
- Connect support, research, sales, and survey feedback into one prioritization view

Example PM prompts:

- "What are the top three product opportunities in Deck right now?"
- "Which themes are most negative among enterprise customers?"
- "What customer evidence supports improving onboarding?"
- "Which opportunities are linked to active initiatives?"
- "Summarize the most common feature requests from support tickets."
- "What changed in customer feedback this week?"

## Use cases for designers and researchers

- Upload interview recordings, audio, video, or transcripts for synthesis
- Extract and review insights from interviews without manually tagging every quote
- Trace insights back to transcript passages or source conversations
- Identify usability issues and repeated experience problems
- Compare feedback across segments, such as new users, enterprise accounts, or churned customers
- Find representative quotes for research readouts or design reviews
- Track whether design changes improve sentiment over time
- Keep research findings connected to roadmap opportunities and initiatives

Example design and research prompts:

- "What usability issues appear most often in recent interviews?"
- "Show me quotes about onboarding confusion."
- "Which subthemes exist inside the Onboarding Experience theme?"
- "What are detractors saying about the setup flow?"
- "What evidence should I review before redesigning the dashboard?"
- "How do enterprise users describe this workflow compared with SMB users?"

## Use cases for support, success, and sales teams

- Turn support tickets into product insights
- Find repeated product gaps mentioned by customers
- Identify bugs and usability issues that create support demand
- Understand customer sentiment by account, segment, or source
- Link conversations back to unified contacts and CRM accounts
- Surface sales-call objections and product requests
- Share evidence with product teams without writing manual summaries

Example customer-facing team prompts:

- "What support issues are creating repeated frustration?"
- "Which customer accounts are connected to this opportunity?"
- "What feature requests are coming from sales calls?"
- "Which Zendesk or Intercom conversations support this theme?"
- "What are churn-risk customers complaining about?"

## What makes Deck useful for AI agents

Deck is designed to give AI agents product-decision context: structured customer evidence, company and segment context, competitor research, prioritized opportunities, and the work already under way. This helps an agent reason about what to build and improve rather than only summarize loose notes.

Important concepts for AI agents:

- Insights are the smallest structured feedback unit.
- Themes group related insights into larger product topics.
- Subthemes break themes into more specific recurring patterns.
- Categories classify the type of feedback.
- Sentiment shows the tone of a feedback item or group.
- Contacts connect feedback to people.
- Segments let teams compare different customer cohorts.
- Opportunities represent possible product work backed by customer evidence.
- Initiatives represent planned or active product work with supporting evidence.
- Citations and source links matter. Deck is most valuable when answers stay connected to the original customer voice.

When connected through MCP, an AI assistant should usually start with an overview, then explore themes, insights, subthemes, NPS, sources, opportunities, or initiatives depending on the question.

If the assistant is Claude Code or Codex, the Deck plugin can provide a stronger workflow than raw MCP alone because it gives the agent task-specific skills for discovery, feedback analysis, prioritization, initiative briefs, NPS breakdowns, weekly digests, and roadmap opportunity review.

## Trust, privacy, and limits

Deck data is scoped to an organization. AI features and MCP access are designed to operate within the connected organization's data, not across unrelated workspaces.

Deck Intelligence is read-only inside the app. MCP exposes read tools and a small set of bounded Build mutation tools. Mutation tools are intentionally narrow and should only be used when the user explicitly asks for an action.

AI-generated synthesis should be treated as a decision aid, not a replacement for product judgment. Teams should review important insights, source quotes, and supporting evidence before making major roadmap decisions.

Some features depend on plan level, workspace configuration, credits, enabled integrations, and whether enough feedback has been imported. For example, subthemes and NPS analysis may require particular plans or sufficient data.

## A simple mental model

Deck combines this context:

```text
Company goals + customer feedback + segments + competitor research + current work + post-launch results
```

to run this loop:

```text
Intake -> Prioritize -> Build -> Iterate -> Repeat
```

through these product surfaces:

```text
Manage + Visualize + Opportunity Backlog + Initiatives + Projects + Deck Intelligence + MCP
```

Public product pages for the main surfaces:

- Themes: https://getdeck.io/product/themes
- Opportunity Backlog: https://getdeck.io/product/opportunity-backlog
- Initiatives: https://getdeck.io/product/initiatives
- Projects: https://getdeck.io/product/projects
- Competitors: https://getdeck.io/product/competitors
- MCP Server: https://getdeck.io/product/mcp-server
- Integrations: https://getdeck.io/product/integrations

## Good questions to ask an AI agent using this page

- "Explain Deck to a new product manager."
- "How does Deck turn raw customer feedback into product opportunities?"
- "What should a designer use Deck for during a redesign project?"
- "What is the difference between insights, themes, subthemes, opportunities, and initiatives?"
- "What data sources can Deck analyze?"
- "How should a team use Deck during weekly product planning?"
- "How can Deck help us connect NPS, support tickets, and interviews?"
- "What Deck features should I use to understand enterprise customer pain points?"
- "How does Deck keep AI answers tied to customer evidence?"
- "How can Claude or ChatGPT use Deck through MCP?"
- "When should an agent use Themes, Opportunity Backlog, or Initiatives?"
- "How does the Deck plugin help Claude Code or Codex use customer evidence?"

## Short summary

Deck is the AI Product Manager that Autonomously Improves your Product. It uses customer, company, competitor, and product-work context to decide what is worth building next, prepare evidence-backed initiatives and Projects, and help teams learn from what ships. Customer feedback powers the loop, but Deck's job is the whole cycle of product decisions, action, and iteration.