Research

Research is where Floe turns raw content and screens into structured product knowledge. After ingestion collects your documentation and exploration maps your product's interface, the research pipeline analyzes both to build the understanding that powers capabilities, plans, and the AI agent's real-time answers.

What research produces

Research isn't a single pass. It runs multiple analysis layers across your content and screens:

  • Feature extraction: Identifies distinct product features from your docs and UI, grouping them by area (e.g., "Reporting," "User Management," "Integrations")
  • Workflow mapping: Connects sequences of screens into logical user flows, such as "create account then configure workspace then invite team"
  • Terminology indexing: Builds a vocabulary of your product's specific language so the agent speaks in your terms, not generic ones
  • Content-screen linking: Maps documentation topics to the screens where those features actually live, so the agent can show what it's talking about

This produces a structured knowledge base that the AI agent draws from in real time during demos, self-serve onboarding, and support conversations.

How research fits the pipeline

The typical flow is:

  1. Create a site for your product
  2. Ingest content from your docs, knowledge base, and videos
  3. Explore your product's interface
  4. Research analyzes the combination of content and screens
  5. Capabilities are generated from research output

Research runs automatically as part of capability generation. You don't need to trigger it separately. When you click Generate on the Capabilities tab, research runs first, then capabilities are produced from the results.

What you see in the dashboard

After research completes, the Knowledge tab in your site shows what Floe learned:

  • Topics: Grouped areas of your product, each with a summary of what Floe understands about that area
  • Features: Individual features identified from your content, with the source material that supports each one
  • Screen associations: Which product screens relate to which topics, so you can verify Floe's understanding
  • Content coverage: A breakdown of how much of your documentation maps to actual product screens, highlighting gaps where Floe has content but no corresponding UI or vice versa

This is the place to check whether Floe has an accurate picture of your product before running demos.

Reviewing and improving research output

Research quality depends on the quality of your inputs. If you see gaps:

Missing features: You likely need more content. Add documentation or help articles that cover the missing area, re-run ingestion, and regenerate capabilities.

Wrong associations: A feature might be linked to the wrong screen. This usually means exploration didn't reach the right page. Re-run exploration with deeper settings or make sure the starting URL can reach the relevant sections.

Vague descriptions: Research mirrors the clarity of your source content. If a feature description is generic, check whether the underlying documentation is specific enough. Detailed docs produce detailed research.

Low coverage scores: If the content coverage section shows many topics with no screen associations, exploration may not have reached those parts of your product. Check whether login or navigation barriers prevented full access.

How the agent uses research at runtime

During a live demo or onboarding session, the agent draws on research in three ways:

Answering questions: When a prospect asks "How does your reporting work?", the agent retrieves the research-derived knowledge about your reporting features and explains it using your product's terminology.

Navigating to the right place: Research links features to screens. When the agent needs to show something, it knows which screen to navigate to because research connected the documentation topic to the specific UI location.

Contextual explanations: As the agent navigates your product, it pulls relevant knowledge for whatever is currently on screen. If the agent lands on a settings page, it can explain each section because research mapped your settings documentation to that screen.

This is what makes the agent sound like it knows your product rather than reading from a script. The research layer is the bridge between your raw content and the agent's contextual help intelligence.

When to re-run research

Research re-runs automatically when you regenerate capabilities. You should regenerate when:

  • You've added or updated content via ingestion
  • You've re-run exploration after UI changes
  • The agent is giving inaccurate or outdated answers during demos

Each regeneration builds on the previous knowledge rather than starting from scratch.

FAQ

Is research the same as ingestion? No. Ingestion reads and indexes your raw content. Research analyzes that content alongside explored screens to build structured product understanding. Ingestion is input; research is analysis.

Can I see exactly what research extracted? Yes. The Knowledge tab shows topics, features, screen associations, and coverage metrics. Everything research produces is visible and reviewable.

Does research work without exploration data? It produces limited results. Without exploration, research can only analyze your content and won't be able to map features to screens. The agent will know what your product does but not how to navigate to specific features. See how it works for how the full pipeline connects.

How long does research take? It runs as part of capability generation. For most products, the full process (research plus capability generation) completes in under a minute.