Understanding analysis

Dashboard tab

High-level overview for the current filter slice: session counts, note volume, and distribution charts (topics, sentiment, intent). Start here to see whether patterns emerge before drilling into individual sessions.

Dashboard KPIs vs. analysis metrics

The dashboard always shows funnel KPIs for the study — public opens, sessions started, completed, and completion rate. These describe participation volume, not what was said in interviews.

Analysis metrics (configured at publish) are separate from funnel KPIs. Overall metrics answer “how many interviews showed X?” Per-session metrics answer “did this interview show X?” for each conversation.

Analysis metrics — overview

When you publish, KogniFeed can derive metrics from your objectives. You choose two optional layers in Analysis settings (gear icon on the Analysis page) or in the builder before launch:

• Overall metrics — aggregated percentages across completed sessions (dashboard cards).
• Per-session metrics — indicators evaluated separately for each completed interview, shown on that session only (Chats list and session detail).

Both are evaluated from participant messages after each session ends. They do not replace structured form fields or assistant notes — they complement them for roll-ups, session review, and structured capture.

Overall metrics

Overall metrics roll up across all completed sessions in your current filter (date range, CRM group, etc.). Each metric appears as a card on the Analysis dashboard with session count and percentage — for example “62% of interviews mentioned pricing friction.”

• Use when you need population-level trends: share of interviews that hit a theme, sentiment mix, or category split.
• Good for product feedback (“how often is onboarding confusing?”), pulse trackers, and executive summaries.
• Not for reading one specific conversation — per-session metrics stay on that interview’s row instead of aggregating.

Per-session metrics

Per-session metrics attach signals to each completed interview individually — they are not rolled up into dashboard percentages. You can define several indicators for the same study; each one is scored only for that conversation.

Results appear on the Chats tab (session list and detail panel) and in the analysis-ready email when enabled. They help reviewers scan many sessions and spot flags without re-reading every transcript.

• Use when you need conversation-specific flags, not population trends.
• Define multiple indicators per study — they are independent, not a single hire/no-hire verdict.
• Hiring example: “Salary expectation above band”, “Technical depth insufficient”, “Strong communication”, each as its own per-session metric.
• Support or QA example: “Issue unresolved”, “Escalation warranted”, “Customer effort high”.

Metric types

When editing metrics, each item has a type that controls how the AI fills it and how results display:

• Yes / no (binary) — whether a theme clearly appeared in the participant’s own messages.
• Sentiment (polarity) — when a theme appears, classify it as positive, neutral, or negative.
• Category — pick one option from a list you define (e.g. plan tier, root cause bucket).
• Yes / no + strength — binary signal plus intensity (low / medium / high).
• Short text (per-session only) — a concise operator-facing sentence when categories are too rigid.

Each metric also has a label and detection hint. The hint tells the model what counts as evidence — write it like instructions to a researcher (“Count only if the participant describes waiting more than 30 seconds”).

When to use which

• Tracking prevalence across many interviews → Overall metrics.
• Flags and indicators for one specific conversation → Per-session metrics (you can add several).
• Exact counts for reporting charts → Structured form fields (Form statistics tab).
• Rich qualitative themes and quotes → Notes tab (always generated when analysis runs).
• You can enable both overall and per-session metrics on the same study.

Configuring and editing metrics

• At publish — review AI-suggested metrics from your objectives; toggle overall and/or per-session generation.
• After publish — open Analysis → gear icon → Metrics tab. Add, edit, or remove definitions anytime.
• Changing a metric definition does not rewrite past sessions automatically. Use General → Reset analysis data if you need a clean slate, then new sessions will analyze with the updated definitions.
• Completed sessions keep chat transcripts after a reset; only computed analysis (notes, metrics, clusters) is cleared.

Reset analysis data

Analysis → gear icon → General tab → Reset analysis data removes stored results for this interview (notes, summaries, metric hits, theme clusters, post-interview form fills). Use this after changing metric definitions mid-study or when testing. Transcripts and live-captured form values from the conversation itself are kept.

Sentiment analysis

Automatic extraction of positive, neutral, and negative sentiment from participant conversation tone, with line-by-line sentiment tagging on transcripts and distribution charts on the dashboard.

Theme clusters

AI-powered semantic embedding clustering that automatically groups hundreds of open-ended answers into master canonical topics. See member counts, share of notes, and representative quotes without reading every transcript.

Notes tab

Structured assistant notes — one card per insight extracted from conversations. Each note includes theme, sentiment, importance, and supporting context. Filter by category, keyword, sentiment, or insight type. Use this for qualitative synthesis and quote hunting.

Form statistics tab

Aggregated charts for chartable structured fields from your blueprint — answer shares, scale distributions, choice breakdowns. Only fields with countable types appear here. If the tab is empty, you may not have structured fields yet, or the current filter slice has no captured values.

Chats tab

Full conversation transcripts session by session. Open a chat to read messages, assistant notes, captured fields, and per-session metric results together. Use this to verify context behind a note or pull exact participant wording.

When analysis is ready

Analysis queues when a session completes (or hits turn limits). You may receive an email when results are ready if your workspace has notifications configured. Refresh the Analysis page if you do not see new sessions immediately.

How to use results together

• Dashboard & metrics — funnel KPIs plus overall metric cards for trend spotting.
• Per-session metrics — open Chats to see session-specific indicators next to each transcript.
• Sentiment & theme clusters — tone distribution and grouped open-ended themes.
• Notes — prioritize themes, severity, and filter insights by category or keyword.
• Chats — validate nuance, read full transcripts, and collect verbatim quotes.