Analysis Workflow

Standard process for analysing conversation data in Tellet.

Analysis Steps

1. Check Data Quality

Navigate to Overview Tab to verify:

• Completion rate (percentage of conversations completed)
• Total number of completed conversations
• Response count per question
• Average completion time

2. Generate Insights

If insights are not displayed:

1. Click "Run" in the Overview Tab
2. A progress indicator tracks each question in real time, so you can follow along without refreshing
3. Insights populate automatically when the run completes

Insights include:

• AI-generated summary
• Themes and key patterns
• Cross-question patterns that recur across multiple questions
• Strategic recommendations
• Notable visuals (images respondents shared, curated by AI)
• In Their Own Words quotes with links back to the source transcript
• Sample demographic profile of the cohort
• Per-question summaries

To re-analyse a project that is already fully analysed (for example, after adding or removing conversations), use the Re-run option. This re-analyses all existing conversations from scratch rather than only picking up new ones.

Screening context in analysis — when a project has screening questions with qualify/disqualify rules, Tellet factors those criteria into the analysis. The AI avoids restating obvious demographic facts that follow directly from the screening, and frames findings as observations about a pre-filtered population rather than the general public.

3. Review Question Responses

Navigate to Questions & Insights Tab. Each question is presented as its own page:

1. Open the question you want to review
2. Review AI-generated summary for that question
3. Review AI-generated categories
4. Refine categories if needed
5. Review response distribution across categories
6. Read representative answers for each category

Categories:

• AI automatically generates categories based on response patterns
• Categories can be manually edited
• Responses can be assigned to single or multiple categories
• Distribution shows the percentage of responses per category as a bar chart; when categories are single-select, you can switch the chart between bar and pie

4. Read Transcripts

Navigate to Transcripts Tab:

1. View list of conversations (Completed or Incomplete)
2. Apply filters if needed (demographics, screening responses)
3. Click conversation to view full transcript
4. Review message-by-message conversation flow

Filtering options:

• Completion status (Completed, Incomplete, Screened Out)
• Demographic data (age, gender, location, etc.)
• Screening question responses
• Date range

5. Ask the Research Assistant

Navigate to Research Assistant Tab:

1. Type a question in natural language or click a suggested question
2. Read the cited answer — respondent references like [R7] link back to the source transcripts
3. Review any artifacts (quote collections, data tables, theme summaries, charts, markdown reports) that appear in the side panel
4. Ask follow-up questions in the same thread — the assistant remembers your conversation

Questions can be:

• Open — "What are the main themes from the pricing question?"
• Specific — "What did respondent 7 say about onboarding?"
• Comparative — "How do Premium customers differ from Free customers?"
• Quantitative — "How many women aged 25-35 mentioned price as a concern?"
• Reportive — "Generate an executive summary of the top three findings"

The Research Assistant works in 64+ languages and answers in the same language as the question.

6. Mark Conversations Complete

Incomplete conversations that you know are substantively finished (for example, a participant finished through an alternative channel) can be promoted to the Completed view without going through the normal interview flow. Select them in the Transcripts tab and use the "Mark as Complete" action. They move to the Completed view immediately and become eligible for analysis.

7. Create Deliverables

Export Raw Data

1. Navigate to Transcripts Tab
2. Click Export button
3. Select the export type (Answers in columns, Full conversation, or Summary & categories)
4. Select format (Excel, CSV, or TXT — Excel is the default)
5. Click Export
6. File downloads to browser

Export types:

• Summary & categories (Excel/CSV/TXT): conversation-level data — IDs, timestamps, duration, summary, categories per question, and any metadata fields
• Full conversation (Excel/CSV/TXT): full message-by-message transcripts plus all metadata fields, the question each message relates to, the order it was shown, and whether each answer was spoken or typed
• Answers in columns (Excel/CSV/TXT): wide-format spreadsheet with one column per question, the presentation order of each question, and which stimulus a respondent saw — so each participant's answers and randomisation line up across rows
• Analysis Report (TXT only): AI-generated insights, summaries, themes, opportunities, and per-question analysis — downloaded from the Overview or Questions tab

Excel files also include a Data Dictionary tab that explains every column.

Analysis at Any Scale

Tellet automatically scales the analysis to the size of your study. Smaller studies are analysed in a single pass; larger studies are processed in stages and then brought together into one consolidated set of findings. This keeps results reliable and consistent whether you have a handful of conversations or many thousands — with nothing to configure.

Filtering Data

Audience filters become available once your project is marked Completed (from the project status menu) — while it is still collecting responses, the analysis keeps changing, so filtering is held back until you close data collection.

Once unlocked, filters apply to:

• Overview Tab insights
• Questions & Insights analysis
• Transcripts list
• Exports

To filter:

1. Click filter icon/button
2. Select filter criteria
3. Apply filter
4. View filtered results

Comparing Segments

Use the Compare tab for a structured side-by-side report between two audience filters. It produces a presentation-quality report covering executive narrative, statistical deltas, demographic profiles, shared themes, segment-unique findings, and a per-question breakdown.

1. Navigate to the Compare tab
2. Click + New Comparison
3. Pick two filters (Segment A and Segment B)
4. Click Compare and wait for the report to generate
5. Read the executive narrative first, then drill into per-question details
6. Copy the full report or download it as Markdown for client presentations

Comparisons are saved in the Compare hub to return to later or share by URL. Re-running analysis for a filter marks any comparison referencing it as stale; stale comparisons still open with their previous data, but a badge prompts you to regenerate.

Read full documentation →