From Stories to Structure
If you've run a survey with an open-ended question, you're probably ready. Here's exactly what happens to your data.
The workflow
Voxil's pipeline has four stages. The first and last involve you directly — preparing your data and reviewing the findings. The middle two are where Voxil does its work.
Text responses are cleaned and prepared — removing duplicates and empty entries while preserving the original voice. A CSV file containing your text responses is then ready for analysis. If you have additional metadata — dates, demographics, geographic factors — these can be included as columns and used to filter and compare findings later.
Voxil uses machine learning to find intrinsic groupings in the data without predefined categories. Topics emerge from patterns of similarity and difference in the responses themselves — not from a coding scheme designed in advance. This means the findings help reflect what communities actually said.
Topics are organized into a nested tree — broad themes at the top, specific subtopics underneath. Each node is labeled and summarized using language models, so findings are legible to non-specialists. The hierarchy can be navigated from the general to the specific, and every node links back to the representative responses that informed it.
The interactive tree visualization allows domain experts and community participants to explore findings, validate interpretations, and identify what matters most. This is where analysis becomes shared understanding — findings that can be discussed, questioned, and acted on together.
What you need to get started
If you've run a survey with at least one open-ended question, you're almost certainly ready. The only strict requirement is a CSV file with a column of text responses.
A CSV file with one column — labeled 'Data' — containing the text responses to be analyzed. Additional columns for metadata (date, age, location, any other factors) are optional but can be used to filter and compare findings.
# Example structure:
Data, Date, Region, Age
I think the main issue is..., 2024-01-15, North, 24-30
We should consider..., 2024-01-15, South, 31-40
- One column 'Data' containing the text to analyze
- UTF-8 encoding recommended
- Additional columns for factors/metadata (optional)
Key capabilities
Voxil combines modeling, interpretation, and exploration in one workflow. These are the core features most teams use to move from raw responses to actionable insight.
- • Analysis to discover themes without predefined categories
- • Hierarchical topic structures that organize findings from broad themes to specific issues
- • Automatic topic labels and summaries for quick interpretation
- • Auto-tuning and corpus profiling to improve parameter selection for each dataset
- • Embedding-aware configuration suggestions that help tune analysis settings
- • Attributes, factors, comparisons, and filtering for subgroup analysis when metadata is available
- • Data provenance tracking from source CSV through generated outputs
- • Interactive tree exploration with in-browser editing and downloadable JSON
What you get back
Two outputs: a D3.js-compatible JSON hierarchy file containing the complete topic structure with labels, summaries, keywords, probability scores, and example documents — and an interactive tree visualization that lets you explore, search, filter, and export findings — for you to build upon.
- • Topic labels and summaries
- • Representative keywords
- • Metadata and factors (if applicable)
- • Typed attribute profiles per topic node (entity distributions, scores, IC measures)
- • Example documents
- • Parent topics and relationships
- • Topic weights
- • Data provenance tracking
- • View topics and their relationships
- • Compare topic weights and probabilities
- • Read topic summaries and keywords
- • Directly edit and export labels and summaries
- • Search across all topics
- • Filter by factors (if applicable)
- • Entity network view linking topics to organizations, participants, and other multi-valued metadata
- • Export as SVG or CSV
- • Link to representative documents
Analysis is a starting point, not an endpoint
The findings Voxil produces are designed to move — from the screen into the room, from the room into decisions. The four stages below describe how that happens.
Before findings can inform decisions, they need to be understood — not just by analysts, but by the groups who will act on them. Visual hierarchies, relative weights, and short summaries make it possible to compare priorities, tensions, and blind spots across audiences or datasets. The goal is a common reference point that helps groups move beyond anecdotes or dominant voices toward grounded discussion.
What do the findings mean here, for this decision, at this moment? This stage maps analytical themes onto existing goals, initiatives, or constraints — identifying where current strategies align with expressed community concerns and where they diverge. Abstract categories become language that resonates with the specific groups who need to act on them.
Findings anchor planning sessions, design workshops, and strategy meetings. Rather than starting from a blank canvas, groups and communities work from evidence of what communities are already expressing. This enables activities like identifying issues for intervention, prioritizing leverage points within a broader system, or stress-testing pre-conceptions against distributions of concerns and experiences.
As new data are collected — through follow-up surveys, ongoing engagement, or open feedback — results can be re-analyzed and compared over time. This makes it possible to track shifts in community concern, assess whether interventions are changing how issues are understood, and build systems of learning rather than commissioning one-off studies.
Ready to get started?
Try the visualization with sample data, or contact us to discuss your specific needs.