Getting Started

Overview

Running a causal analysis in Cauzen takes three steps, corresponding to the three phases shown in the top navigation bar. You always move left to right: Data Wrangling → Causal Modeling → Causal Inference.

Your work is automatically saved in your browser as you go, so you can leave and come back without losing progress.

Step 1: Upload your data

Click Get Started on the home page, or click Data Wrangling in the top navigation bar.

On the Data Wrangling page, drag and drop a CSV file onto the upload area, or click it to open a file picker. Cauzen accepts CSV files up to 50 MB.

Once uploaded, you'll see a preview of your data and a table of columns. Spend a few minutes reviewing the columns — give them readable names and descriptions, and uncheck any columns you don't want to include in the analysis. This metadata helps the AI features throughout the rest of the workflow.

Cauzen also marks columns that are likely to cause modeling problems, such as row IDs, all-unique identifiers, sequential dates, empty or constant columns, duplicates, and linearly dependent columns. These recommendations appear in the column status area. You can keep a recommended column if you have a reason to, but the recommendation remains visible so you know why Cauzen flagged it.

When you're ready, click Continue to Causal Modeling at the bottom of the page.

Step 2: Build a causal model

On the Causal Modeling page you'll see your columns displayed as nodes on a canvas. Your task is to draw directed edges between nodes to represent cause-and-effect relationships.

The fastest way to get started is to click Discover. Cauzen checks whether the kept columns are suitable for discovery, sends the current data to the backend, streams discovery progress, and then applies LLM refinement to improve the graph. You can keep, remove, or modify any returned edges.

If discovery would fail because the encoded data matrix is singular, Cauzen lists recommended exclusions. Choose Apply exclusions and run to exclude those columns and continue discovery, or Review columns to go back and decide manually.

You can also draw edges yourself by holding Shift and dragging from one node to another.

When the graph reflects your understanding of the causal structure, click Continue to Causal Inference.

Step 3: Ask causal questions

On the Causal Inference page, you can ask questions in natural language using the question input at the top, or build a query directly using the Query Builder.

To ask a natural language question, type it in the input field (for example, "What would be the probability of Y if we set X to 1?") and click Interpret. Cauzen sends the current dataset, graph, and metadata to the backend so the generated query can use values that make sense for your data. It then fills in the Query Builder automatically.

Click Estimate to run the analysis. Cauzen will return an estimate of the causal effect, a confidence interval, and an AI-written explanation of what the result means.

Navigation tips

• The workflow stepper in the top navigation bar always shows your current phase. Click any phase to jump to it.
• The Cauzen logo in the top-left returns you to the home page.
• The Docs link opens this documentation in a new tab.
• The Feedback button sends a message with page and workflow context so issues are easier to diagnose.
• Your data, model, and questions persist in the browser between sessions.
• Replacing your dataset in Data Wrangling will reset the causal model and inference history.