Find the drivers. Explore the what-ifs.

UCMT Labs helps you investigate causal relationships in tabular data, compare discovery methods, explore what-if scenarios, and explain the findings in one workspace.

Causal AI investigates how variables may influence one another, rather than only identifying patterns that occur together. UCMT Labs uses causal discovery methods to build a graph of candidate relationships from your data. You can then investigate paths, explore possible changes, and inspect the evidence reported by each method. Results depend on the data and the assumptions of the selected method.

A dashboard or chart helps you monitor values and spot patterns. UCMT Labs adds methods for investigating causal structure and exploring how an outcome could differ when a factor changes. You can compare discovery methods, trace paths between variables, run what-if analyses, compare scenarios, and inspect contributions to a selected outcome. These analyses support investigation; they do not automatically prove causation.

You can start with a built-in sample dataset or upload your own CSV or JSON file. Save Excel sheets as CSV before uploading. Your table should contain observations in rows and variables in columns. The platform checks for potential data issues and offers an explained starting-method suggestion. You can follow the suggestion, compare methods, or choose your own. File-size and storage limits depend on your plan.

The recommended starting point is a table with 5–30 numeric columns and at least 500 rows. Suitability also depends on missing values, variation, relationships between variables, and the assumptions of the selected method. More rows alone do not guarantee a reliable result. Review the data checks and method guidance before interpreting your analysis.

Data is hosted in Germany, within the European Union. Uploaded dataset files are encrypted at rest, backups are encrypted, and the platform uses HTTPS. Access to protected data is controlled through account and organization permissions. Two-step verification is available for user accounts and required for platform operators. See the Security and Privacy pages for details about implemented controls and data handling.

The available evidence depends on the selected method. Some runs measure how stable relationships are when the data is resampled; others do not report confidence. Missing measurements are labelled "not reported." Stability is not the probability that a relationship is causally true. A passed engine check means a defined condition was met, not that causation was proven. Some edge directions may also remain unresolved. Review important findings using domain knowledge and appropriate validation before making consequential decisions.

Yes. The Free plan has no subscription fee and no expiry date, and no credit card is required to start. New accounts also receive 14 days of welcome access to Professional features after email verification. Welcome access allows up to 200 analyses in total, files up to 100 MB, and one analysis at a time with a maximum processing time of 10 minutes per analysis. When welcome access ends, your account continues on the Free plan unless you choose a paid subscription. See Pricing for current plan limits.

UCMT Labs offers UCMT v2 Global, UCMT V3 (experimental), and UCMT V4 (experimental), alongside 11 established methods: PC, FCI, GES, GRaSP, BOSS, SCORE, DAS, NoGAM, DAGMA (linear), ICA-LiNGAM, and DirectLiNGAM. Method comparison is available on every plan, including Free. Run selected methods on the same dataset and inspect their shared relationships and differences. Professional and Enterprise plans also include consensus discovery, which retains relationships meeting your selected agreement threshold across methods. Agreement is not proof of causation. See pricing

You can explore relationships, trace direct and indirect causal paths, run counterfactual what-if analyses, and compare alternative intervention scenarios. The XAI view explains contributions to a selected target using SHAP from a fitted linear model when the required dataset rows are available, or structural attribution from graph weights otherwise. The calculation method is identified. You can also review available checks, generate an executive brief, export graphs as PNG, SVG, or PDF, and download relationship tables as CSV. Interpretations depend on the graph and the method used.

You can begin with sample data and Executive mode, which provides a guided workflow with plain-language labels. Data checks and explained method suggestions help you get started. Technical mode provides detailed methods, metrics, and analytical views. Both modes use the same underlying analysis. You still need to understand your data and the question you are investigating. Important causal conclusions should be reviewed with relevant domain or methodological expertise.