Connect your systems and RootCause.ai resolves them into a governed ontology, joins curated outside data to it, learns the causal structure over it, and answers what happens if you change something. Each part earns its keep alone. The reason to have all five is that they hand the work forward.
Your systems and curated external data resolve into one set of business entities. The ontology layer resolves them, the enrichment layer attaches outside context to those same entities, and the causal layer adds direction and strength between them. Agents query the result over MCP, and planners run simulations against it.
Churn over 90 days if the renewal discount holds
Lead time −3 days
Teams arrive at different points. Some need the semantic layer before anything else can happen. Some already have one and want the causal model. Each part is useful on its own, and each one makes the next cheaper to reach.
Connect warehouses, databases, SaaS apps and spreadsheets. RootCause.ai resolves entity, time and location across them and produces a governed semantic layer your team and your agents can query in plain language. Starting takes no glossary, no catalog and no dbt project.
Foundational Ontologies arrive as columns on the entities you already have, at your time grain and your geography. The sourcing, the licensing and the maintenance are done. A question about whether rainfall moves footfall takes an afternoon.
Claude, ChatGPT and your own agents connect over MCP and query the governed layer instead of raw schemas. Access rules are applied before the query runs, and every answer comes back with its sources, its metric version and its row count.
Discovery runs across thousands of variables on standard compute. Supply your own hypotheses and each one comes back supported, contradicted or underdetermined. Relationships the data cannot orient stay marked as unoriented.
Run interventions, counterfactuals, attributions and optimizations against the same causal model. Each result states the assumptions it depends on and carries an interval that widens as the scenario moves away from conditions you have observed.
Not sure which one describes you? Answer six questions and we will point you at the right one.
Five separate tools can each do a version of this. Every boundary between them is a rebuild, and the rebuilding is where the months go. Three things travel the whole chain here, and each one is work your team does once.
Each one is created at a single stop and consumed at the other four. Assembled from separate tools, all three get rebuilt at every boundary, and the rebuilding is where the numbers stop agreeing.
Three keys get mapped once, and nothing downstream writes its own version of them.
A forecast extrapolates the past, which is the right instrument for planning capacity and for setting a baseline. A decision is a change, and the past holds no version of that change to extrapolate from.
Causal models estimate the effect of an intervention. That is what lets them rank levers, attribute a movement to its drivers, and separate what your team did last quarter from what the market did.
Cost is the reason this has not already been standard practice. Conventional causal discovery scales badly in the number of variables, so it has been rationed to a handful of expert-run projects a year. RootCause.ai runs discovery at sub-quadratic scaling, which is what makes thousands of models practical instead of a few.
Our engine is founded on peer-reviewed research in combinatorial optimization and causal inference, led by Ph.D. experts from the intersection of AI and industrial application.
Every prediction is explainable.
Trace every insight back to the raw data evidence.
Enterprise-grade security for your most sensitive data.
BI tools visualize past data. RootCause.ai explains why the past happened and simulates what happens if you change something, so the output is a decision rather than a chart.
No. The automated data ontology handles entity, time and location resolution, so business leaders can ask questions in plain language and get causal answers. Teams that do have data scientists can bring their own hypotheses and test them against the same data.
Databases, warehouses, object storage, SaaS APIs and flat files. Nothing has to be centralized first, and no source has to be migrated for the ontology to work. There is no requirement for an existing glossary, catalog or dbt project.
Yes. RootCause.ai runs a standard MCP server, so Claude, ChatGPT, Cursor or your own agents connect without custom integration. Agents query governed metrics by name, access rules follow the user rather than the client, and every answer carries lineage back to the rows that produced it.
The digital twin is a causal model of your business learned from your own data. Because the relationships were estimated and tested rather than assumed, you can move a lever and see the effect propagate, with the assumptions stated and the uncertainty quantified.
Every answer resolves to the rows behind it: which sources, which metric version, which model version, how many rows. Every edge in a causal model carries the tests and conditioning sets that put it there, and where the data cannot settle a direction the model reports it as underdetermined rather than guessing.
SPARC-fast is our causal discovery engine, a Screening, Pruning, Agreement and Reconciliation Cascade. It replaces the quadratic conditional independence tests conventional approaches rely on with a test of O(n log n) complexity, which is most of the reason discovery runs at enterprise variable counts rather than on a handful of variables per project.
Bring one real question and the data around it. We will show you the causal model behind the answer.