You can easily model cause-and-effect relationships

Prior Knowledge Configuration & Causal Discovery

It automatically proposes a causal structure as a directed acyclic graph (DAG) based on the selected data. VELDT employs a system that allows users to choose from multiple causal discovery algorithms. While these algorithms can uncover causal relationships, no single algorithm can perfectly derive the structure; therefore, the system includes features that enable experts to easily make manual corrections. Furthermore, by pre-configuring known cause-and-effect relationships as rules, the system can present highly reliable models from the very first run. In addition to configuration features, the system supports the use of causal knowledge extraction LLM technology to extract causal relationships from text data such as manuals and encyclopedias. (Custom support available) Pre-configured prior knowledge compensates for the limitations of causal discovery.

You can determine whether a causal relationship exists and assess the effectiveness of the measures

Causal Inference

We present the effects of implementing a particular measure versus not implementing it (i.e., when a change is made versus when it is not) in concrete numerical terms. xCausal™ calculates these effects by applying the "adjustment formula" from a theoretically established "structural causal model" to eliminate the influence of other variables on the results. In addition to the Average Treatment Effect (ATE: the average causal effect across the entire population) and the Stratified Average Treatment Effect (the average treatment effect by group), we can calculate the Conditional Average Treatment Effect (CATE), which estimates the effect that varies by individual or entity. We can also identify both direct and indirect effects.

You can see the causes of anomalies and changes

Root Cause Analysis (RCA)

It is possible to identify factors that have a significant impact on key indicators. When an anomaly or malfunction occurs, or when trends in results change, you can check what influenced the outcome and by how much, ranked by contribution rate, from among the other factors within the causal graph. Its true value is realized through system integration.

You can see conditional treatment effects and "what-if" scenarios

Conditional Average Treatment Effect (CATE) and Counterfactual Analysis

The effects of interventions vary around the mean. In other words, there are individuals for whom the effect is high and others for whom it is low; some experience a positive effect, while others experience a negative one. If you are interested in these specific individuals, you can use the conditional average treatment effect (CATE) to estimate how the effect of the intervention would change if the conditions were made similar. It can also address effects under hypothetical conditions that differ from reality—for example, answering a counterfactual question such as, "What if a person in their 60s were in their 20s?"

You can assess the model's reliability and gain insights into potential improvements

Robustness Test and Causal Variables Recommendations

It comes standard with a feature that tests the robustness of the causal graph you have created. You can identify weaknesses such as the need for a third variable—known as a confounding factor—in the relationships between variables, as well as issues like insufficient data or lack of balance. You can also use Smallytics™, a feature that recommends third variables likely to be causally related based on time-series data, and the Assistant feature to get hints for improving your model.

With an interactive UI, you can delegate tasks to AI agents, making analysis more efficient

Assistant (Agentic AI)

Through the conversational UI, you can ask the AI agent to handle time-consuming tasks. For example, evaluating the robustness of a model you've created or exploring improvement strategies based on that evaluation can be time-consuming, but you can leave these tasks to the agent.