Baresquare

When the revenue for a product category is considered an anomaly on a given day (compared to the expected values either calculated by time series analysis or based on business input/targets provided), an analysis is triggered and conducted in two steps: 

The funnel is analyzed by the Funnel Explainer Agent (FEA) which calculates the contribution of each funnel step to the reported revenue anomaly/deviation and expresses it in monetary terms (“incremental impacts analysis”). The step with the highest value in line with the revenue change is the top contributing funnel step.

Depending on the top contributing funnel step identified before, the Root Cause Analysis (RCA) is triggered. This analyzes both internal data from the data warehouse (e.g., additional data from GA4) and the website (e.g., checks on the pages for various issues), as well as external data (e.g., news, special calendar dates). The RCA is conducted in the form of hypotheses that are formulated and checked by the system; they are either retained or rejected. In case the AI Agent is not confident enough to either retain or reject a hypothesis, then the hypothesis is marked as ‘inconclusive’.

1. The funnel is analyzed by the Funnel Explainer Agent (FEA) which calculates the contribution of each funnel step to the reported revenue anomaly/deviation and expresses it in monetary terms (“incremental impacts analysis”). The step with the highest value in line with the revenue change is the top contributing funnel step.
 
2. Depending on the top contributing funnel step identified before, the Root Cause Analysis (RCA) is triggered. This analyzes both internal data from the data warehouse (e.g., additional data from GA4) and the website (e.g., checks on the pages for various issues), as well as external data (e.g., news, special calendar dates). The RCA is conducted in the form of hypotheses that are formulated and checked by the system; they are either retained or rejected. In case the AI Agent is not confident enough to either retain or reject a hypothesis, then the hypothesis is marked as ‘inconclusive’.

The results of the analyses and the tested hypotheses are passed to the AI summarization engine (Natural Language Generator (NLG) Agent). This produces a natural language description of the issue (enriched with visuals), including short-term and longer-term recommendations.