We pointed DecisionBox at a data set it had never seen. Here's what it found under an hour.

Most AI analytics tools need you to ask the right questions. We wanted to see what happens when you don't ask anything at all.

We took a public Kaggle dataset, a 109 million raw e-commerce dataset from a multi-category store. No documentation, no data dictionary, no context. We pointed DecisionBox at it and hit "Start."

The AI agent figured out the schema on its own, wrote and executed dozens of SQL queries, and iterated based on what it found. 30 minutes later: 33 validated insights and 6 prioritized recommendations, from conversion funnel bottlenecks to revenue concentration risks to catalog quality problems. The key difference: every single finding is validated. Two independent passes run fresh SQL against your actual data to confirm or adjust the numbers. No hallucinated metrics. Every query and reasoning step is fully auditable.

We wrote up the full end-to-end walkthrough from setup (3 commands) to loading data into Redshift to exploring the results: https://decisionbox.io/blog/first-discovery-ecommerce-tutorial

DecisionBox is open source (AGPL v3). Clone it, docker compose up, connect your warehouse, and see what your data knows that you haven't asked yet.