When you connect a data source, WoEngage's AI agent builds an internal map of your schema — tables, columns, relationships, and inferred meaning (for example, recognizing a column named last_login as a recency signal).
When you write a prompt, the agent breaks it into a plan: which tables are relevant, what filters apply, and what the resulting audience should look like. It then generates the underlying query and runs it as a read-only operation.
Every plan is inspectable. You can open the Reasoning panel on any campaign to see exactly which tables the agent used, the filters it applied, and why it excluded records that didn't match your intent.
The agent improves over time within your workspace: as you approve or edit its suggestions, it factors that feedback into how it drafts future campaigns for your team.