◆ Autonomous Worker
Tracks every live campaign's funnel, attributes outcomes, and produces a post-mortem plus a next-best-action recommendation. Gates a 'winner' call on statistical significance and reports a causal lift estimate with a confidence interval, traced to its data sources.
Memory
Working The campaign and metrics window being analysed.
Episodic Historical campaign benchmarks by segment and channel.
Semantic Attribution models, funnel definitions, statistical-significance thresholds.
Store Results warehouse + benchmark library
Orchestration
MCP
Harness · Managed Agents … scheduled + event-driven runs with a reporting sandbox.
Tools
{ } Campaign + web analytics API ›_ Attribution modelling sandbox Code exec { } Reporting + dashboard service API
Evals & guardrails
- Statistical-significance gating … no 'winner' called without sufficient sample.
- Agent-as-judge review of attribution logic on flagged anomalies.
- Full trace of data sources behind every reported number.
Frontier edge
- ▲Causal attribution: runs counterfactual 'what would have happened without this campaign' analysis rather than crediting last-touch correlation.
- ▲World-model simulation: projects next-best-action lift forward so the recommendation comes with a forecast, not only a post-mortem.
- ▲Proactive anomaly surfacing: flags a funnel break or attribution drift the moment it emerges, before the scheduled report would catch it.
In numbers
under 10 min
Median reporting turnaround
100%
Campaigns under continuous measurement
Handoffs
Fed by ← Campaign Generation Agent