◆ Autonomous Specialist
Clusters customers on behaviour, lifecycle and product fit to define targetable segments, and tracks drift to refresh them as behaviour shifts. Enforces consent and marketing-eligibility so opted-out customers never enter a segment; emits segment membership to the campaign agent.
Memory
Working The segmentation run and feature set in play.
Episodic Prior segment definitions and their campaign yield.
Semantic Lifecycle stages, product taxonomy, consent/eligibility rules.
Store Feature store + segment registry
Orchestration
MCP
Harness · Managed Agents … sandboxed code execution against the analytics warehouse.
Tools
{ } Customer analytics warehouse API ›_ Clustering + modelling sandbox Code exec { } Consent + marketing-eligibility service API
Evals & guardrails
- Consent and marketing-eligibility enforced … opted-out customers never enter a segment.
- Fairness check that segmentation doesn't proxy for protected attributes.
- Segment-stability monitoring with drift detection.
Frontier edge
- ▲Causal reasoning: distinguishes behavioural cohorts from spurious correlations, so segments reflect why customers act, not co-moving noise.
- ▲Confidential compute: clusters behavioural features inside a secure enclave; downstream campaigns get segment membership, never raw customer records in clear.
- ▲Proactive segment discovery: anticipates emerging cohorts (e.g. early churn-risk drift) before the campaign agent requests them.
In numbers
240
Active segments maintained
daily
Refresh cadence
Handoffs
Hands to → Campaign Generation Agent