Segmentation Agent

⬜ Proposed — not built. Tier 3 — blocked

What it would do

Proposes customer cohorts and the moment to reach each one, inferred from behaviour rather than fixed rules.

Why an agent

The weakest agent case of the six, honestly. Executing a segment is SQL. Only proposing segments from behaviour is judgment — so scope it as a proposer sitting on top of a rules engine, not as the engine.

What it needs first

CRM integration. Not started, and it gates all of BTL and Program.

Roadmap source

A3 Segments (page 2) — marked x on the whiteboard, alongside micro-segmentation, real-time engagement and intelligence — see roadmap_ref_extracted.md.

Design notes and open questions

No PII into prompts. Records must carry cohort attributes, never identifiers.
Never auto-execute on a segment — propose, a human approves.
Consent, DND and suppression are absent from the roadmap entirely and are regulatory requirements for Indian outbound, not features.

Subagents

Cohort proposer — segments inferred from behaviour rather than fixed rules
Trigger picker — which moment to reach that cohort
Fatigue checker — has this customer already been contacted this week, by any program

Deterministic, deliberately not subagents: eligibility rules (SQL, and they must be auditable) and the PII shield — records carry cohort attributes only, never identifiers. No customer identifier reaches a prompt.

Stub spec. Expand to the full overview.md + technical.md split (as the three built agents use) when this moves from proposed to scheduled.