4 insights
Qualitative + quantitative segmentation of D2C "-Web" voice callers, now expanded with a RICE-scored Product Roadmap (Now/Next/Later, owned by Design/FE/BE/AI-ML) and a Cross-Functional Research Agenda (gaps a prompt engineer, data scientist, UX researcher, consumer researcher, and voice-AI engineer would each flag), on top of first-time vs returning behavior, the Whale/Returner Archetype, and the Token-Warning & Retention finding that hitting zero tokens is a 1.1%-return churn event.
"- Web" agents should run on two messages only — silent-pause and explicit re-up. Both underperform (62.3% / 3.7% compliance), the re-up warning's timing is inconsistent (6-98s before call end), and a lander-only message is leaking onto 2 agents that shouldn't have it.
Fleet-wide before/after review across all 34 "- Web" agents: false-positive rate on end_call collapsed from 68.9% to 11.6% around Jul 1-2 2026; "prolonged silence" false positives went from 266 to zero.
Working dashboard for GTM-1771. STXT licensed creators ranked by voice call activity. Live voice call metrics from BigQuery, profile data from D2C API. Surfaces blockers (voice training, OpenPay, WhatsApp access) per creator. Auto-synced via GitHub Actions from synthetic-turing/creator-crm.