MARY | Machine-Assisted Retention Yielder | Customer Retention System

MARY (Machine-Assisted Retention Yielder) is a human-in-the-loop customer retention system designed to support clarity, cost mitigation, and responsible decision-making in customer experience (CX) and service operations. It is intended for environments where automated or AI-assisted responses must remain subordinate to human judgment, governance, and auditability.

At its core, MARY provides a structured framework for generating customer-facing responses that balance empathy, technical accuracy, and business constraints. The system is not an autonomous agent and does not replace human operators. Instead, it functions as an operator-assist framework that improves consistency, reduces cognitive load, and mitigates escalation risk while preserving human authority at all decision points.

This release (v3) includes three primary components:

• CRG v1.1 — a lightweight, zero-shot Customer Response Generator optimized for portability, rapid deployment, and routine customer interactions without deep system integration.
• CRG v2.8 — a structured, system-prompt-oriented Customer Response Generator that enforces clarity, tone discipline, and consistency across multi-turn interactions and retention workflows.
• CRG v4.0 (GABRIEL) — a mentor-level emergent abstraction engine designed to guide operator judgment during complex, ambiguous, or high-stakes scenarios. GABRIEL provides decision framing and contextual synthesis without issuing commands or replacing human agency. Read More: GABRIEL∞

MARY emphasizes explicit constraints, including strict source-of-truth anchoring, hallucination disallowance, business-goal prioritization, and unconditional human override. All outputs generated using this framework are intended to be reviewed by human operators prior to delivery. No automation, outcomes, or performance improvements are implied or guaranteed.

This artifact is released as a frozen, citable publication. Cryptographic integrity (SHA-256) and authoritative identifiers are maintained externally via the Zenodo registry.

Intended audience:
CX operators, operations leaders, AI practitioners, system designers, and researchers interested in human-governed AI application frameworks, particularly in customer service and retention contexts.

License:
Creative Commons Attribution–NonCommercial–NoDerivatives 4.0 International (CC BY-NC-ND 4.0).

Canonical record:
DOI: https://doi.org/10.5281/zenodo.18115752
SHA-256: 8ffe45bc3c894cf5fd6a6f86e8a0f14e8956239084581ffbb49e833aa36ca177

Website:
sunshynelabs.com