Agentic AI for Data Quality: Probabilistic Pivot for Data Quality Management
Year: 2026
Type: conference-paper
Abstract: Poor data quality imposes substantial and measurable costs across industries: organizations lose an estimated ${\$}$12.9 million annually to data-driven decision failures, while the U.S. economy sustains aggregate losses exceeding ${\$}$3.1 trillion per year. Although rule-based validation systems have long served as the primary defense against data integrity failures, they are inherently static, brittle under schema drift, and incapable of detecting previously unseen anomaly patterns. This paper proposes a conceptual framework applying Agentic AI to data quality management, synthesizing three established design patterns: Reflection, Tool Use, and Multi-Agent Collaboration into a unified five-phase validation pipeline. The framework introduces four novel governance mechanisms: a memory fast-path routing layer, a bidirectional Rejection Memory store, a mandatory human-in-the-loop escalation pathway, and a 24-hour rolling source quality monitor that autonomously flags and suspends compromised data sources. Two industry scenarios - insurance and EdTech: illustrate practical application. Conceptual evaluation demonstrates theoretical advantages in adaptability, anomaly detection, and proactive source-level governance over legacy rule-based approaches.
Language: English
Topic: Data Quality and Management
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7202270784