Kakao Recommendation Team Unifies AI Analysis Standards with Data Catalog and Analyst Persona

Details

Designed a common team context to resolve the issue where results vary depending on log selection and criteria, even when analyzing the same metric
Separated three contexts—Data Catalog, Analyst Persona, and Task Context—to explicitly define data definitions, analysis methods, and task backgrounds
The Data Catalog documents usage criteria and aggregation rules per table to eliminate data tacit knowledge, while the Analyst Persona standardizes a 6-step analysis process and pitfall checks
The Task Context briefs the purpose and current status of the task through Jira issues to determine the direction of the analysis
In an A/B experiment case, applying this structure led to an intermediate conclusion that there were no loss signals but gains were unconfirmed, and a proposal to extend the subsequent observation period
The role of the analyst shifted from repetitive data extraction to focusing on problem definition, establishing success criteria, and making final judgments

Key point