Best Practices
Guidelines for effective AI and data governance with Qarion.
AI System Governance
Registration & Classification
Register AI systems as dedicated AI System product types rather than generic tables or datasets. Assign a Risk Classification aligned with the EU AI Act (unacceptable, high, limited, minimal). Document the model's purpose, training data sources, and intended use cases in the product description.
Drift Monitoring
Set up continuous monitoring checks for AI training data and model outputs. Configure data drift checks to detect distribution shifts in input features, model drift checks to catch accuracy degradation, and concept drift checks for evolving relationships. Set alert thresholds that balance sensitivity with false-positive risk.
AI Lineage & Traceability
Use lineage graphs to map the full AI pipeline — from raw data sources through feature engineering to model outputs and downstream consumers. This traceability is essential for EU AI Act conformity assessments and helps quickly identify which AI systems are affected when upstream data changes.
Use Case Documentation
Document every AI project as a structured Use Case before requesting data access. Include the business justification, data requirements, risk assessment, and expected outcomes. This creates the governance paper trail that regulators expect for high-risk AI deployments.
Catalog Management
Documentation Quality
Write clear, concise product descriptions that include business context, not just technical details. Document the update frequency and data sources, and ensure descriptions remain current as products evolve. Avoid leaving descriptions empty, using placeholder text, or assuming users know internal acronyms.
Naming Conventions
Establish consistent naming patterns, such as using lowercase with underscores for technical names. Include a domain prefix (e.g., finance_monthly_revenue) to provide cross-team clarity and avoid abbreviations that aren't universally understood.
Tagging Strategy
Organize products effectively by creating a tag taxonomy before mass-tagging. Apply tags consistently across similar products and review them periodically to clean up unused ones. Limit the number of tags per product to maintain a high signal-to-noise ratio.
Data Quality
Check Design
Effective checks should test for specific, actionable conditions with clear pass/fail criteria. They should run at an appropriate frequency and include meaningful descriptions. Common types include Freshness to ensure timely updates, Volume to detect unexpected row counts, Uniqueness for primary key verification, and Null rates to monitor data completeness.
Alert Management
Triage alerts efficiently by reviewing them daily (or according to your SLA) and linking them to issues for tracking. Close resolved alerts promptly and adjust thresholds if you encounter persistent false positives. To avoid alert fatigue, tune thresholds to reduce noise, disable checks during known maintenance windows, and prioritize critical products.
Quality Improvement
Follow a continuous improvement cycle: Monitor to identify failure patterns, Investigate to find root causes, Remediate to fix issues at the source, and Prevent by adding checks that catch recurrence.
Access Governance
Request Quality
Good access requests provide a clear business justification, are specific about the access needed, honest about duration requirements, and include relevant project or use case context.
Timely Reviews
Approvers should review requests promptly within the agreed SLA. Provide clear reasons for any rejections, consider the principle of least privilege, and track patterns in requests to identify broader needs.
Access Hygiene
Request only what you strictly need and report when access is no longer required. Respond to periodic access reviews and keep your role assignments current to maintain a secure environment.
Collaboration
Meeting Effectiveness
Before meetings, set clear agendas, invite relevant stakeholders, and share pre-read materials. During the meeting, take notes in real-time, create action items as decisions are made, and assign owners immediately. Afterward, follow up on action items, update relevant documentation, and share decisions with the broader team.
Issue Management
When creating issues, write descriptive titles, include reproduction steps for bugs, link to related products or alerts, and set an appropriate priority. To resolve issues effectively, update the status as work progresses, document the resolution for future reference, complete debriefs for significant issues, and consider if new quality checks can prevent recurrence.
Lineage & Impact Analysis
Before Changes
Use lineage to understand the impact of your work. View the product's lineage graph to identify all downstream consumers, then notify affected teams before making changes. Plan your migration or communication strategy accordingly.
Maintaining Lineage
Keep source system connections current and update product metadata after schema changes. Document manual lineage relationships where automatic discovery isn't possible, and periodically review lineage accuracy.
Space Organization
For Admins
When setting up a space, define clear ownership and purpose, configure appropriate permission levels, establish naming conventions early, and set up quality check templates. For ongoing maintenance, review membership quarterly, archive unused products, monitor space activity, and gather feedback from users.
For Users
Keep your profile information current and configure notification preferences. Use preferred tags for quick filtering and report any issues or suggestions to administrators.