How can AI improve ERP reporting (280+ SQL reports, PostgreSQL, 10+ years growing data, multi-client security)?

We run a multi-client ERP system (PostgreSQL backend) with 280+ SQL-based reports. Users select reports via Java UI, but we face challenges: high maintenance, limited flexibility, performance bottlenecks, and lack of deeper insights.
Context:
• 10+ years of growing store data (daily additions)
• Multi-client setup → strict data privacy/security required
• Heavy daily reporting usage
Looking for input on:
1. Benefits AI can bring to ERP reporting
2. Recommended tech stack (LLM, RAG, vector DB, Java integration)
3. Handling of parameters & report intent (summary/detail, financial/operational)
4. SQL strategy – dynamic AI SQL vs optimized templates
5. Extra insights (trends, anomalies, predictions)
6. LLM cost management for frequent queries
7. Data privacy & security best practices
Would love to hear experiences, recommendations, or case studies.