Exploring AI-Powered Reporting for ERP Systems

[Senthil96]

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.

[natashasturrock]

I’ve seen AI really shift ERP reporting workflows. Instead of maintaining hundreds of SQL reports, users can query in natural language and get either summaries or dynamic SQL behind the scenes. A hybrid approach works best: keep optimized templates for heavy reports, but let AI handle ad-hoc needs. Adding anomaly detection and trend insights on top of ERP data has been especially valuable.

For stack: LLM + RAG (pgvector with Postgres plays nicely) + Java integration. To manage costs, cache common queries and reserve bigger models for complex asks. And if data privacy is strict, an on-prem LLM is the safest bet.