Dakri Cartons (Mauritius)
How Real-Time RAG & Predictive Forecasting transformed raw intelligence into a 1,295% Year 1 ROI.
Executive Summary
The Challenge: Factory managers were spending 3–4 hours per day manually pulling shift reports, querying multiple disconnected systems, and chasing anomalies reactively—usually after production losses or stockouts had already occurred.
The Solution: BazzAI deployed a domain-aware RAG pipeline over the factory's telemetry. An n8n orchestration layer coupled with Chroma Vector DB and Claude API allowed natural language querying, paired with Holt-Winters forecasting to automate inventory signals.
The Result: Stockouts dropped 40%, OEE improved by 15%, and the payback period on the BazzAI implementation was achieved in just 35 days.
Snapshot
Reduction in Stockouts
40%
OEE Improvement
15%
Payback Period
1.2 months
Financial ROI Framework
Before BazzAI
- Stockouts: 40/month (costing ~$2,500 each in lost production)
- OEE (Effectiveness): 65% utilization
- Manual Reporting: 15 hours/week ($750/week)
After BazzAI (Month 1)
- Stockouts: 24/month (Savings: $40,000/month)
- OEE: 75% (Avoided downtime: ~$60,000/month)
- Manual Reporting: 3 hours/week (12 hours freed)
Investment & ROI
1,295% Year 1 ROI
Intelligence Architecture
Domain-Aware RAG
Instead of rigid SQL queries, operations managers now ask complex questions in plain english: "What caused the Line 3 shutdown?". The system references live SCADA logs, extracts the specific anomaly via Pinecone Vector DB, and summarizes the actionable response through GPT-4o-mini in 60 milliseconds.
Holt-Winters Forecasting
Through n8n orchestration, the system continuously analyzes stock flow cycles, identifying seasonal variations and trend logic using Triple Exponential Smoothing. This actively eliminated 16 stockouts per month without requiring a single human intervention prompt.
Stop Firefighting. Start Forecasting.
Your enterprise data already holds the answers. Let BazzAI build the autonomous pipeline to extract them.