Artificial Intelligence & ML Predictions CEO / Admin
Kaltiv integrates 6 machine learning models trained on your real operational data.
Access
ML predictions appear directly in the relevant modules (dashboard, agriculture, sales). No additional configuration is required.
Refresh Cadence
Predictions are automatically regenerated every week (Monday at 04:00 UTC) by a scheduled job that re-scores the models with your latest data (weather, harvests, market prices, sales). The last-update date is displayed in Dashboard → Analytics → "ML Predictions" tab. A manual re-run within the same week never creates duplicates.
Available Models
| Model | Type | Accuracy | Data | Module |
|---|---|---|---|---|
| Yield prediction | CatBoost (ONNX) | R² = 0.79 | Weather + harvest history | Agriculture |
| Quality prediction | CatBoost | 73.7% | Harvest conditions | Agriculture |
| Palm oil price | LightGBM | Trend | Market + seasonality | Sales |
| Palm nut price | LightGBM | Trend | Market + seasonality | Sales |
| Papaya price | LightGBM | Trend | Market + seasonality | Sales |
| Customer scoring | ML algorithm | Score 0-100 | Order history | CRM |
Yield Predictions
Where: Dashboard → "Predictions" section
The CatBoost model analyses weather data (800+ readings) and harvest history (78 records) to predict:
- Expected yield per plot (kg/hectare)
- Optimal harvest period
- Risk factors (drought, excessive rainfall)
Quality Predictions
Where: Agriculture → Plot detail
Assesses expected harvest quality based on:
- Temperature and humidity over recent days
- Bunch maturity stage
- Plot's historical quality
Price Predictions
Where: Sales & CRM → Analytics
Three LightGBM models provide price forecasts for:
- Palm oil: Price per litre, weekly trend
- Palm nuts: Price per kilogram
- F1 Horizon papaya: Price per kilogram
Intelligent Customer Scoring
Where: Sales & CRM → Customer Scoring (/dashboard/sales-crm/customer-scoring)
The algorithm scores each customer on a scale of 0 to 100 by analysing:
- Order frequency and volume
- Payment regularity
- Length of commercial relationship
- Growth potential
AI Advisor — Digital Chief of Staff
Where: Floating button in the bottom right of each page + Settings > AI Advisor
The Kaltiv AI Advisor uses Claude (Anthropic) with 44 specialised tools organised in 4 layers:
| Layer | Tools | Domain |
|---|---|---|
| L1 — Core | 12 tools | HR, leave, payroll, operations, recommendations |
| L2 — Lean | 15 tools | PDCA, 8D, QRQC, Kanban, SPC, 5S |
| L3 — Knowledge | 7 tools | RAG documents, facts, semantic search |
| L4 — Advisory | 10 tools | ML predictions, memory, scheduled reports, external sources |
Advanced Features
- RAG Knowledge Base: Upload documents (PDF, Excel, text) that are automatically analysed and indexed
- Teaching Mode: Teach business facts that the advisor retains and uses
- Proactive Recommendations: Automatic detection of anomalies, trends and opportunities
- Personalised Memory: The advisor learns your preferences over time
- Scheduled Reports: Daily, weekly or monthly briefings generated automatically
- External Sources: Connect Google Drive and RSS feeds for continuous enrichment
For the full guide, see the AI Advisor page.
Each prediction displays a badge indicating its source: ML (trained model), Heuristic (business rule), or Hybrid (combination of both). Accuracy rates are shown to help you assess reliability.
Quality models require 200+ labelled harvests to reach optimal accuracy (76 available currently). Predictions improve with every new data point recorded.