AI Transparency and Explainability
Learn how our AI works, its accuracy, and the safeguards we have in place.
What AI Model Do We Use?
Amazon Chronos-2 (Public research model)
Architecture: Large Language Model fine-tuned for time series
Training set: General time series data (NOT your data)
Your data: Used ONLY for generating predictions, not to train the model
How It Works (In Plain Language)
- You provide up to 24 months of sales data
- Chronos-2 analyses patterns (trends, seasonality, anomalies)
- The model predicts the next 30/60/90 days
- We display the confidence interval (MAPE %)
Is It Accurate?
Average deviation: 15-25% (industry standard). Unforeseen external factors (market changes, external stock shortages) may increase this margin significantly
MAPE is a measure of deviation. The lower the percentage, the more closely the prediction matched your real history
Example:
Actual: 100 units
Prediction: 110 units
MAPE: 10%
We display MAPE for each prediction.
No Dark Patterns
We do NOT hide uncertainty
We DO show confidence intervals
ApisDom is a data visualisation tool, not a business manager. You decide
Predictions are EXPLAINABLE (based on your data)
Bias Testing
Model tested across more than 50 industries
Consistent MAPE across all product categories
No demographic bias (we do not see customer data)
Human Control
YOU make ALL decisions
The algorithm is a decision-support tool, not a decision maker
You can reject ANY prediction
Predictions do NOT execute orders automatically
Audit Log
Every prediction logs:
- Input data (dates, quantities)
- Model version
- Confidence interval
- Timestamp
Prediction logs are kept for 12 months for technical efficiency and are available upon request through the control panel.
Contact
Questions about our AI?
legales@apisdom.com