White Box AI

While AI applied correctly to Business Forecasting can remove significant error, the implicit black box nature of AI hinders explainability and hence trust in AI forecasts. Deep Learning Models (such as Neural Networks) create a very high number of non-linear relationships across inputs and outputs. Resulting AI Forecast is difficult to comprehend and adopt if it differs substantially from what Planners expect in terms of level, trend, and seasonality.

vyan.ai takes an approach to optimally blending all the forecasts in its Forecast Universe. This means all the forecast models generate forecasts at all the hierarchy levels in all the time buckets. All such forecasts are then allocated optimal weights that is shown to produce the most accurate signal historically.

vyan.ai ranks all individual forecasts for a given time series and a given lag in terms of the least error forecast as rank #1, next higher error forecast as rank #2, and so on across all forecasts in its universe.

vyan.ai also reports forecast specific weights leading to the composition of Optimal Forecast. This combination of ranks and weights across forecast cycles enable Planners to clearly understand how the AI Optimal Forecast is put together: where Human inputs have been included or not, which models are performing better, at which levels, in which time buckets, etc.

This provides a white-box AI experience for Demand Planners: leading to better comprehension, higher trust, and sustained adoption of the forecast.