Technology

Most of the internal and external data used by raw material buyers belongs to the category of streaming data, or "time series": time stamped, sequentially revealed, with the time factor a critical part of the message.

Datapred's market dynamics module and demand prediction module use a proprietary, award-winning time series modeling engine. Notable benefits include:

• Minimization of model risk via model aggregation
• Ability to work with "small data", thanks to the aggregation of frugal models
• Reactivity to market shifts, thanks to "sequential" learning process
• Explainability - total transparency on model and input contribution to performance
  

For cost/constraint optimization and purchase order suggestions, Datapred include an optimization engine built on top of OR Tools, an open-source optimization suite maintained by Google.

Datapred's "secret sauce" in that respect consists in:

• Adapting that framework to the context of streaming data
  
• Adding a few procurement-specific optimization tweaks
  
• Smoothly integrating optimization and prediction
  
• Optimizing computational performance

But the basic genericity of OR Tools is preserved, meaning that Datapred can integrate any business rule or optimization model you may already be using.

Datapred is a modular software, and users with data science competencies can take advantage of that modularity to play with the market analysis and the optimization engines, using a standard Python console:

• Add/substract contextual data streams and check the impact on predictive performance
• Test new predictive horizons
• Add/substract models to our standard portfolio of parametric and non-parametric models
• Add/substract parameters to the optimization engine...