Caprari cuts forecast error by 69% with machine learning on AWS

17 Nov 2026 | Case Study, Data Platform, Machine Learning, News

In questo articolo

    Condividi

    How a pump manufacturer with more than 6,000 products moved from spreadsheets to automatic, reliable demand forecasts


     

     

    Customer

     

    Caprari designs and manufactures pumping systems for water management, used in civil, agricultural and industrial applications all over the world.

    With about 6,140 products and very different sales patterns, forecasting demand was a monthly struggle. One planner spent about 160 hours a month on spreadsheets and statistical methods, and the forecasts still missed by 78% on average. The result: stock-outs on some items, excess inventory on others and production plans that kept changing.

    Solution

    Omnys replaced the manual process with an AI forecasting engine on AWS. Instead of one model for everything, an ensemble of machine learning models, including Chronos, Amazon’s time-series foundation model, learns the behaviour of each product family. Every month, as new sales data arrive, the forecasts for the next six months are ready with no manual work.

    Results

    Caprari wanted forecasts more accurate than its statistical methods and a forecasting process reduced to a quick review. It got both:

    • • 69% lower forecast error, from 78% to 24%.
    • • 90% less effort: from about 160 to 16 hours a month.

    Planners now spend their time on decisions instead of spreadsheets, and inventory follows real demand more closely.