Lumeis
All cases

Consumer goods · Food manufacturer

Backtesting cut forecast error by 10.7 pp

Regional demand forecasts went into S&OP without any error measurement. Lumeis OS ran a backtest on the sales history and selected the algorithm with the lowest error.

in production

Results

  • −10.7 ppin demand forecast error, with the algorithm selected by backtesting.
  • 23 of 24weeks in which the algorithm missed less than the regional forecastPeriod: 24 weeks
  • 3data sources connected in one base

The challenge

A food manufacturer with national distribution. Regional managers built the demand forecast by region, in spreadsheets, and it fed the monthly S&OP.

The planning team wanted to know which forecasting method missed less, measured week by week against invoicing.

The solution

Lumeis OS connected these sources in one base, with the source of every number:

  • ERP: orders, invoicing and stock by distribution center.
  • Regional S&OP forecasts, version by version.
  • Promotion calendar agreed with distributors.
“Seeing the algorithm miss less in 23 of 24 weeks changed the S&OP conversation. Now every adjustment has to show it improves the number.”
Demand planning manager, food manufacturer

The results

What we found

  • 23 of 24 · At a food manufacturer, the backtest tested forecasting methods against 24 weeks of invoicing. The selected algorithm missed by 36.4%, against 47.1% for the regional forecast, and won in 23 of the 24 weeks.

    Period: 24 weeks

What changed in the routine

  • The algorithm selected by the backtest became the S&OP baseline.
  • A regional adjustment on top of the baseline comes with a logged reason and has its error measured afterwards.
  • The backtest runs again every cycle, and each forecast’s error is published every week, by region.

The numbers were measured in the client's operation, in the period shown. The client stays anonymous, with a description of the type of company.

Talk about a consumer goods process

Tell us which process hurts the most today.