top of page

Artificial Intelligence for Demand Forecasting

Today we are discussing demand forecasting.

Put simply, demand forecasting is the forecasting of demand.

Boom. Discussion over.

Well no, not really.

Demand Forecasting for Supply Chain is a fundamental tool for order planning and general strategy. It helps Inventory Managers plan their monthly orders, understand seasonal trends, save time on re-ordering and reduce Stock-Outs. It’s important to recognise that the demand forecasts are only predictions, and that all predictions have room for error. But modern machine learning methods have achieved up to 50% increase in accuracy over previous methods such as ARIMA and Exponential Smoothing methods. Some of this is achieved because new methods are able to consider many variables simultaneously.

Machine Learning Platforms like Remi AI’s Intelligent Inventory incorporate data streams such as:

Current Price

Future Price Promotions


Public Holidays


Google Analytics

Web crawlers are replacing syndicated data sources

The demand forecasting models take in these variables and then output the expected sales of a single SKU at either a Monthly, Weekly or Daily Level.

Traditionally, it has been difficult to run forecasts at a daily level for low velocity products. This is because many of the machine learning methods are inclined to predict a zero every day for such products. Newer methods developed in-house at Remi AI have allowed us to solve this problem and we can now generate highly accurate daily forecasts for high-value, low-velocity products like vehicles, whitegoods and other products.

This approach is also useful for perishable goods.

Demand forecasting will also help you to delineate between trends and seasonal variations. Seasonality is a critical part of any demand forecasting and it’s one of the reasons that the more historical data you have, the greater the accuracy of the forecast.

Even the most accurate forecasting tools are only valuable if they are actionable. As such, a huge component of a roll out is setting out the methodology and pipeline to actually act upon the forecast. For example, if you run a weekly forecast for 10,000 items, on what planet will you able to view the forecasts in any meaningful way? As such, from running the forecasts they then need to be served in a way that you can generate the most value from them - whether that be through pushing awareness through a rules-based approach, or automation of parts of the pipeline through further A.I integration - A.I Autoreplenishment (which will be explored in an upcoming blog).

If you’d like to make the first steps toward ai-driven demand forecasting head over to our Demand Forecasting API to see what it possible.

Alasdair Hamilton


Remi AI


bottom of page