AI case study 1: demand forecasting using artificial intelligence

AI case study 1: demand forecasting using artificial intelligence

How machine learning helps manufacturers cope with volatile demand, with examples from Danone and L'Oréal and the benefits, impact and data requirements.

by Tradecloud Marketing Team

June 29, 2021

Illustration for Tradecloud's AI case study on demand forecasting

The business problem

Demand volatility is one of the biggest challenges for business leaders. More data is available than ever, yet customer purchasing patterns are becoming more complex and harder to detect or predict (Symphony Retail).

So many factors influence demand, from the weather to social media influencers, that customers often change their minds. The events that shift customer intentions usually happen unexpectedly (AltexSoft, 2019).

Traditional forecasts are only as good as the data, models and resources behind them, and the people who interpret them (Symphony Retail). So how do you respond to these challenges?

The solution: AI-powered demand forecasting

Traditional demand forecasting is a form of predictive analytics that estimates future demand from historical data (Dilmegani, 2021). Machine learning (ML) goes further. ML algorithms can detect changes in customer demand, recognise patterns, uncover complex relationships in large datasets and pick up signals of demand fluctuation.

Organisations use AI forecasting mainly to avoid the inefficiencies that come from a mismatch between supply and demand. A forecast will never be 100% accurate (AltexSoft, 2019). Still, AI can significantly reduce supply chain costs and improve financial planning, workforce planning, profit margins and risk assessment (Dilmegani, 2021).

Traditional versus machine learning forecasting

CriterionTraditional forecastingMachine learning forecasting
Handling many variables and data sourcesAdding extra variables and sources takes substantial effortMultiple variables and sources can be added smoothly thanks to a high level of automation
Manual workHighLow
Data requiredSmallLarge
Maintenance complexityLowHigh
Technology requirementsLowHigh
Best fitMid to long-term planning, established products, stable demandShort to mid-term planning, new products, volatile demand

Source: AltexSoft, 2019.

Example: Danone

French food manufacturer Danone implemented a machine learning system to improve its demand forecasts. The company needed more accurate and reliable forecasts because its fresh products have a short shelf life and demand changes quickly (Brosset et al.).

More than 30% of Danone's total volume is sold through promotions such as discounts and leaflets. That made forecasting somewhat ad hoc (Best Practice AI). The ML system improved the forecasts and the planning between sales, supply chain, finance and marketing. It also improved efficiency and the balance of inventory, which helped Danone reach its target service levels for channel and store inventories (Brosset et al.).

Benefits

These are the main benefits of AI demand forecasting for a manufacturing company:

  • Accuracy improves over time. Machine learning keeps learning from the data it receives.
  • Higher customer satisfaction. Products that are out of stock lower customer satisfaction. Consistent availability builds loyalty and a better brand perception.
  • Better workforce planning. Forecasts help HR find the right balance between full-time and part-time staff, which optimises HR costs and effectiveness.
  • Better markdown and discount planning. Products that stay unsold longer than expected tie up cash in stock, raise holding costs and increase the risk of obsolescence. Better forecasts reduce this and the need to sell at low prices.
  • Overall efficiency. Teams can focus on strategy instead of constantly adjusting inventory and staffing (Dilmegani, 2021).

Impact

According to Dilmegani (2021), AI-driven forecasting can:

  • reduce errors in supply chain networks by 30 to 50%
  • lower warehousing costs by roughly 10 to 40%
  • cut lost sales due to out-of-stock situations by up to 65%

The total impact of AI on manufacturing and supply chain planning is estimated at 1.2 to 2 trillion dollars (Dilmegani, 2021).

For Danone, the results were (Brosset et al.):

  • 30% fewer lost sales
  • 30% less product obsolescence
  • 20% fewer wrong forecasts
  • 50% less workload for demand planners

Accessibility and requirements

An ideal AI forecasting system is trained on a variety of data sources. Think of weather data, financial data and third-party data such as social media, historical sales and macroeconomic data. The system learns how combinations of past events affected customer demand, and uses that to predict future demand (Brosset et al.).

Example: L'Oréal

French beauty manufacturer L'Oréal combined social media data, weather and financial market indicators and point-of-sale data (collection, receipts and inventory) to anticipate trends and predict demand. By finding out which combinations of variables drove customer demand, L'Oréal could target end customers more effectively and respond better to demand volatility (Brosset et al.).

Human expertise remains essential

AI does not replace people. You still need human judgement to assess the outputs of an AI system and to apply common sense and domain expertise. Companies should invest in industry specialists who decide which factors belong in the predictive models. The combination of human and artificial intelligence gives you the best basis for planning ahead (AltexSoft, 2019).

Also read AI case study 4: machine learning in the manufacturing process.

Sources

Curious how better forecasts and supplier data come together in your supply chain? Book a demo.