Most businesses are beset by numerous challenges today such as shorter customer tolerence times, high-quality expectations, short product life cycles, uncertain demand and supply, high costs, and margin pressures. What can help them succeed amidst these challenges is to execute an effective supply chain management process, and demand forecasting can form the foundation on which businesses can execute their supply chain systems, from procurement and inventory and warehouse management to distribution management.
Making accurate forecasts has always been a pain point for businesses. But with artificial intelligence (AI) and machine learning (ML), they can make specific predictions. In its report, Hype Cycle for Artificial Intelligence, 2018, global research firm Gartner found that ML is at the peak of inflated expectations and is central to supply chain analytics in the future.
Challenges in traditional forecasting
Forecasting accuracy has typically been hindered by siloed operations, scattered data in legacy systems, exaggerated or conservative manual adjustments, and misaligned strategic and tactical planning. Moreover, the use of old-fashioned methods such as spreadsheets and a lack of collaboration among sales, marketing, and planning teams has only added to the complexity.
Traditional forecasting techniques such as the Naive model, Moving Average model, Holt-Winters model, and Auto Regressive Integrated Moving Average (ARIMA) model measure few variables like trends, seasonality, and cycles. They fail to account for internal and external factors in the real-world that significantly impact demand forecasting.
In the era of Industry 4.0, where firms are more connected and have better data visibility, manufacturers are moving away from traditional time series forcasting to more advanced demand methods that leverage ML, cognitive computing, and real-time internet of things sensor data.
Forecasting is all about how efficiently companies use the available data and derive actionable insights. Manufacturing companies have plenty of structured and unstructured data at their disposal. Cutting-edge technologies like ML can mine such vast data pools and can provide accurate forecasts to date.
How ML facilitates demand forecasting
When ML is applied to demand forecasting, it not only analyzes statistical input, such as historical sales order data, but also considers internal and external causal factors that affect demand. This improves forecast accuracy.
Internal factors refer to new product launches, distribution network expansion/reduction, product price changes, product seasonality, new sales channels, lost sales due to stockouts, promotions, discounts, and more.
To better understand the external factors affecting demand, let’s group them under PESTEL:
- Political factors: Addition or reduction in taxes, government policies of newly imposed regulations and public holidays among others.
- Economical factors: Unemployement rate, fuel prices, labor wages, and currency fluctuations
- Social factors: Customer sentiments on social media platforms, media coverage, competitors’ promotions, and more.
- Technical factors: Product life cycles and upcoming technology trends.
- Environmental factors: Weather conditions, natural and human calamities, and disasters among others.
- Legal factors - New safety and environment regulations
While traditional demand forecasting methods largely ignore external factors, ML goes beyond a firm’s immediate context when analyzing demand, as these scenarios illustrate:
- Manufacturing companies can use ML and big data to examine tweets and posts on websites and social media to understand customer sentiments about their products.
- ML can teach self-learning algorithms to analyze the past impact of currency fluctuations and then predict better forecasts.
- ML can predict future weather patterns at the local level and identify how it connects to local demand patterns. ML can also determine if a lag exists between the weather changes and the demand of products on a real-time basis.
- The life cycle of a product plays a critical role in demand forecasting. Using ML to understand how a solution performs, right from the time it is introduced to its growth and decline in the market, leads to improved predictions of demand.
ML can not only provide forecasts on demand, trends, and seasonality but, unlike traditional statistical forecasting techniques, it can also make predictions over and above random demand patterns.
Benefits of automating forecasting
The upside of using ML is plenty as it directly impacts the supply chain end-to-end. It reduces working capital and the number of obsolete materials.
It also ensures that sales opportunities aren’t lost due to stockouts, reduces premium frieght costs, and improves perfect order fulfillment and cash-to-cash cycle time.
Large organizations are also deploying ML in their supply chain processes. This year, food and beverage major PepsiCo, cosmetic giant L’Oréal, and manufacturing conglomerate 3M were among leading organizations in Gartner’s Supply Chain Top 25 rankings that use ML in their daily forecasting and planning. While these companies are more mature in terms of data availability and readiness for ML, other manufacturing companies must catch up quickly, as the machines look poised to take over the future of demand forecasting.