Industry
Manufacturing
TCS Neural Manufacturing: Driving Agility & a Competitive Edge
Highlights
- While traditional forecasting methods help in short-term production and purchase planning, they don’t take into account a host of other factors that impact demand, which is why longer-term forecasting may not be as accurate.
- Cognitive demand forecasting uses artificial intelligence (AI) and machine learning (ML) to integrate real-time data and identify new patterns.
- AI-led forecasting can help manufacturers improve working capital management, customer service levels, and supply chain responsiveness, while opening up new business avenues.
On this page
On this pageinpage
Modernizing demand forecasting
Demand forecasting has evolved significantly from the manual methods of the 1970s and enterprise resource planning (ERP) to the integrated business planning of the 2000s.
Consumer behavior, too, has changed significantly over time, influenced by factors such as product quality, innovations, sustainability concerns, promotional activities, and competitive market dynamics. Manufacturing companies must leverage new-age technologies to improve their demand visibility and meet dynamic market needs. Forecasting accuracy can be significantly enhanced with advanced modelling techniques using artificial intelligence (AI) and machine learning (ML).
However, many manufacturers still rely on traditional forecasting methods, mostly because it is believed that AI requires better quality data than what is already available. For example, a manufacturing company might use historical sales data to generate forecasts for different time horizons. While this approach helps in short-term production and purchases planning, medium-term manpower planning, and long-term strategic capacity planning, it overlooks several other factors that impact demand, which in turn may reduce forecasting accuracy. This results in overstocking and blocked cash reserves or understocking and subsequently, missed opportunities.
Enhancing forecast accuracy with AI
AI-driven forecasting integrates real-time data and identifying new patterns.
Cognitive demand forecasting uses a wide range of demand signals that can be factored in by complex algorithms. Different weights are assigned to the various factors depending on their impact on demand forecasting. With current technological advancements, forecasting engines can process huge volumes of data to deliver usable insights.
In steel manufacturing companies, for example, the demand for steel can be directly forecasted by historical sales data. However, to increase the accuracy of this forecast, multiple other factors that directly and indirectly influence this demand must also be considered. Direct factors could include missed opportunities, product quality, product pricing, customer service quality, and customer satisfaction. Indirect factors could be competitor performance, fluctuation in raw material prices, regulatory policies, geopolitical influencers, and performance of other related industry verticals.
Similarly, in the automobile industry, demand forecasting is used for forecasting sales of vehicles and spares. Typical data used in forecasting sales are historical sales data, confirmed booking, and dealer projection. Additionally, a wide range of real time, unstructured, and meta data can be used to capture data such as market condition, customer sentiment, promotion, oil prices, and sales data of competitors (see Figure 1).
Figure 1: Indicative real-time demand signals
Manufacturers are unable to adapt to the dynamic nature of these factors using the traditional forecasting methods. Hence, they must turn to advanced modelling techniques that leverage AI and ML, to forecast accurately.
Consider this: a global beverage manufacturer used real-time data and demand sensing, resulting in nearly 25% improvement in forecast accuracy. This was brought about by incorporating the effects of promotions and seasonal variations in their forecasting methods.
In another instance, a retailer benefited realized substantial reductions in its out-of-stock and inventory levels, thanks to cognitive demand sensing. The retailer was able to bring down forecasting errors to less than 30% by using AI-led forecasting, which has enabled approximately 10% reduction in safety stock.
By integrating its demand signals with sales forecast, a leading healthcare product manufacturer was able to optimally allocate stock across its distribution chain, cutting inventory by some 18% and obsolescence by 40%.
Adopting AI algorithms
Even as manufacturers seek to make better use of data to improve demand forecasting, they are exploring ways to adopt AI algorithms effectively.
First, it is important to identify critical demand signals and product categories. These signals should then be classified into structured and unstructured types. When dealing with varied data sources, the different mechanisms for extracting, transforming, and loading the data must be determined. Integration strategies must be defined for seamless data availability.
The selection of an ML model should be based on specific objectives, its ability to process incoming information, self-learning capabilities, and the accuracy of its output. These models can be evaluated and fine-tuned using historical data, and the most suitable ML model can thus be chosen. The significance of the factors or signals on the product category can be determined in discussion with business stakeholders.
Once the model is finalized and forecast is generated, it can be integrated into other input or output systems. ML models can further improve forecast accuracy by analyzing their outputs against actual results. Once stabilized, the organizations can consider taking it to the next level by including more demand signals or extending the solution to cover different product categories, regions, or other factors.
Figure 2 illustrates the step-by-step approach a manufacturer can adopt to implement a cognitive demand forecasting solution.
Figure 2: An approach to implementing cognitive demand forecasting
Digital maturity is critical
The ability to harness data and extract meaningful insights depends on the digital maturity of an organization.
Relying solely on demand forecasting is insufficient for any manufacturer. When coupled with supply and demand visibility, it offers an overarching picture of how the forecast will be fulfilled.
The tenets to achieve cognitive demand forecasting and supply-demand orchestration can range from simple to complex depending on business requirements. This is made possible through coordination of multiple IT systems aligned with organizational capabilities. These systems include engagement, intelligence, record, security, and monitoring. Inputs are received from engagement and record-keeping systems while systems of intelligence help to process data and extract actionable insights.
It is imperative to have a robust mechanism to manage the large volume of data received from multiple sources. Given the exposure to data from numerous public entities, the implementation of robust security and monitoring mechanisms becomes mandatory.
We propose a roadmap for a manufacturer to assess its current capability in demand forecasting and pave the way to reach the highest level (see Table 1). Often, however, moving from one level to the other may require IT investments, which ought to be evaluated judiciously, through a comprehensive cost-benefit analysis.
Parameters |
Level 1 Brick and mortar Manual planning in xl |
Level 2 Systems-driven process Standardization system-led planning |
Level 3 Integrated demand and supply balancing Key customers and suppliers onboarded
|
Level 4 Extended market intelligence, demand sensing |
People |
• Process control based on tacit knowledge • Lack of defined roles and responsibilities |
• Demand planner role is defined with responsibilities |
• Global demand planner, regional and product planner roles are defined |
• Roles across the extended enterprise are defined including channel partners, service providers for providing the inputs and taking decisions |
Process |
• Manual forecasting for high runners for reporting purposes |
• Statistical forecasting is done • Forecast is published based on historical sales • Aggregation or disaggregation of forecast is done |
• Integrated with new product introduction process, sales and operation planning and supply planning process • Consensus based demand forecast is published • Sales and marketing inputs are included before publishing the forecast |
• Real-time visibility across the extended enterprise • Integration of real-time inputs in demand planning process • Demand sensing and publishing the forecast on a daily basis |
Technology |
• Manual |
• Use of standalone commercial-off-the-shelf (COTS) products • Variety of applications for different product lines |
• Advanced statistical forecasting techniques are used • Automation of workflow and gathering of inputs |
• AI-ML based modelling of real-time signals to predict changes in demand • Ease of integration with external and internal stakeholders |
Data |
• Minimum use of data sources |
• Use of historical sales data • Manual planner-based inputs |
• Use of historical sales data • Market intelligence inputs • Data from key customers are considered for demand planning • Inventory positioning |
• Use of a variety of structured and unstructured data sources |
Table 1: A framework for digital maturity assessment
Forecasting for growth
Retail and consumer goods industries have realized significant benefits by implementing cognitive demand forecasting.
The manufacturing industry can also embark on this journey considering the myriad benefits AI-led forecasting offers. By evaluating the organizational maturity from both business and technology standpoints, the approach can be customized to a manufacturer’s specific needs. While the benefits are substantial, there are also a few limitations. Organizations must be cautious in selecting as well as periodically revalidating the factors that influence demand. This approach will help manufacturers improve working capital management, drive better customer service, ensure supply chain responsiveness, and explore new business avenues.
