Demand Forecasting and Inventory Optimization with AI

Are you considering elevating your demand planning using artificial intelligence to the next level to:

• optimize collaboration between production planners and algorithms,
• reduce inventory and depreciation costs,
• ensure your delivery capability, and
• save time in manually planning requirements?

On this page, we provide information about demand forecasting, clarify important terminology, offer methodological insights, and demonstrate how we tackle such challenges.

Starting point Supply chain & demand planning: the challenge

Having products or materials ready in meaningful quantities and on time is essential for a functioning supply chain. This should also operate efficiently, cost-effectively, and guarantee high customer satisfaction. Effective demand planning is crucial here. What can you do to get closer to these goals and further optimize the supply chain using data science and AI?

What to do How AI can optimize the supply chain

Artificial intelligence (AI) and data science can significantly enhance supply chain optimization by analyzing large data volumes, recognizing patterns, and enabling informed decisions. Valuable actions include:

Automate planning through forecasting

Manufacturing companies often produce numerous products of different types, including items with regular, seasonal, or intermittent demand. New products with short data histories and small quantities, as well as long-standing best-sellers with high volumes, trendy, and stable products, typically coexist in the portfolio.

To manage numerous planning objects, it is advisable to automate planning as much as possible. The core of many of our demand planning projects is a forecasting system that generates forecasts for products or product groups for the upcoming weeks or months. We emphasize not using one method for everything but accommodating different types of planning objects: whether long or short, seasonal or intermittent demand. We rely on a portfolio of over 30 forecasting methods from statistics and machine learning to address specific challenges, such as intermittent demand frequently occurring in demand planning contexts.

Different forecasting challenges can arise depending on the company and industry, such as:

• What forecasting level is best? Is it worthwhile forecasting every single product, or should groups of products be aggregated, with detailed numbers derived as needed?
• Some orders are known in advance. Such open orders can suitably be considered in the forecasting system to improve forecast accuracy.
• Often, actual demand depends heavily on the number of working days per month or week. There is potential for more accurate forecasts by appropriately incorporating working days into the forecasting system.
• Internal or external factors can influence demand, such as marketing activities, economic indicators, or weather. It is essential to identify and integrate these relevant factors into forecasts appropriately.

Provide tailored tools for planners

The large number of products and forecasts can make navigating through the numerous numbers challenging. Planners face questions such as:

• Which products can be forecasted accurately and which less so?
• Should some products still require manual adjustments?

We always focus on the question: How does a company truly want to work with these results? How can it achieve the greatest benefit? How can the new demand forecasting system best integrate into existing processes and simplify them ideally? The tools and mechanisms applied can vary. Crucially, they should match the company, its culture, and especially the people working with them. Here is a selection of tools that have already made a decisive difference in many projects:

Traffic-light system

A classification mechanism identifies which product groups are forecastable effectively through automated methods and where additional validation by experts is meaningful. We often see particular emphasis placed on planning products with large volumes. Yet, these are precisely the products that can often be forecasted very effectively through data-driven automation. Time spent on their planning might better serve other areas. A traffic-light system presents the results clearly, directing planners’ attention to areas requiring further action.

Performance tracking

By tracking historical forecast performance, planners can assess how well AI-based forecasts have performed in recent months compared to manual planning. Experience shows that adopting a new system becomes significantly easier when its strengths are clearly demonstrated.

Dashboard

A dashboard or frontend clearly presents results. From past demand planning projects, we know what matters and how results can be effectively visualized. Dashboards can effectively highlight product groups worth reviewing and possibly adjusting manually. Companies often already have existing reporting tools like Power BI available, which can also be utilized.

Intelligent AI assistants for demand planning: chatbots and more

Intelligent AI assistants can create Excel reports, presentations, or custom graphics through text or voice input – a game changer for often difficult-to-grasp topics and dry numerical results. This provides an intuitive and easy access to data. Customers are already productively using AI-based chatbots for their demand forecasts. Example questions that an AI-based chatbot for demand planning can answer include: “By what percentage will sales of this product change next year?”, “Please create a table of products expected to experience declining demand!”, “Please create a forecast graphic for product A for the next 6 months!” AI assistants can also actively provide signals and recommendations integrated into existing reporting tools.

Optimize inventory management

Precisely forecasting future demand lays the foundation for inventory optimization, typically focusing on:

• Determining accurate safety stock levels ensures delivery capability throughout lead times.
• Balancing inventory holding costs and customer satisfaction – maintaining sufficient stock without excessive accumulation.

Various approaches and formulas calculate safety stock, based on parameters such as targeted service levels, replenishment lead times, and projected demands. Ideal safety stock values ensure reliable supply and efficient warehouse resource utilization.