Advancing Demand Planning Decisions Through Causal AI

Overview

The primary goal of this initiative is to harness the power of Causal AI to overhaul and enhance the demand planning operations for companies in the apparel and consumer goods industries. This endeavor aims to accomplish three pivotal objectives:

• To cultivate a deep and systematic comprehension of the causal chains affecting demand, thereby facilitating superior decision-making in real-time.
• To minimize the over-reliance on crucial personnel, whose expertise currently serves as the linchpin of demand forecasting.
• To transition from complex, antiquated spreadsheet models to a dynamic, sophisticated Causal AI-driven platform.

Achieving these goals is expected to bolster forecast precision, curtail instances of stockouts and surplus inventory, and ensure a supply chain that is both nimble and attuned to market dynamics.

Problem Statement

Our client is a multinational consumer goods manufacturer with a highly complex supply chain. The demand planning department oversees thousands of SKUs and places production orders 12 months in advance, based on customer indications and historical data. The stakes are incredibly high, with the manufacturing process initiating 18 months before products actually reach retail shelves. At this stage, there is a risk-laden gamble of committing to large-scale orders before actual demand is fully understood.

The complications don’t end there. Once orders are placed, goods are manufactured in the Far East and shipped, usually by sea, to warehouses in the US and UK. The shipping process itself is fraught with its own set of variables like weather conditions, port closures, and capacity issues. Furthermore, fluctuating costs of raw materials, fuel and containers can drastically affect the cost of goods sold.

Currently, these multi-faceted causal relationships are not adequately captured or understood by the existing spreadsheet-based system. As a result, forecasting inaccuracies are common, leading to financial setbacks, delayed shipments, and stock imbalances. To make matters worse, a significant portion of this complex decision-making resides in the minds of a couple of key staff members, making the entire process vulnerable to the proverbial ‘key person risk.’

Key Research Questions for Causal AI

Understanding the causal relationships between various factors and demand is essential for improving the reliability and responsiveness of demand planning. The chosen research questions serve as the foundation for our causal analysis. They target key variables like promotions, market fluctuations, competitor pricing, and product portfolio changes, all of which have significant impacts on demand. By investigating these questions, we aim to sharpen our forecasting methods and gain actionable insights for strategic planning.

1. What effect do promotional activities have on product demand?
2. What is the most likely cause of sudden spikes or drops in demand?
3. What effect does competitor pricing have on our market share?
4. How does the introduction of a new product affect the sales of existing products in the portfolio?

Solution Overview

We’re tackling the demand planning challenges by using Geminos Causeway’s Causal AI technology.

The goal is to move away from heavy reliance on spreadsheets and a few key people. Instead, we’re using real data and causal analysis techniques to make better decisions in real time. Here’s how we’re doing it:

1. Subject Matter Expert Collaboration

We invited representatives from the shipping, manufacturing, and sales departments for workshops to capture insights into how shipping delays, raw material costs, and promotional campaigns can directly or indirectly affect demand forecasts.

2. Causal Model Validation and Refinement

Initial model results showed that raw material costs are linked to customer demand. However, upon closer inspection alongside project teams and Geminos experts, it was realized that seasonal trends (a confounder) were actually driving both variables. Adjusting for this confounder provided a more accurate model.

3. ChatGPT Virtual SME

Utilizing the Geminos ChatGPT Virtual SME assistant to run preliminary analysis identified a potential new causal relationship between gas prices and raw material costs. Afterward, we consulted with SMEs in the procurement department to verify the finding.

4. Validation of Data Feasibility

Workshops were conducted with the IT team to ensure that real-time shipping data can be integrated into the model. If a key variable like “shipping delays” couldn’t be reliably measured, consultation with third-party shipping partners would be required to explore alternative data sources. In this instance data was available.

The same applied to inventory data from third-party warehouses which wasn’t immediately available, a simplified model using quarterly inventory snapshots was developed as an interim solution, while simultaneously outlining requirements for IT to start collecting more granular data.

5. Creation of an MVP Causal Model

The following shows one of the early versions of the causal model used:

6. Data Hydration

Configured ETL processes to pull in sales data, inventory levels, and raw material costs into Geminos Causeway. This step also involved normalizing different units of measure or time zones to make the data congruent.

7. Perform Causal Edge Strength Analysis

8. Causal Model Refinement

The Causal Edge Strength Analysis previously discussed gave the team valuable information about the model’s accuracy. Our primary concern was to minimize the level of noise affecting the outcome variable relative to other causal factors.
A higher noise level would signal an inaccurate model, necessitating a re-evaluation of the chosen causal variables. In this instance, the analysis showed a manageable level of noise, which, when compared to other variables, was deemed acceptable.

9. Undertake Causal Analysis

After the model had been refined and data had been fully integrated, the project team leveraged various types of causal analysis to confirm and challenge existing hypotheses about demand planning. Each type of analysis served a specific purpose in dissecting and understanding the complexities involved.

• Root Cause Analysis:
  We used the model to pinpoint the underlying reasons behind low demand for a particular product line. A specific component’s quality issues had been negatively impacting overall product perception. 
• Intervention Analysis:
  Simulation was performed to show what demand would look like if shipping times were reduced by 20%. This helped inform decisions about investing in faster shipping options or closer manufacturing facilities, weighing up the cost balance. 
• Outlier Analysis:
  Identified anomalies were there was as a sudden increase in demand for winter clothes in the summer. This helped spot unusual trends early on, in this case a fashion influencer making winter coats trendy out of season.

Causal Variables 

Understanding causal variables in demand planning is crucial for optimizing inventory, improving resource allocation, and ultimately meeting customer demand more effectively. Here, we group these variables by department to give a more targeted perspective:

Understanding these variables will enable a thorough analysis and allow for a more nuanced and effective demand planning strategy. With the assistance of Geminos Causeway and the Causal AI technology it offers, you can develop a robust, actionable, and data-driven plan that considers all these variables.

Challenges

Integrating causal AI into demand planning is not without hurdles, and each challenge requires its own set of solutions:

• Initial Skepticism from Key Team Members: The greatest initial challenge came from the very people responsible for demand planning. There was a natural resistance to replacing the techniques they had developed. The concern revolved around the loss of control and skepticism about the new model’s efficacy. Overcoming this resistance required a tactical approach—directly involving these key stakeholders (who are also Subject Matter Experts) in the project, from inception to completion.
• Data Quality and Availability: Accurate, comprehensive data is crucial. Incomplete or outdated data can skew the model, leading to faulty demand planning.
• Model Complexity: Capturing the intricate cause-and-effect relationships in demand planning is complex and calls for both domain expertise and data science skills.
• Scalability: As the business scales, the model too must adapt to manage larger data sets and more variables.
• Adoption and Trust: Beyond initial skepticism, the model’s broader adoption hinges on clear communication and possibly training sessions to ensure everyone involved understands its workings and benefits.

Key Benefits/Results

• Accurate Demand Forecasts: Through causal AI, the organization can pinpoint the exact factors affecting demand, resulting in more accurate and actionable forecasts.
• Optimized Inventory Levels: Accurate demand forecasts mean better inventory management, reducing “stockouts” and overstocking.
• Strategic Resource Allocation: The insights derived guide more efficient use of resources, from manpower to storage space.
• Alignment with Business Goals: The deep understanding of demand variables can also be aligned to serve broader organizational goals, like cost reduction or market expansion.

By incorporating this causal AI approach in demand planning, the organization went beyond minor adjustments to fundamentally revamp its understanding and management of demand.
The key outcome, to remove key-person risk, was achieved but a significant positive by-product was the ability to make decisions almost in real time, whereas previously an update to the demand planning spreadsheets often took over a week with varying degrees of accuracy.

Conclusion: Harnessing Causal AI for Strategic Demand Planning

In conclusion, the integration of Causal AI into demand planning represents a significant leap forward for businesses in the apparel and consumer goods sectors. By adopting this advanced analytical approach, companies are not only able to decode the intricate causal relationships that drive demand but also to enact more informed and timely decisions that keep pace with market dynamics.

The tangible business benefits of employing Causal AI have been substantial for the organization and include:

• Enhanced Forecast Accuracy: With a more nuanced understanding of causal factors, companies can predict demand with greater precision, reducing the likelihood of stockouts or excess inventory.
• Reduced Personnel Dependency: By decreasing reliance on a few key individuals for demand forecasting, the risk of disruption due to personnel changes is mitigated, leading to a more resilient operation.
• Operational Agility: The transition from static spreadsheet models to a dynamic Causal AI platform allows for real-time responsiveness to market changes, ensuring the supply chain remains flexible and robust.
• Cost Efficiency: More accurate demand forecasting translates to optimized inventory levels, which in turn reduces holding costs and minimizes wasted resources.
• Strategic Resource Utilization: Insights from Causal AI enable businesses to allocate resources—be it labor, capital, or warehouse space—more strategically, ensuring they are invested where they will yield the highest return.
• Alignment with Business Objectives: The depth of analysis provided by Causal AI supports the alignment of demand planning with broader business goals, such as market growth, customer satisfaction, and sustainability

Incorporating Causal AI into demand planning is not merely an upgrade to existing processes—it is a transformative move that positions businesses to thrive in a competitive landscape by leveraging data-driven insights for strategic advantage.