Let's Talk

Why Your Demand Forecasting Model Doesn’t Work and Proven Ways to Fix It

Advanced Analytics
Demand Forecasting
30 Jul 2024
7 min read

Ed Coronado

Over the past three years, accurately forecasting demand has become more challenging due to external factors such as the COVID-19 pandemic, global trade wars, and skyrocketing inflation. As manufacturers grapple with additional disruptive elements like climate change, geopolitical tensions, and the need to realign supply chains, it is clear that returning to the previous state of “normal” is unlikely. This evolving landscape requires a reevaluation of forecasting strategies to build resilience and efficiency in the face of ongoing uncertainties.

Old forecasting models, which rely heavily on deterministic forecasting and historical data to predict future demand, have proven insufficient in today’s volatile environment. Evidence shows that these models often fail to account for rapid changes and unexpected events. According to a 2021 study by e2open, the rate of planning errors is alarmingly high, with more than half (54%) of manufacturers’ forecasts being wrong. This significant margin of error underscores the need for more sophisticated, adaptable forecasting methods that can handle the complexities of modern supply chains.

In this interview, demand forecasting expert Ed Coronado shares insights on the benefits and challenges of demand forecasting, as well as how modern methods and AI are helping to overcome these challenges and improve accuracy. Ed is an experienced data, analytics, and AI expert who has worked with global organizations to leverage the power of data, analytics, and AI to improve forecasts. Ed’s expertise has led to improved production, optimized inventory management, enhanced profitability, and happier customers.

Why is accurate demand forecasting important for businesses?

Demand forecasting is nothing new for businesses. Accurate demand forecasting is like an operational crystal ball, helping businesses anticipate future needs and align everything from production, to staffing, to inventory levels, and more. For example, a retailer can use forecasts to ensure they have enough stock of popular items during peak shopping seasons, minimizing the risk of stockouts. This minimizes costs associated with excess inventory or stockouts and allows companies to respond swiftly to changes in consumer behavior, economic conditions, and competitive landscapes.

What are some of the challenges of accurate demand forecasting?

Accurate demand forecasting relies on numerous technical and organizational factors, including a clear alignment with the business goals, access to diverse datasets, and an understanding of market fluctuations. However, the biggest challenges for most organizations are their approach to sourcing and integrating quality data and the introduction of external data. Traditional modeling approaches, such as a deterministic approach, often rely on simple statistics and historical data, which struggle to address factors like seasonal variations, economic shifts, or sudden changes in consumer preferences. For instance, a sudden economic downturn can drastically change consumer spending patterns, which traditional models might not predict. They are also difficult to adapt and scale to today’s complex markets and high-volume datasets.

How do organizations address the challenges?

The first thing you want to do is make sure you have the right strategy in place; this includes having clear business goals, including desired outcomes, and ensuring you have access to the data you need to run your models. You also need a fast, secure, and well-architected data foundation to keep them running smoothly.

Once you have your strategy and foundation in place, you’ll want to consider your approach. As I mentioned earlier, many demand forecasting programs are plagued with inaccuracies due to the use of outdated models and techniques. I often recommend a solution that blends more flexible stochastic models with the addition of machine learning and AI. This approach continuously learns and adjusts based on new data, enhancing forecast accuracy and reliability. Additionally, I recommend taking an iterative approach to applying different machine learning models, validating and comparing the results of each.

What would you say are the key components of a modern forecasting program?

A modern forecasting program must incorporate several key components to ensure accuracy and adaptability. First, having the right data is essential. Forecasts need to move beyond just sales history and include a comprehensive view of both internal and external factors. This means integrating sales, customer, vendor, and inventory data with external factors like climate, macroeconomic conditions, and supply chain dynamics.

Second, stochastic optimization is a game changer. Unlike deterministic approaches that assume a single outcome, stochastic optimization acknowledges the variability in future demand. This allows businesses to better understand and evaluate a full range of possible outcomes, enhancing your ability to make informed decisions.

Third, leverage AI and machine learning to process vast and diverse data and extract complex insights and patterns that are too intricate for manual approaches. This leads to more accurate and reliable forecasts and, at the same time, allows for scenario planning.

Fourth, adopting a holistic viewpoint is vital. Viewing demand as a single value ignores the inherent fluctuations of today’s global market. By considering demand as a probability distribution, companies can set their risk tolerance and plan accordingly, which helps mitigate risks and seize opportunities.

Finally, having clear business goals is imperative. Organizations that optimize their forecasting toward specific goals—such as reducing costs, managing inventory, optimizing price, optimizing supply chain, or preventing stockouts—are better able to tailor their activities and leverage forecast results to support their desired financial outcomes. This strategic alignment ensures that forecasting efforts directly contribute to achieving business objectives.

How do you recommend getting started with a modern approach to demand forecasting?

A lot of the specifics will depend on the industry, the organization’s technical foundation, and their previous experience with demand forecasting. However, here are a few general best practices to consider when getting started:

Define objectives and scope: Clearly define program objectives and focus points (For example, are you focusing on inventory management, production planning, or sales forecasting? What’s the benefit to the company as well as the customer?), and determine the full scope of your forecasts, including the level of detail needed for decision-making.
Collect and prepare data: Identify your relevant internal and external data sources, cleanse and validate that data to ensure quality, and centralize it in a common repository or data warehouse for accessibility and analysis.
Select forecast models: Consider your product characteristics, market dynamics, and accuracy needs, as well as the benefits and drawbacks of statistical and machine learning models. Determine your needs and assess model performance using historical data to select the most suitable approach.
Implement and calibrate models: Deploy your chosen model(s) using appropriate tools and fine-tune parameters to fit specific data patterns and business context. As a best practice, avoid overfitting the data to the model, as it can potentially skew the results and limit the ability of the model.
Generate and evaluate forecasts: Regularly produce forecasts for defined scopes and timeframes and continuously monitor their levels of accuracy and potential biases. Adjust models as necessary to better fit needs and increase accuracy.
Integrate and automate forecasting: Integrate forecasting with ERPs, CRMs, inventory systems, and other technologies for seamless data flow and automate data collection, updates, and forecasting to streamline operations.
Monitor and improve: Continuously monitor performance and refine models, update data sources, and incorporate feedback to enhance accuracy and drive adoption.

Ed Coronado is an experienced leader in data, analytics, and AI. In his role at Wavicle Data Solutions, Ed has helped companies design and implement technical solutions, including demand forecasting, to solve complex business challenges. If you’re interested in learning more about how a modern approach to demand forecasting can help you make better operational decisions, reduce risk, or increase profits, please reach out to Ed on LinkedIn or at info@wavicledata.com.

This article was originally published as Why Your Demand Forecasting Model Fails and Proven Ways to Fix It on July 30, 2024, on CDO Magazine.

WIT Leader

Ed Coronado

Ed brings 25+ years of experience building cloud and data platforms across Snowflake, Azure, AWS, and Databricks. He leads Wavicle’s tech COE with deep architecture and solution design expertise.

View all my Posts

Blog

Databricks Genie: Transforming Enterprise Conve...

16 Sep 2025
7 min read

Prantika Sinha

Blog

Navigating Cross-Industry Moves in Data Science...

27 Aug 2025
21 min read

Beverly Wright

Blog

Modernizing Healthcare Governance and Security ...

07 Aug 2025
6 min read

SERVICES

SOLUTIONS

PARTNERS

FEATURED SOLUTIONS

RETAIL

Retail​

HEALTH & WELLNESS

Healthcare

MANUFACTURING

Manufacturing​

COMPANY

News

Behind the Badge: A New Milestone in Our Microsoft Data & AI Journey

CAREERS

CASE STUDIES

INSIGHTS

Why Your Demand Forecasting Model Doesn’t Work and Proven Ways to Fix It

Ed Coronado

Why is accurate demand forecasting important for businesses?

What are some of the challenges of accurate demand forecasting?

How do organizations address the challenges?

What would you say are the key components of a modern forecasting program?

How do you recommend getting started with a modern approach to demand forecasting?

WIT Leader

Ed Coronado

Related Topics

Related Posts

Databricks Genie: Transforming Enterprise Conve...

Prantika Sinha

Navigating Cross-Industry Moves in Data Science...

Beverly Wright

Modernizing Healthcare Governance and Security ...

Data Team

Smarter Tech, Leaner Systems: Reducing Cost Wit...

Beverly Wright

How Power BI Auto Aggregations (on Databricks) ...

Cloud Team

Computer Vision for Health: Living Longer

Beverly Wright

What We Get Wrong About AI Expertise

Beverly Wright

5 Major Benefits of Amazon Quick Suite That you...

Prantika Sinha

AI Doesn’t Negotiate Tariffs – But It Can Help ...

Data Team

Exploring Amazon Quick Suite Community, Feature...

Prantika Sinha

Tips and Tricks to Get the Most Out of Amazon Q...

Prantika Sinha

Rethinking Healthcare Data Governance: From Sil...

Duane Lyons

Building the Future of Healthcare Through Flawl...

Duane Lyons

Leveraging AI to Optimize Energy Consumption of...

Beverly Wright

Predicting the Unpredictable: Leveraging AI to ...

Tom Lin

The AI Storefront: How Retail and CPG Leaders C...

Retail Team

When and Where GenAI Actually Makes Sense in Kn...

Beverly Wright

Why Enterprises Must Shift from General Intelli...

Data Team

Navigating Ethical Issues of AI in Retail

Retail Team

How Generative AI is Transforming Retail Custom...

AI Team

AI in the Real World: Balancing Innovation Risk...

Beverly Wright

Back to Basics: Essentials for Product Developm...

Beverly Wright

A Perfect Match: SAP Business Data Cloud and Da...

Data Team

How Text Analytics and Generative AI Are Unlock...

AI Team

Transforming BI Reporting and Visualization Wit...

Data Team

Mastering Cloud Cost Optimization for a More Ef...

Cloud Team

How Generative AI is Transforming the Retail Ex...

Retail

Manufacturing