Written by Tali Remennik, CEO & Co-Founder at Granularity
For companies striving to stay ahead of the competition, demand forecasting provides an invaluable lifeline. By delivering predictive insight into consumer trends and interests, businesses can make crucial decisions in uncertain times with absolute clarity on their resources – positioning them firmly for success no matter what emerges from today’s ever-changing business landscape.
AI and machine learning technologies play a significant role in demand forecasting by providing more accurate and precise predictions of future demand. One of the key benefits of using AI and machine learning algorithms in demand forecasting is their ability to consider a wide range of factors that can impact demand, including seasonality, promotions, and market activities.
With the ever-growing demand for customer satisfaction, incorporating a ‘customer first’ mentality into forecasting methods has become increasingly essential. This article examines how this approach can be leveraged to determine and predict consumer needs more accurately while taking various influencing factors as well as potential benefits of upgraded forecasts into account. Furthermore, it explores how Artificial Intelligence is revolutionizing the field by enabling analysis of large amounts of data points with utmost precision.
What is demand forecasting?
Demand forecasting is the process of predicting how much of your product will be purchased over a future period. Generally speaking, this has historically been calculated by using past precedents to predict future demand.
Successful demand planners take a holistic approach that revolves around customer segment. Trying to predict future sales without predicting customer demand is like driving a car with a broken GPS. In both cases, you’re navigating based on outdated information and not accounting for real-time changes and variables, resulting in a higher likelihood of getting lost or making costly mistakes.
The best definition of demand forecasting is then:
Demand forecasting is the process of predicting how much consumers are interested in buying your product.
The main difference between the two definitions is the demand planner taking a customer-first mentality. Consumers drive product demand, and forecasts need to model consumer interest and behavior as much as demand for their particular version of the product.
Factors influencing demand forecasting
Demand forecasting requires modelling a chaotic variable: human behavior. The factors that influence demand for your product are extremely important, but also number in the hundreds potentially. It can be daunting to determine, let alone gather, all of these factors.
Traditional forecasting used factors centred around what’s called autocorrelation and autoregressive modelling. Essentially, the factors considered focused on similarity of a time series against itself over successive periods. Frequency patterns allowed for analysis that better predicted this coming December’s demand for chocolate reindeer using autocorrelation of many sub-time periods.
This clearly isn’t the full picture. Using the customer-first mentality definition of demand forecasting above, you can see we’re missing something: the customer.
Imagine you decide to purchase a travel backpack. First of all, how did you come to the realisation that you wanted to purchase a travel backpack?
Then how do you go about the process of buying a travel backpack? You google for travel backpacks and research which one is right. You visit the website of some bag companies and maybe visit several outfitter stores to ask the experts.
That example demonstrates key factors that drive consumer demand behavior: interest and action.
Interest can be measured in many ways, depending on the type of product. In today’s interconnected online environment, social media is often the introduction of a product or re-introduction of a product to a consumer’s interest. They saw their favourite influencer backpack, so maybe that would be fun. Weather data has a surprising impact on interest for many product types as well.
Action is the most available factor. This includes all of the current factors such as autocorrelation and frequency patterns we talked about above. But there’s more available: you can gather search frequency on search engines for key terms related to your product. And action may be limited by macroeconomic factors; wallets might be tighter for extravagant purchases such as kayaks in a recession.
Expanding the scope of factors such as we just did causes many headaches for traditional methodologies as we’ll see below. That’s where AI demand forecasting is needed.
What are the 4 demand forecasting approaches
There are four key demand forecasting approaches: Top-Down / Bottom-up / Middle-out approach, Product Lifecycle Approach, Account Based Forecasting, and Product Classification.
Top-Down / Bottom-up / Middle-out approach
The Top-Down approach builds forecasts at the product-level and then sums them up to have category forecasts. The Top-Down Approach builds forecasts at the category level and then breaks them down to the product level based on expected assortment. The Middle-Out Approach is a hybrid approach that reconciles the forecasts.
Product Lifecycle Approach
The Product Lifecycle Approach buckets products into different stages, and then forecasts based on the product stage.
Account Based Forecasting
The Account Based Forecasting builds individual forecasts for each customer, and typically includes prioritization based on the size of the customer.
Product Classification Approach
The Product Classification Approach buckets products into categories of importance or difficulty to forecast, and creates different approaches for each bucket.
At Granularity, we support a mix of forecasting approaches. Our expertise is on short- and medium-term, hybrid AI forecasting. We build massive amounts of data features based on qualitative data sources, such as social media and economic data, and pair it with classical quantitative data like sales and weather, in advanced AI models that make sense of this high-dimensional data.
Benefits of demand forecasting
The benefits from improving demand forecasts allows businesses to stay ahead of the curve and make informed decisions. Here are some of the primary benefits of demand forecasting, categorized by the time horizon:
Short-term forecasting benefits
- Streamlined inventory management: more accurate forecasting optimizes the inventory you hold in the fulfillment centers, avoiding stock outs while keeping costly overstock to a minimum, which is essential for small business owners.
- Improved production planning: while many companies rely on more medium- and long-term forecasts to plan production, the growing trend of reshoring manufacturing and additive manufacturing, not to mention direct-to-consumer channels, means a more accurate short-term forecast can greatly improve labour and production planning.
- More cash flow insight: short-term financial planning and modelling, including planned investments or financial decisions, can gain insight from a better picture of forecasted sales.
- Targeted marketing efforts: Understanding short-term demand fluctuations helps marketing teams identify trends, seasonal patterns, and customer preferences, allowing for more focused marketing plan, ad campaigns and new product promotions. This can lead to increased customer engagement and higher sales.
Medium to long-term forecasting benefits
- Strategic planning and decision-making: Medium and long-term demand forecasts provide valuable predictive analytics and insights for strategic planning, such as product development, resource allocation and expansion into new markets.
- Enhanced supply chain coordination: Accurate medium and long-term demand forecasts allow businesses to collaborate more effectively with their suppliers, ensuring timely delivery of raw materials and finished goods across all supply chains and distribution centers.
- Informed pricing strategies: Understanding how demand is likely to change over the medium and long term can help businesses adjust their pricing strategies accordingly, leading to increased retail sales, improved in store customer experience, and higher profit margins.
- Risk mitigation: By anticipating fluctuations in demand over the medium and long term, businesses can prepare for potential risks and challenges, such as economic downturns, changing consumer preferences, or supply chain disruptions.
AI in demand forecasting
Demand forecasting with AI is a rising trend, buoyed by the improvements in the past decade in machine learning and artificial intelligence. Look no further than the rise of ChatGPT and its associated massive AI open models at how transformative and disruptive the field is becoming. The benefit of AI demand forecasting lies in three key improvements: dimensionality, scalability and learning.
Before we get into the details of AI demand forecasting, a quick disclaimer that in this section we refer to artificial intelligence as equal to machine learning, rather than what is called artificial general intelligence (AGI). This means we’re not talking about scary sentient machines, but simply incredibly powerful algorithms that can reason not like a human, but faster and at a scale a regular human cannot.
Dimensionality in Demand Forecasting
In terms of dimensionality, AI demand forecasting can reason with an incredible number of features. Some machine learning models, such as deep learning neural networks, perform better with dozens, hundreds or thousands of variables (aka “features”). This is a huge enhancement to univariate statistical models, where the only variable is prior sales.
In practice, dimensionality means you can go past normal variables like prior sales or even interesting variables like the temperature or Google searches. You can include word embeddings: the secret art of ChatGPT models that breaks human words and phrases into discrete number sequences. These embeddings can be quite long. For example, a popular ChatGPT model embedding for the phrase “demand forecasting” is an array of 1536 numbers.
Scalability in Demand Forecasting
For a field such as demand forecasting, where a point of sale system may have millions of records each month, scale is an unparalleled advantage of AI demand forecasting. AI demand forecasting can run smoothly, even when dealing with terabytes or petabytes of data. The growth of big data collection and the ability of cloud computing to handle these workloads means that AI can ingest, understand, and predict better on these massive workloads. As one demand planner put it to us recently – it would otherwise have to be broken into thousands of excel workbooks.
Feedback Loop in Demand Forecasting
The feedback loop in demand forecasting refers to the process of continually updating and adjusting forecast models based on the actual sales and customer behavior data. This loop helps to improve the accuracy of the forecasts over period of time, as the model becomes more attuned to changes in the market. Through the integration of AI and feedback loop, retailers can build more agile demand forecasting systems that are capable of adapting to changes in the market. AI plays a critical role in this feedback loop, by automatically processing large amounts of available data and identifying patterns and trends that might not be immediately apparent.
AI demand forecasting benefits greatly from the dimensionality, scale and learning available in demand-related datasets. In many ways, demand forecasting is a perfect field for the area of artificial intelligence.
Challenges in demand forecasting
As one CEO put it, demand forecasts are the underlying hypothesis that runs the entire business. Transforming that hypothesis can make massive differences. At Granularity, we’ve spoken to over a hundred demand planners from different companies and industries.
- Difficulty and Quality of Data Reporting: Time and effort required to string data together from disparate sources to have the reports needed to make decisions.
- Lack of Insights on Customer Trends & Being Too Reactive: Difficulty of staying ahead of customer needs in a rapidly evolving market, and the loss of margin from being too reactive.
- Heavy Workloads & Talent Gaps: Pressure of managing heavy workloads, high turnover rates, and re-training new talent.
- Building Consensus: Importance of cross-team collaboration among a wide range of stakeholders on forecasts and decisions.
- Lack of Visibility on Duration of Peaks: Challenge of predicting the duration of peaks to understand if a trend is long-term or an anomaly.
What is unconstrained demand?
Unconstrained demand is the maximum demand for a product in a given market without any external limitations, such as supply constraints.
At Granularity, we support forecasting based on unconstrained demand by providing market research and intelligence on end-consumer behavior. We provide data that estimates unconstrained demand based on indicators across the market.
What is Bullwhip effect?
The bullwhip effect is a phenomenon that occurs in supply chain management, where small changes in customer demand can result in significant fluctuations in demand further up the supply chain. The term “bullwhip effect” was coined because the effect resembles the way a flick of a bullwhip can cause a large, travelling wave.
The prominence of the bullwhip effect has increased since COVID19 and the presence of social media trends that cause spikes in demand changes. Managing the bullwhip effect requires monitoring market indicators to catch trends as early as possible.
The difference between demand planning software and demand forecasting software
The software for demand forecasting is employed to anticipate future demand, whereas the software for demand planning is utilized to create a strategy for fulfilling that demand. These two softwares are used together to enhance decision-making regarding inventory and production.
Demand forecasting software is used to predict future customer demand for a product based on historical sales data, market research and trends, along with other relevant factors. The goal of demand forecasting software is to provide accurate estimates of future demand, which can help companies make better decisions about inventory levels, production schedules, and supply chain management.
Demand planning software takes the demand forecast generated by demand forecasting software and uses it to create a plan for meeting that demand. This includes determining the appropriate inventory levels, scheduling production runs, and making other supply chain decisions to ensure that the company can meet expected demand while minimizing inventory costs and other expenses.
About the author
Tali Remennik is a CEO & Co-Founder at Granularity and a data scientist with 5+ years of experience building and deploying machine learning use cases across North America. Seeing first-hand the challenges with demand forecasting in a time of fluctuating consumer demand, she set out to see how market data could help. Granularity is an AI-powered demand forecasting platform that catches trends as they happen.