Unraveling Demand Forecasting Challenges in Retail

Optimizing inventory is critical to retail success, ideally moving up and down in line with demand. The challenge is the variability of factors that influence the market, from the more predictable, such as seasonality and promotions, to the spontaneous, such as weather changes and global issues. While we can estimate needs based on averages and historical data, accurate forecasts are difficult to achieve. 

Why Is It Hard to Predict Demand?

When a planner indicates that an item sells 213 units per day, they likely refer to ADU (average daily usage), emphasizing the term "average." In reality, daily sales over the past week or month may not have included a single day with precisely 213 units sold; instead, it could be a range like 199, 220, 233, 187, 210, 228, etc. This variability is why we use averages. However, if we were to order precisely the average 213 pieces, what would happen? The Gaussian distribution curve provides insights into this question, familiar to those versed in probability and statistics.

[Read more: "How Target Uses AI to Deliver Superior Customer, Associate Experience"]

The average of 213 units establishes the chart's peak, representing a 50% probability. In other words, actual demand is equally likely to be below or above this figure. Accepting a 50% probability of shortages is impractical in retail, affecting sales and profits. Hence, the logical inclination is to secure additional goods for reassurance.

The paradox arises in aiming for a high level of availability, necessitating orders significantly above the average. Crossing the 95% probability boundary requires a substantial increase in stock for each subsequent percentage.

Forecasting complexities in retail stem from natural distribution curves having shapes more intricate than the ideal Gaussian curve. Even for fast-moving goods, these curves are asymmetric, and less frequently sold items deviate further from the perfect shape.

For an item averaging 0.5 units per day, fluctuating between zero, one, two or three units, ensuring 98% availability demands ordering considerably more than the average sales. The pursuit of high availability, however, can lead to excessive overstock, potentially reducing sales and resulting in various economic repercussions.

Nevertheless, in the realm of retail, analyzing distribution curves for inventory planning is a rarity. The challenge lies in the individuality of these curves, not only varying across SKUs but also within the same SKU across different locations. For instance, managing 1,000,000 curves for a retail chain with 10,000 SKUs and 100 stores is a time-intensive endeavor. Consequently, this task is often tackled using Excel, with specialists relying on their unique formulas and methods. Even if an ERP (enterprise resource planning) system handles the calculations, its algorithms are typically straightforward. The prevalent tendency is toward order increases to prevent sales loss. As a result, some retailers may not meticulously oversee their inventory economics, often neglecting this task until it becomes a critical issue.

The Cost of Forecast Inaccuracy

In the initial phase of a company's growth, the sales department typically serves as the primary driver, and the demand-planning model often follows a familiar pattern: average sales plus adjustments for seasonality, trends, promotional campaigns, and a safety buffer to ensure robust availability and, consequently, high sales.

[Read more: "How Price Chopper/Market 32 Leverages AI to Combat Food Waste"]

While this strategy can achieve availability rates of 97% or even 98%, it often results in excessive overstock, leading to elevated logistical and operational costs. This surplus inventory not only occupies valuable shelf and warehouse space, but also adversely affects overall sales. As items designated for sale take precedence, customers may bypass other, potentially more profitable, goods. Although turnover may increase, profit margins decline. Additionally, products risk reaching the end of their shelf life, necessitating write-offs and incurring additional costs.

For nonfood products, the surplus may be shuffled between retail outlets and warehouses, eventually sold at significant discounts or written off. This process contributes to rising logistics expenses and further diminishes valuable shelf space for more marketable goods. As storage capacity becomes strained, companies are compelled to initiate expansion projects.

While the formula for calculating sales losses is straightforward, determining the comprehensive costs and losses tied to excess inventory poses a more significant challenge due to their diverse and not always apparent nature. The common belief that "we'll eventually sell this product" may hold true for isolated cases or specific SKUs but proves inaccurate when applied across a company grappling with numerous overstocks.

What Does This Have to Do With AI?

The sheer volume of data to analyze surpasses the capabilities of employee experience, intuition, and simplistic Excel or ERP models. Striking a balance between the perspectives of "salespeople" and "finance people" in this age-old battle requires a robust methodology and sophisticated mathematical models.

Fortunately, AI has emerged as a powerful equalizer in this arena. AI algorithms impartially perform calculations and tactical decisions based on complex models, aiding managers in crucial strategic decisions by discerning clues and patterns within vast datasets that may elude human perception. Thus, modern companies are no longer wondering whether to implement AI for demand forecasting. The only question is the speed of decision-making and the choice of appropriate solutions. Undoubtedly, the market competition will be won by those retailers that will jump on this departing train the fastest.