AI-Powered Forecasting for Seasonal Demand

Artificial intelligence is being adopted in fashion retail faster than almost any other consumer industry. Despite this, discarded product costs the sector $500 billion every year, and has unparalleled financial and environmental implications across the supply chain. 

This two-part series focuses on how the future application of AI will be central to enterprise success in fashion by enabling retailers to more effectively calibrate their supply to demand. 

Multivariate forecasting should be a crucial part of their supply chain strategy. In this first article, we discuss how our models enable sales and operations planning (S&OP) to reduce financial loss from overstock and prevent missed opportunities through stockouts – not just in peak seasons – but throughout the annual sales cycle. 

Another year, a similar story 

The average clothing retailer will make 30% of their annual revenue over Christmas while also seeing 30% of their seasonal stock go unsold. After two winters of COVID-19 restrictions and a further two overshadowed by inflationary pressures, December and January will be a financial inflection point for the sector. 

This year, cost-of-living challenges made consumer behavior as difficult to predict and will add to overstock across many retailers' product lines.

In a best-case scenario, excess inventory will be sold in post-holiday sales at a discounted price, after taking up valuable space in warehouses and stockrooms. Some items will stay in ‘deep storage’ and be remarketed next winter. 

At worst, inventory will go to waste – and join the 1 in 12 clothing products that do so annually. Overstock is not a problem exclusive to Christmas; it reflects a year-round struggle fashion retailers face in effectively matching supply to demand.

Global clothing sales doubled between 2000 and 2015 – but waste from inaccurate sales planning and execution in wider retail still runs at over $500 billion annually. In recent years, the fashion sector has provided consumers with a breadth of product choices, distribution channels, and purchasing options. However, this vast quantity of waste only adds to the carbon footprint of an industry that has environmental implications at every tier of the supply chain. 

‍

‍

At an enterprise level, understocking has a similar impact on bottom lines. Research indicates that missed sales cost North American retailers alone $130 billion every year.

‘One big gambling machine’

In fashion, dynamic purchasing lies at the heart of this disconnect. Today, consumer preferences change at breakneck speed. What people buy in December is radically different from what was in style in the summer – when retailers placed their purchase orders. 

With it taking well over six months for a product to go from a design concept to being available for order, larger online outlets renew their inventory between three and six times a year to meet fast-evolving demand. 

But even with the most connected procurement strategies, decision-makers are often blind to what shape and form their consumer demand will take. 

It's not enough to say: “Product A sold X number of units in Q3, so we expect to sell Y number of units in Q4.” External variables ranging from weather to popular culture significantly reduce the accuracy of these conventional approaches to sales and operations planning. 

Industry leaders even describe the sector as “one big gambling machine” – where demand forecasts are often tied together using past sales orders, market analysis, and well-placed human predictions. 

Leveraging AI in fashion retail 

It’s no surprise that fashion is turning to artificial intelligence faster than almost any other consumer industry. An already sizable solution market, valued at $5 billion today, will be worth $31 billion by 2028. From scraping social media for consumer sentiment to color and pattern analysis models, AI-powered solutions are being developed for every corner of the market. 

Yet accurately calibrating supply to demand – and addressing this $500B problem space – is still a challenging task. In fashion, demand forecasting poses new and unchartered challenges for machine learning, and many retailers remain hesitant about its integration, application, and adoption when it comes to their supply chain operations.

Most supply chain forecasting solutions will predict demand using historical sales data through time series and regression analysis. In retail, this almost immediately disadvantages smaller and younger players in the market, who lack year-on-year data for their proven market demand. 

Models do not start producing accurate forecasts overnight, they take time to learn and train from enterprise data. This delays time to value in a fast-evolving industry, where when it comes to new sales and distribution channels, speed of adoption can either transform a retailer’s position in the market or prove fatal.

Fashion outlets, like most logistics-based organizations, also grow organically. Each department often has its own data cataloging software and practices. While some use the latest ERP solutions, others will be more reliant on spreadsheets and rule-based methods. 

These separate and often siloed systems mean that even the largest organizations lack the supporting infrastructure to bring together disparate, multimodal datasets and effectively apply machine learning.

Rethinking retail supply and demand 

At Pendulum, we build artificial intelligence products for fashion companies to address these very supply chain planning and execution challenges. Our software takes a holistic view of supply and demand so that retailers can effectively integrate their sales and operations planning with their end-to-end procurement, distribution, and marketing strategies. 

‍Predict\Products – our AI-driven solution for intelligent forecasting – empowers retailers with more accurate predictions tailored to specific enterprise objectives.

It can be applied across demand, inventory, and sales ecosystems to better forecast required inventory levels and inform downstream product distribution and allocation strategies.

A multivariate approach to forecasting

Predict\Products enables retailers to go beyond using just individual products’ past sales data to predict future demand. Instead of analyzing individual unit sales in isolation, it integrates historical data from across multiple product lines, locations, and sales channels, to create far more granular forecasts for any given SKU. 

This multivariate approach is designed to reduce overstock and understock in industries where inventory is defined by a high number of product variables. The size curve has long been an ‘Achilles heel’ for fashion retailers – where accurately predicting demand for smaller or larger-sized items can lead to both excess product and missed sales.  

But it doesn’t matter whether it is size, color, design, or fit – Predict\People can detect patterns across all of these variables and then create forecasts at an individual SKU level. This significantly outperforms the standard distribution curve many retailers rely on and empowers them to make purchase orders that more closely reflect end demand. 

Multimodal data for better insights 

Whatever their existing data ecosystem looks like, Pendulum’s software is designed to meet retailers where they are. Predict\Products integrates and leverages multimodal data sources and ‘glues together’ their complex digital landscapes. 

In a North American study by IHL Group  – an estimated 60% of financial losses were attributed to disconnected data systems. Whether it is contained in tables, text, graphs, or even images, Predict\Products is designed to learn not only from more data but from the data sources most relevant to creating accurate forecasts for retail enterprise objectives. 

Disconnected operating systems are only one facet of this data challenge. Mergers, system overhauls, and improper cataloging mean that many retailers work in ‘data-scarce’ environments, where out-of-the-box models simply do not have enough of the right enterprise data to learn from.  

Where that historical data is incomplete, inaccurate, or even unavailable in any format, Predict\People can supplement it with publicly available data, third-party sources, and secondary indicators.

Pendulum has ten years of experience creating forecasts across supply chain problem spaces defined by the absence of complete data. This means Predict\Products has benefited from a decade of human-in-the-loop interventions specific to optimizing outcomes around overstock and stockouts. It uses these learnings to provide high-accuracy predictions from complex and often lower-quality data landscapes. 

Today, Pendulum’s predictions perform 39% better on average when compared with the forecasting methods used by customers before engagement.

Complex predictions, all through an API 

The success of AI and machine learning solutions is not only measured by their accuracy – but by their adoption. Forecasts are only impactful when they are understood and effectively leveraged by employees. 

Predict\Products can integrate with the ERP, TMS, and WMS platforms most familiar to procurement, distribution, and sales teams, reducing any friction surrounding change management or uptake.

Rather than requiring employees to train with and use new dashboards and interfaces, it provides these forecasts within the operating systems they are most familiar with – all through a simple API.

Turning insights into actions 

Multivariate forecasts can empower supply chain and logistics decision-makers with the insights they need to make more informed procurement strategies – not only for peak sales periods but to drive year-round profitability. 

But product forecasting is only one side of the retail supply and demand challenge. How outlets can optimize existing inventory – within what can often be short windows of relative demand – is also crucial to maximizing sales, effectively leveraging stock-on-hand, and reducing product waste.

The second installment in this series will look at how artificial intelligence can empower retailers with machine-learning allocation recommendations across their distribution channels and how they can put these forecasts into action with Plan\Products. 

If you’d like to learn more about Predict\Products or Pendulum, please send an email to predictproducts@pendulum.global

About Pendulum

Pendulum is an AI company that optimizes critical supply and demand networks. We make ubiquitous systems more intelligent, maximizing the impact of resources available. Pendulum products are deployed via APIs that navigate, predict demand, and optimize supply  – continuously and autonomously improving on their own.‍‍