Forecasting for basic items, such as a white T-shirt, is relatively easy, as forecasts can be based on the sales history of similar items. However, consider forecasting for a new fashion item, such as a floral-printed neon dress. That’s when things get more complicated. Fashion items have short life cycles, long lead times and no historical data to draw upon. 

Rapidly changing customer preferences, new competition, macro influences and ‘see now, buy now’ trends make it incredibly hard to predict demand accurately in the long run. That’s why judging how many units a fashion retailer will need to order from the supplier becomes more like guesswork. If you guess wrong, you will either run out of inventory, which is a deal-breaker for many consumers, or stock too much inventory that will need to be marked down later. To our knowledge, there isn’t ‘one right way’ to accurately forecast demand for new items in fashion. But these days, data is plentiful, and retailers are applying different approaches.

Retailers have traditionally made these decisions manually, which can be time-consuming and often lead to poor results. This blog post will discuss how AI can help retailers make better decisions faster and improve their bottom line!

Here are six commonly used demand planning techniques

1) Relying on designers, buyers and merchandisers’ opinion

Despite all the developments in AI-based demand forecasting, many fashion retailers still use a gut-based approach and trust their buyers, merchandisers and designers to make pre-season forecasts.

Merchandisers read the market, buyers visit production and design houses and designers use their observations of what people will buy. In this method, long-term forecasts are limited by intuition. It is more of an art and creativity-based method than an accurate science-based approach.

Besides, every designer or buyer can work on a narrow segment of merchandise. For example, one can work on scarves, whereas the other can work on crop tops. Therefore, using this method alone, fashion retailers can’t accurately foresee effects such as cannibalization or product substitution. 

 optimization involves deciding what products to sell at a discount and when. It can be complex, as several factors determine the optimal time and depth to mark something down.

2) Finding similar items in the past and projecting from there

Fashion retailers might have similar products that are close enough to make comparisons. Think of a retailer who wants to forecast demand for a ‘never-out-of-stock product’ like a black dress for the next season.

Typically, the retailer can access historical data on existing or previously sold black dresses for the past few years. Looking at previous years’ data can help forecast demand at sufficient levels for existing black dresses. However, they can’t be 100% efficient in predicting demand for a new item. Because of the fast-changing nature of the fashion industry, it’s pretty impossible to fulfill the demand of tomorrow’s consumers if forecasts are based solely on yesterday’s data of similar products.

3) Working with a trend forecasting agency

Unlike other retail industries, fashion is heavily trend-driven. Fashion retailers can work with data-driven trend forecasting companies that offer predictive AI-driven insights on upcoming trends and products.

Using trend forecasting to predict fashion direction and analyzing social media and fashion week trends can help with demand planning. However, this method will likely work for short-term forecasting, as trends can change overnight. The question remains: Is this data applicable and reliable for forecasting items in the market six months or a year from now?

4) Using product attribute and image processing data

As fashion retailers don’t have detailed information about the products for the next season, they try to take advantage of image processing and attribution data to find similar items.
They use the features, design attributes, such as patterns, color, fabric, materials and so on, and visual similarities of these items to generate long-term forecasts using machine learning models.

5) Getting customers’ insights on upcoming fashion trends

Today, retailers meet their customers through various channels, survey them to understand their preferences better and use the input to identify the trends they’re likely to follow in the upcoming season. They then use this data for the next season's planning.

It’s always good to use customers’ opinions, analyze where they are inspired and use this data to predict trends. This method was efficient in the pre-AI and analytics revolution era. However, today, future demand cannot be modeled purely based on customer preferences and predictions alone.

6) Combining all the methods into one holistic approach

Next season forecasting has long been an art form, but with the growth of AI-powered advanced prescriptive, predictive, diagnostic and descriptive analytics, it now becomes more of a science. Although forecasting continues to be a complex task with a track record of high error percentages, one thing is clear: Fashion retailers must continuously look for new ways and methods to become more efficient and agile in forecasting demand for the next season.

Better pre-season retail demand forecasting, better planning