The fashion industry is known for its notorious high waste production. Retailers attract consumers through frequent collections with shorter life cycles and wider assortments. With today’s headlines on the increase of dead inventory in the industry, sustaining this strategy poses a major risk to the existence of many retailers.
In one of my previous articles, “Rethinking Traditional Formats to Stay Relevant in Today’s Post-Covid Modern Retail Environment”, I discussed the post covid-19 global challenges to improve full-price sell-through. As of 2023, it is estimated that around 30% of all clothes made worldwide are never sold (Fashion United). This benchmark is very bad news for our planet and terrible news for retail managers that contemplate tactics, given that the next two seasons were already planned. More discounts will put the company’s short-term profitability at risk but also risk long-term brand positioning.
The puzzling question is: why do retailers go so wrong in their plans when the technology available to them is so advanced? Today’s ready-to-use AI-based prediction models can use long-term historical data and predict variables like market sentiments and fashion trends which used to be limited to the opinion of experts. The somewhat disappointing reality is, that while AI-based planning has become so prominent today, it didn’t change much the benchmark of in-season sell-through, nor the unsold inventory share.
In this article we will use common sense to challenge the convention that retailers need better season planning to finish seasons without unsold inventory. We will describe the inherent negative spiral of the fashion industry and use a cause-and-effect approach to find its root cause. Eventually, we will illustrate the in-season methods and AI tools that have the power to considerably mitigate the dead inventory catch. On a personal note, I truly hope that this article will serve as some relief for central merchandisers, planners and supply chain directors that are operating in over-stressed environments and are unjustly held accountable for poor season results.
Why is Fashion Retail in a negative spiral?
Let us assume a retail environment with no noise factors. New inventory is delivered to the distribution center on time and in full. Consumers visit the stores, browse through the assortment of new products, and buy according to their preferences. We can call the products that are sold well fast movers and those that do not sell well slow movers. We, of course, want to produce only fast movers, but the reality is that fast movers and slow movers are distributed according to the 20-80 rule when a collection is composed of multiple styles, colors, and sizes. Much of the collection will turn out to be slow movers, while only a few products will turn out to be fast movers. Since the fast movers naturally sell faster than planned, they run out of stock for replenishment fast, and their share in the store decreases over time. With more collections coming in, the shelf is occupied by more slow movers, and less fast movers, and less space is left available to display new products – consumers’ chance to find the product that they desire in their size reduces sharply. Luckily for retailers, the End of Season Sale period is the time for the store to liquidate unsold inventory through deep discounts and start afresh.
Over a decade of analyzing fashion retail data, we learned that over 50% of the stock on a shelf before the end of the season belongs to SKUs that contributed to merely 5% of the seasonal sales. Different criteria to identify slow movers will not change the distribution much, eventually, retailers were more wrong than right! Were the designers wrong? Or perhaps the planner? Thankfully, we have unlimited access to data today to know that this outcome had little to do with both. If we mapped inventory between different stores, we find that slow movers in one store can be categorized fast movers across numerous other stores. Fast movers and slow movers are mostly store-specific definitions. Tragically, the same fast movers that ran out of stock early in the season and led to loss of sales in one store, end up also losing margin in other stores when liquidated at end of season’s discounts. A simple cross-store analysis will reveal that stores had the capacity to sell more than 50% of the slow-moving inventory during the season if they had it at the right time and in the right place.
What is the root cause, then?
We can safely assume now that the plan was approximately right, or at least not precisely wrong. Our search for root cause narrows down on in-season execution.
The problem starts as early as retailers allocate new collections to stores. Statistics regarding sales are not yet available, and we cannot use our intuition either when it comes to predicting the sale rate of a specific style-color-size in a specific store. A common practice of retailers is to set size-curve norms to categories in stores. For example, if a pivotal size of a shirt sells two pcs a week on average in a high-street store, retailers can assume two pcs as an initial norm for the pivotal sizes of the same category across high-street stores. But we understand now that this average is a fallacy. It is an average of the few super-fast movers that can sell even ten pieces a week and many slow movers that did not sell any. Starting with over-inflated inventory in stores generates a high inventory of stores’ slow movers and restricts the inventory remaining at the warehouse for replenishment of the stores’ fast movers.
Knowing that the allocation plan is not rocket science, retailers attempt to adjust SKU norms based on actual demand in the season. The common approach is conventional forecast algorithms like rolling forecasts and dynamic MIN-MAX targets. The unfortunate mistake here is that the collection life can be as short as 4-6 weeks, and the average sale rate of a typical product is as low as 1-2 pcs in 2 weeks. In such a reality, these algorithms will either fail to respond to demand trends or constantly respond to noise. In simple words, by the time we get sufficient statistics for an accurate forecast, the season is over, and most inventory is misplaced in the wrong stores.
How can we use AI to break out of the industry’s negative spiral?
I’ve worked with data scientists over the years in search for the right mix of technology that will allow retailers to considerably improve their in-season operations through automation. We narrowed down on three areas in which AI led to breakthroughs:
- Behavioral Clustering: A common AI technique that identifies and divides unlabeled data into different groups such that similar data points. If applied to SKU-Location level demand, this approach has the power to enable an effective transition from conventional forecasting into a fast-response Short-Term Prediction. The algorithm clusters products in real-time based on their identified demand pattern and adapts their target accordingly.
- Products Similarity: Learns the impact of product attributes and various combinations of product attributes on the demand for products. Well-known methods like “Features-Embedding” translate the semantic information in the data into a numerical form that the algorithms can understand and use to find which products are similar to each other. This capability leads to a far more intelligent allocation of new products based on learning gained from the behavior of similar products in previous seasons.
- Extreme Demand Prediction: Uses past seasons’ data to predict the potential demand changes and their impact on store inventory during special events or price-offs. While this topic was not discussed in this writing, it is another area that can fuel inventory mismatch between stores. The rapid increase in traffic calls for a proactive increase in inventory in stores to protect sales. Without the tools to guide retailers on which products are likely to absorb the demand increase, the common human approach is to error on the safe side and proportionally increase the inventory of all products unnecessarily.
Retailers that implemented these tactics demonstrated a 10%-20% increase in in-season sell-through, enjoying higher margins with less inventory in stores and higher freshness in the season. Finally, establishing more scalable and predictable operations where Merchandisers and Planners operate in harmony and extract more joy from their work.
