According to a report by McKinsey Global Fashion Index, the fashion industry will observe a growth of 3.5 to 4.5% in 2019 itself. A major part of this growth will be fueled by fast fashion, which is basically what the likes of Zara and H&M deal in. This trend in fashion is about getting the right stock at the right store at the right price, thus bringing in agility and value to fashion retailers, and their customers.
However, there are various challenges that the fashion retailers face, of which predicting the demand for the fast fashion product is a major one.
How will you know whether the product you are planning to launch this season will cater to the audience’s needs?
How to allocate the stock to the different stores, as the audience preferences and needs may vary?
How to price the apparel or assortment right, and ensure discounts are introduced for the right products?
Demand forecasting is a must if you want to plan way ahead of the season, and improve the operational efficiency of your business.
The few challenges that demand prediction helps overcome:
Overstocking: When you understand the exact demand for a particular product, you will stock them accordingly. This way, you will not end up spending too much on a product that has low demand, and increase the retail cost
Understocking: You would have had to mention “out of stock” for certain products during peak sale period. This not only causes user attrition, but also induces friction in the user experience. It can also lower your sales. With forecasting, you will know whether the customer is likely to purchase the product or not, and you will be able to manage your inventory wisely
Obsolescence: Fashion has the shortest shelf life. If you don’t sell it during the peak season, you might find that product in your shelf, even after the fashion is obsolete. These products will not fetch you profits; instead they will keep adding to the overall cost, and inventory. You will have to get them out at a discounted price, which will eventually reduce your earning margin.
Demand forecasting ensures a quick sell of the product, and reduces the cost to company while increasing the profit margins. It is a must for the fashion industry.
However, predicting demand for this industry can be quite challenging.
Fashion is perishable, which means the products that are in fashion today, may or may not be trending tomorrow. Certain historical data such as how products in this range did during the particular season would help you predict the demand for the new fashion. However, you will also need to consider geopolitical factors as well as the other external dominating elements, while predicting the demand, which poses as a challenge
It is important for the fashion retailer to offer an assortment, catering to every consumer. It is difficult to predict which type of product will get maximum demand, as tastes are fickle and fashion keeps changing fast. If the colour blue did well this year, it may or may not perform in the coming seasons
If your brand is associated with a particular segment of product, and you release a product from another segment, then forecasting the demand would be difficult. The reason being, there is no historical data to support the forecast. In this case, the demand can be assessed, and cannot be forecasted.
There are too many parameters to consider when you predict the demand for a fashion apparel. Festivals, promotions, marketing efforts, bundling, the events surrounding the launch of the fashion as well as cannibalization are few of these factors that you cannot ignore.
You will need to extrapolate the relationship between the products and the sales, while keeping these factors in mind.
Here are two different scenarios where forecasting can help
In case of the other kind of fashion demand predication, you are releasing a new assortment for the category that is already performing. You will need to take into consideration the attribute’s performance in the history, and how the user will take to the new assortment.
AI models learn from the history. They study the past performances of the product, their attributes. They even consider what worked and what didn’t for the particular product to produce the forecasting results.
There are several models that work in demand forecasting. You can take the “one size fits all” approach here, especially when it comes to granularity at the store/channel or at the style level.
One of the many ways in which you can approach it is by assembling the models through an ensemble algorithm which optimizes the nuances of forecasting that applies to the level of granularity required, at a specific snapshot in time.
You may measure the performance of your fashion apparels after the implementation of AI models.
Rate of Sale: The total quantity of the particular fashion apparel sold during the time period
Average Basket Size: The total order amount and the quantity of the products in the cart helps calculate the average basket size. It also helps know the quantity sold, and eventual profits earned
Full-price sell-through: The proportion of styles sold in full price as opposed to selling at discounts, over the available stock. This number can be analyzed at various levels like style-wise, store-wise or overall, and signifies how much consumers are willing to pay for the brand/style.
These are important metrics if you want to know how the particular fashion product performed in the market. The businesses can easily evaluate the performance and ensure profits, if they have predicted the demand with greater accuracy.
Demand forecasting is a challenge, as fashion is not just a fast moving, but also a fickle industry. Change in preferences, and seasonal demands need to be considered, along with the external factors, before predicting the demand.
However, with AI models, you will be able to allocate the right stock at the right store at the right time and price, thus maximizing your absolute profits.
The factors will be different for a new category versus the factors for a new product in the same category. In the latter case, you will need to consider the attributes and the assortment while, in the former you will be considering brand association, marketing and competition along with the experimental week’s data to help predict the demand
There are different approaches that you can take to complete the demand prediction. If you are considering a fashion product that is time-based, then LSTM method should help
In some cases, you can feed certain scenarios, which eventually helps predict the demand for the product.
Published on Oct. 8, 2019