"Big data enables intelligent management of the fashion stores supply chain"
2/4 - How can artificial intelligence be useful to fashion? The answers, supported by concrete examples, in this second part of the interview of Romain Chaumais, executive leader of the French company Fashion Data, specializing in AI applied to fashion.
By Ludmilla Intravaia
Le Boudoir Numérique: You mentioned in the first part of our discussion (read here), that eco-profitability must be at the heart of the concerns of fashion brands. Does forecasting fashion trends by artificial intelligence, one of the solutions offered to companies by your start-up, promote this eco-profitability?
Romain Chaumais, executive leader of Fashion Data: The analysis of the web and social networks by computers allows us to infer how fashion trends are evolving. It can be relative to a cut, a material, a color, a type of product, a way of wearing it, for example if the black skirt is back, if it is above the knee, below, if it is pleated or straight, etc. Analysis curves show whether one trend is going down and another is going up. A trend that is starting to fall but is still relatively important should not be forgotten. You need to have the items that can meet this need, even if they will target from now on a clientele that follows the trend rather than making it. But if a market remains for these items, fashionistas will no longer buy them. Therefore, when you will send them your next email, it should no longer be offering these products but new models. Note that predictive analytics doesn't tell you how many products you will sell next year, but whether you will sell more or sell less than last year. It gives the upward and downward trend in sales for the following years. One of our solutions, demand forecasting, specializes in predicting future sales, based on past sales data. Say you want to market a green V-neck sweater with black stripes. We are going to study all the similar products that you have sold in the past, which will give us a good indication of your ability to sell these kinds of items. We analyze past sales by correcting them against full potential sales, based on certain phenomena, such as out of stock. We can, for example, estimate that you could have sold 15% more coats, by identifying as the main obstacle a lack of sufficient quantities to sell. Or maybe you have sold 10,000 pieces but in the end we can determine that your real sales potential was rather 7,500 pieces because you had to make a lot of discounts to succeed in selling everything that you had made. By combining sales, corrected past sales and fashion trends, we can help a brand more precisely adjust the quantities to buy to produce, reduce unsold items and avoid stockouts. This is where we meet this eco-profitability objective of a brand that has not manufactured too much, has no waste and has sold well, without discounting, therefore is profitable.
Once the product is launched, can you follow its life cycle to adapt it to the circumstances of the moment?
As soon as an item is put in the store, real sales start and it's like the release of a movie. The first day of a movie or its first weekend in theaters, you see pretty quickly whether it's going to do great or not. It's the same thing with fashion. Three to four days of sales are enough to estimate whether the product is up to expectations or not. We can set up a monitoring during the season, which consists of detecting as quickly as possible whether you will have enough products, if you will have too many, if you will not have enough. If you have enough, it's fine, let's not discount. If you have too many, no need to make more, stop the orders and start reducing the price to speed up sales, so that we don't end up with a bunch of unsold items at the end of the season. If you don't have enough you may be able to trigger productions to be delivered to stores within a fortnight to keep up with demand, otherwise you will have to control scarcity, putting items in certain stores only or online, so that sales are made in the right places.
In terms of store supply, what can be done?
Stores do not need to have the same quantities and the same items at the same time because each one is different. One will sell more plus sizes and sportswear, the other more small sizes and casual items. Each store has its own features that we identify in our identity analysis, in order to understand how to better adapt the offer and adjust the stocks. The question is knowing the quantity for each reference, size and color to place in each store and thus have the right quantities in the right place. It goes without saying that retailers did not wait for us to optimize their stocks in stores. But it was done with rules that weren't big data, quite simple rules in fact, that could fit in an Excel spreadsheet. With AI and big data, we are able to bring much more subtlety and precision in the intelligent management of the fashion stores supply chain, for every quantity, for every store, every day and every item.
You also propose a solution to prioritize orders?
Indeed. If you have a container to fill in China with your items made there, knowing that there will be no next delivery for ten days for example, it is essential that you know the products to put first in this container. The goal of this order prioritization solution is to get your winter coats to arrive in stores when it's cold, but not too early. And therefore, before sending the coats, to make room in the container for the sweaters intended for milder weather. If it's the other way around, you risk missing out on your season. Calculating a priority level for each order gets products to the right places, at the right times, and at the best prices.
Can artificial intelligence slow down parcel returns?
The return of purchases is a very serious environmental problem, foor instance in Germany. In this country, 50% of fashion orders online are returned, because German consumers are used to being able to buy online, while having a period of three weeks for the return of goods, without penalty and free of shipping costs. The same item can be sold two or three times, before someone decides to keep it permanently. In terms of carbon footprint, it's a disaster. It comes back to knowing the customer, since it is about his consumption habits. If a customer has add three red sweaters, in three different sizes, in his shopping cart, it is just that he has doubts about his size. He will order three different sizes, so after trying them on, he will only keep one and he will return the other two sizes. This is where artificial intelligence can come in because it knows the size the customer needs, let’s say medium. It is then possible to simulate a shortage of the small and large sizes on the e-commerce website to send only the medium size. Or to adjust the transport costs, according to the profiles of these consumers taking advantage of the free return system, in order to discourage their behavior. Or even avoid this type of customers, by no longer sending them promotional emails. In fact, a customer who brings in 100 euros in purchases but costs 150, in materials and transport for the brand and the planet, is not eco-profitable.
Are you able to demonstrate to your customers, with supporting figures, the impact of your eco-profitability solutions?
Yes. Many brands still stick to their intuition and do not take the time to really quantify the impact of their actions on the evolution of their sales, their customers, etc. For our part, we impose ourselves to measure the benefits, in terms of additional margin and reduction of CO2 emissions, generated by our solutions. Quantifying results appears to us to be a desirable and necessary step because in order to improve, nothing better than knowing clearly where we are.
* Read the first part of Romain Chaumais' interview on Le Boudoir Numérique: "Smart data at the service of fashion eco-profitability".
* Fashion data website is here.
* Continue reading on Le Boudoir Numérique with these following papers:
- What will we wear post-lockdown ? The answer with Heuritech’s AI
- Covid-19 – Heuritech launches a solidarity program for fashion brands
- “Cross-fertilization between tech and fashion is the strength of Heuritech”
- “To help fashion brands decision-making with AI during the Covid-19 crisis”
- Will leopard print pleated skirts be trending any time soon ? AI already knows it !
- “AI can contribute to the virtuous circle of sustainable fashion”
- “Our algorithm is the link between the parfumer and the customer”