Data Science in Fashion
Fashion industry is an extremely competitive and dynamic market. Trends and styles change with the blink of an eye. Data Science can be used here on historical data to predict the trends which will be “Hot” hence potentially saving a lot of time and money.
By Preet Gandhi, NYU Center for Data Science
How many stores that you grew up visiting aren’t around anymore? Remember Sears, Zellers, American Apparel, Wet Seal, The Limited, etc. Many other prominent retailers like Aeropostale, Bebe, A&F, Guess, J.C Penney, Payless, Rue2, etc. are closing hundreds of stores nationwide to deal with dwindling sales. Forbes estimates that in last sector of 2017, 21 retailers are closing 3,591 stores. They can’t compete with the e-commerce sector as more and more customers resort to shopping online as reduced prices from the comfort of their homes in lieu of physically going to retail stores. Retailers are forced to leverage data in order to upgrade their infrastructure and services to give their customers a better experience. Many job postings at Macy’s, Coach, Kate Spade, Nordstrom, etc. show how serious and desperate the retail and fashion giants are in face of competition.
Fashion industry is an extremely competitive and dynamic market. Trends and styles change with the blink of an eye. A collection or trend takes thousands of man-hours from the most creative minds and the lucrative outcome depends on the simple Hot or Not judgements from fashion pundits, bloggers and celebrities. Data Science can be used here on historical data to predict the trends which will be “Hot” hence potentially saving a lot of time and money. For example, training good models on previous sales data can help us predict whether Kanye West’s new Yeezy Season 6 collection will be a success or not. Data Scientists can use concepts from predictive algorithms, visual search, capturing structured data from photographs, natural language processing and many more.
Data is abundant in the fashion and retail industry. The vast historical data from retailers and department stores about the spending habits of customers is a traditional source. With the advent of social media, engagement on posts, Instagram trends, Twitter hashtags, clothing style of the most popular fashion bloggers, celebrity fashion styles, the “likes/reactions” on popular celebrities, etc. provide us a rich collection of insight-ridden data. A new technique to test the reaction before the deployment of the collection is to release photos on Social Media (Facebook, Twitter, Instagram, Pinterest) and study the comments to make changes to the collection before launching. Sentiment analysis is used here to get insights from the public opinion. The public API of these sites are open and easy to use. Data can be scrapped in real-time through these API’s and converted to usable form. Online data sources are raw and uncensored and voice the public opinion. This data has a lot of potential if harnessed properly. Most data you find online is unstructured : texts, images, audio and YouTube videos. Unstructured data can be a challenge to use in its native form and needs to be cleaned and transformed.
An interesting source of data are the wifi signals of the customers in a store. The pattern of the customers is tracked to see how long they stay, which sections they visit sequentially, how often they come back and how long they stay in each section. This kind of data can be useful to arrange the collection in stores and to place closely the items frequently purchased together.
Zara is one of the most popular and successful stores in the fashion world. They adopt the concept of “fast fashion”, where the whole process of designing a collection to shipping it to stores takes a maximum of three weeks. The success of this brand is attributed to this dynamic concept where the retailer studies the choices and preference of the customer to create a collection catering their tastes. They create what the customer craves instead of selling what they design. The customer themselves may not know what they are particularly looking for, but the smart business analysts and data scientists at Zara make use of the data to create a collection which the customers will automatically want because it’s their “taste”. Moreover Zara has stores all over the world and their customers have different demographics which means something as simple as size, body shape, colour preferences and quantity will vary greatly. Producing the right quantity of right products helps minimize wastage.
Key Problems Areas
1. Color options
Using Big data, we can find the colors preferred by the customers to curate a best-selling collection. The range of colors for a particular style, the combination of colors purchased together, etc. can be mined from sales data and online retail data. Many times customers purchase a piece of clothing in one color and then exchange it for another. The data from the returns/exchange can be used to create more items of the preferred color.
2. Men’s or Women’s clothing?
Each designer targets a different demographic or gender to increase their popularity or sales. Designers need to decide how much items in each collection and the kind of variety they need to create. They have a fixed set of resources like budget and display space and they need data-backed guidelines to decide how much to allot to each category. It is common in many stores like Forever21, H&M, etc. to see two/three floors of space given for women’s merchandise and only one for men’s. These retailers know to provide more options for a particular category of customers to increase their sales. These insights are derived from historic sales data.
3. Turning runway styles to retails merchandise
Many styles featured on the runway are not “wearable” in real-life. Trends on the runway are exaggerated and too over-the-top for retail. The outfits need to be altered before they can be curated for sale in stores. Training algorithms to suggest which features to change like color, fabric, cut, length, combination, etc. can ensure that the product sells well when they hit the racks. Moreover each country/region has a different taste. Hence each product must be tweaked to suit the local preferences.
4. Garment Price
For each garment, the designers need to understand the prices the customer would be willing to pay given the quality, style, popularity and the brand value. Big Data should be used to average previous sales data to generate suggested pricing. Data from competing brands can also be used to set prices which aren’t too high but still contribute to good revenues.
5. Uncovering new product categories
A brand needs to find new products that will be successful in the market and which product aren’t very lucratively promising. Designers need to think whether making a unique new product will be accepted or rejects by customers. For example a creative bright print may work for yoga pants but may be deemed too gaudy for sneakers. Big Data can be used to decide which category to venture into and whether to continue selling a particular previous product.
6. Store arrangement
Customers exhibit a particular behavior while shopping which can be studied to arrange merchandise in a manner which increase the chances of sales of majority of the pieces. Associative data mining can help us decide where to group products together so customer is likely to pick up most of them. You may have noticed that in many clothing stores, the accessories are placed when we stand near the billing area which causes the customers to pick them up. Wifi data can be used to track the customer movement in the stores to arrange the stock in an optimal manner.
Without a doubt we can say that data powered decisions will give you an edge in the competitive world of fashion. Before creating any product, data needs to be consulted to see it is economically feasible and promising. Selectively using your data to create and convert product lines your customers are sure to buy in the future would help the retails houses survive in the wake of e-commerce. Some may say that AI can dull the creativity of the collections by just creating what the customer want. But that is why the outcomes should be used only to supplement the human creative insight instead of entirely replacing it. But it doesn’t hurt to create the right product at the right price at the right time.