blog

A Guide to AI Solutions for Fashion E-commerce

In the last ten years, we’ve seen fashion brands steadily gravitate towards digitalization, automation, and technology innovation. From client-facing solutions to optimized processes behind the scenes, artificial intelligence (AI) has gradually trickled down into every industry layer. Those who succeed in redesigning their entire processes with technology and data intelligence at its core are the ones that will jump — and stay — ahead of the curve.

The AI journey may have started in the industry a few years back, but the COVID-19 pandemic accelerated the market’s need for better and faster online experiences. As brick and mortar stores closed their doors and physical commerce worldwide declined, retailers relied on their digital channels entirely. So, as revenue migrated to digital, they started to take AI seriously.

Still, the adoption of AI in the fashion industry is in its early stages. Katia Walsh, Chief Strategy and Artificial Intelligence Officer at Levi Strauss reflects:

quotes

This industry (fashion) is still quite analog, very manual and driven by creativity and intuition, but because of that, imprecise.

Katia joined Levi’s in 2020 on a mission to revolutionize and transform the brand into a data and artificial intelligence powered business and become a truly innovative digital apparel company.

Other leaders in the e-commerce space, such as The RealReal, the largest online marketplace for authenticated, resale luxury goods, consider that AI and data are key factors that enable them to scale and grow more efficiently and effectively.

In their Q4 2020 Investor’s Earnings Results, the brand announced that they were automating the pricing of 80% of unit volume, copywriting of 84% (including product title and description), and photo retouching of 85%. The company is also using AI and image recognition technologies to enhance its authentication efforts, a key differentiating factor of their offering.

But, what exactly is AI? How does it apply to fashion and resale companies? Read on as we break down the what, how, and why of digital innovation for fashion e-commerce.

AI for Fashion E-commerce

AI stands for artificial intelligence, and you'll often see it used to mean what we call Machine Learning. To put it in simple terms, Machine Learning is a technology that makes computers learn from patterns in data and consequently solves problems.

In a nutshell, AI allows us to:

  • Reduce costs and increase efficiency.
  • Make intelligent and consistent decisions.
  • Automate and scale operations.

First, AI relies on the collected data of customers when they interact with digital platforms. Engines can track relevant events like when a customer makes a purchase, interacts with a page, or shows interest in a product, and AI learns from these.

Because of its ability to discover relevant patterns in the data, AI is a great tool to make critical business decisions, such as calculating the right price for an item, recommending similar products, or forecasting future shopping trends.

At Tryolabs, we have worked with fashion retailers and fashion resale brands, developing solutions that tackle some of the most significant pain points in their e-commerce operations. We found that these AI-powered engines quickly became vital ingredients to their success and scalability.

In the process, it became clear: AI should be a first-class citizen in online retail operations.

Pricing Intelligence

How you price an item and scale pricing operations intelligently, consistently, and efficiently is critical for many e-commerce fashion brands. AI and data intelligence dramatically improve pricing decisions.

Following, we’ll go through some AI solutions we have created at Tryolabs to improve pricing in fashion and resale e-commerce.

Price Automation

An effective pricing strategy is especially relevant in the booming fashion resale industry, where every pre-loved item is unique and scaling pricing operations becomes challenging.

In resale e-commerce, every item is potentially unique (single SKU). Brands have to learn how to price these unique items without direct access to information such as production cost or even how much these items are worth new. Even though external information can be accessed, it can be costly to do so for every single item published in a resale marketplace.

Usually, brands hire people with fashion knowledge and task them to price the items based on their available information, plus their market value estimation. This manual process sometimes leads to inconsistent results, as different people value the same item differently. The human bottleneck is also a factor: every piece published on a site needs to be priced by a human, impacting costs and publication speed. It is also difficult for a person to consider the various factors influencing demand and price.

At Tryolabs, we have extensively worked in Machine Learning models that automatically and consistently price every new item by looking at data. The machine leverages image recognition technology to understand the look of the product and consider other aspects such as brand, condition, fabric, category, description, historical sales data, and past pricing decisions of similar items. With these data points, it computes an optimal price for each SKU. It is also possible to expand the model to include other features, including competitor price if it’s available.

Some of our largest customers have used such models to automate the vast majority of their pricing operations and scale to more items effortlessly.

Price Recommendation

Sometimes, it’s not just about automating internal company operations. For some peer-to-peer resale marketplaces, such as Depop or Thrilling, the consignor or garment owner ultimately defines the resale price. Therefore, even if you are certain of the right price for the item they publish, it’s ultimately your job to convince them that the price you set is indeed fair.

For these cases, a price recommendation model can be very useful.

In this model, an intelligent algorithm can automatically determine a suggested price point and range, and give hints to the seller as to why the price is fair. This is done by showing them similar items (category, designer, style) sold at similar price points. The goal is to give the seller confidence in the price recommended by the AI. The seller gets information that relates the item they are selling to the fair price suggested by the AI. With this, they are more likely to accept the automated price.

Suggested price and range for a dress, with reference products.

Price Optimization

Traditionally in the fashion industry, pricing strategy is based on an initial fixed price, followed by a series of aggressive discounts as the season unfolds.

These discounts are often applied to groups of items (for example, specific categories) based on business rules. This strategy has the consequence that a product that has sold well throughout the season may be discounted, but in reality, it doesn't need the price reduction to sell. Equally, an underperforming item may benefit from a price reduction earlier than the estimated sales period, losing out on weeks of sales opportunities.

This results in an avoidable loss of profit for companies.

A more innovative approach is using price optimization. The goal is to understand how clients react to items at different price points and estimate the right price for each product if they want to sell it in a certain period, ultimately maximizing profit.

The machine forecasts demand and considers many factors to advise on dynamic price adjustments. Some examples of these factors include:

  • sales performance
  • pricing strategies of the competition
  • weather and seasonality
  • special events such as the holiday season or Black Friday
  • macroeconomic variables
  • inventory and warehouse information

Price optimization is a more complex integration than price automation or recommendation and a perfect second step once you have AI baked in some of the operations. Nonetheless, there is massive potential in this pricing model, and it is likely to become the norm in the future, especially for fashion e-commerce brands.

Get a deeper look into pricing optimization on our dedicated article or go directly to an e-commerce case study here.

Operations Automation

The fashion e-commerce business has many opportunities for improvement in its processes and operations with the help of AI. We'll explore solutions that can automate tedious, repetitive tasks, saving a lot of time (and money) while uncovering new opportunities and possibilities.

Tagging and Copywriting Automation

Suggested prices for different dresses.Images extracted from Farfetch.

Searching for items on a crowded site can be a daunting experience. The constant incorporation of new products and the rapid flow of e-commerce makes filters a browser's best friend. But before shoppers can filter through categories and find what they are looking for, somebody has to put in the hard work.

Fashion retail teams spend days writing descriptive copy for items and tagging them according to their primary attributes and preferred fashion taxonomy. Traditionally, this is a manual process that relies on each person's time and criteria, often resulting in a lack of consistency in data.

Considering that for a fashion expert there might be 14 different types of necklines, it is challenging for any single person to know and remember all the criteria to differentiate each one, let alone considering that this is just the tip of the iceberg of a more complex taxonomy. It’s also very difficult to keep the data and criteria consistent among other people.

With the traditional manual process, it is common to face some of these issues:

  • Inconsistent descriptions among similar items (some too long and detailed, some too short)
  • Missing attributes
  • Material descriptions that do not add up to 100%
  • Lack of human alignment in criteria for specific attributes or values

AI has the power to completely transform this task, freeing people from the repetitive and error-prone tasks necessary for product tagging. Not only can it automate and speed up the process, but it can also make it more consistent and reliable. This results in better data quality, which is critical to unlocking a multitude of data-driven business cases.

It can also improve the process by making it more comprehensive since AI can learn a more complex taxonomy and attributes.

For example, it can automatically define physical attributes (such as the silhouette, necklines, type of sleeves, length, patterns, color, etc.). Still, it can also learn other more subtle attributes such as "bohemian," "nautical," or "artsy." This data enhancement opens the windows for creating smarter search, recommendations, and personalization engines.

Tryolabs has developed an AI solution that uses computer vision technology to automate item tagging and descriptions, obtaining consistent and standardized data. The information comes directly from the image. The machine automatically fills product attributes and tags them accordingly to sit in the correct category and hierarchy on the website by simply uploading the photo. This can be tuned for each particular taxonomy unique for each fashion retailer.

Copywriting of coat.
Copywriting of top.
Copywriting of boot.
Results of copywriting automation using AI.

Image retouching

Images ensure online shoppers get as close as possible to fully understanding product features. Professional-looking photos can give an edge to an e-commerce site, making customers more likely to buy, and brands happier with the portrayal of their products.

Usually, pictures are taken in a photo studio and then sent out for retouching. The retouching process includes several different steps, depending on each particular site. The most common ones are calibrating colors for accurate depiction (you don’t want customers returning items because the colors don’t match their expectations!), removing backgrounds, aligning/rotating items, smoothing creases, and erasing mannequins. Unfortunately, this process can take several days, delaying the whole item uploading process.

But AI is increasingly used to cut corners and automate most of this time-consuming process.

Retouched image of a shoe.
Original image of a shoe.
Retouched image of a dress.
Original image of a dress.
Results of automated photo retouching using AI.

Tryolabs has worked on automated image retouching using AI that cuts both time and costs to a fraction. The system replicates the traditional “Photoshop pipeline” done by artists but on a much larger scale. The opportunity is huge: one of our largest clients reduced the turnaround time for retouching hundreds of thousands of images from three days to only 30 minutes.

Recommendation engines

Recommender systems have become a vital part of the users' experience on e-commerce sites.

These systems look for which products to recommend for each item, based on information from the product itself and how the users interact with it. For example, a recommendation engine can display the most similar items for each product, or those that can complete the look. For a baseline recommendation system, all users of an e-commerce site will see the same recommendations for each product.

A more sophisticated approach is personalization. This layer adds an extra to the mix, allowing brands to recommend not only based on set criteria but also personal user preferences (no two users will see the same!).

Following, let’s break down different recommender systems we have developed at Tryolabs.

Buyers find style inspiration in social media and search for similar styles online. But, what if they could directly upload a photo they downloaded from Pinterest or their favorite influencer on Instagram and search for matching items in your catalog? This is what Visual Search is all about.

Buyers can upload an image, and the machine will identify similar garments and accessories (top, bottom, shoes, bag, etc.) and return a set of recommendations with matching items from your catalog. You don’t have to type anything — an image is enough.

Model wearing a grey coat and 3 similar resultsResults of visual search in fashion e-commerce.

Similar Products

Imagine that you have a product on your website that has been sold, so you want to show similar items that you do have in the inventory and match in the price range.

The machine can compute the most visually relatable products in the inventory and then show filtered results prioritizing brand, tags, or price range, among others.

For one of Tryolabs' top customers, we implemented our AI-powered similar products solution that doubled the conversion rate and tripled the revenue directly associated with the recommender system.

Similar products based on look and style.

Outfit recommendation: complete the look

Unlike similar products, outfit recommendations expand the horizon of product categories recommended to the users. Based on a single item that a user is viewing (for example, a handbag), it can suggest complementary products that can be a good match in a complete outfit (for example, a top or skirt).

AI systems can learn matching styles by looking at images of human-validated looks or having some input from fashion experts expressing their preferences.

By suggesting matching combinations, retailers can generally increase the average value of the shopping cart since the users can also purchase matching items for which they hadn’t explicitly looked.

Jeans, boots, sweater, coat and earrings to complete the look. Images from Nordstrom.

Hyper-Personalized Recommendations

One of the significant drivers of the growth and adoption of AI in fashion is consumer demand for a more personalized online shopping experience. These shoppers expect product and outfit recommendations personalized to their unique style and personal taste.

The AI-powered personalization, based on user interactions, allows for a unique experience, which means in a fully automated platform, no two viewers should see the same recommendations.

Innovative fashion e-commerce brands such as Stitch Fix and The Yes are some examples of how brands leverage the power of personalization through AI. They designed their experience around personalized recommendations, allowing shoppers to narrow down the platform to what is relevant to them and encounter higher customer engagement and conversions, repeat purchases, enhanced shopping experience, and increased revenue.

A personalization layer shows different recommendations for several users for the same product based on their interests, likes, purchases, browsing patterns, and more. Depending on each user's preferences, you can customize the site experience by showing selected offers, trends, or styles.

Conclusions

It is still early for the fashion industry, as most prominent players show static landing pages, offer generic digital experiences, and even lack applied state-of-the-art search technology to improve product discoverability. However, those who adopt AI as a pillar to their business thrive in a forever evolving market.

As we continue to move towards a more responsive and personal digital experience at Tryolabs, we are confident that there is a solution for each unique brand. We like partnering up with like-minded companies to discover it.

Check the Tryolabs blog as we continue to unveil the latest trends in AI for the fashion industry.