Commentary: Savitude and StyleSage leverage AI, machine learning in fashion retail

Savitude brings speed and analytics to fashion design

Savitude brings the approach popularized in “Moneyball” to the traditional business of fashion — that is, it brings data and analytics to an inexact business. In the process, Savitude hopes to have at least as significant an impact on the fashion industry as the Oakland Athletics had on baseball.

I asked Camilla Olson, Savitude’s CEO, to explain the problem the startup solves.

“One of the biggest problems in the fashion design/development phase is that much of a designer’s work is done by hand,” Olson said. “The manual nature of the work calls for an idealized female form — an hourglass body shape, of average height — as the standard for design and sizing, limiting the ability to add improvements. Further, the work is highly subjective. There have been no improvements, and no opportunity for technical advancements nor analytics.”

One result of this approach is that consumers often feel dissatisfied, leading to a large volume of unsold inventory and returns.

Olson added: “Technical improvements have happened after ideation — 3D design and computerized pattern making, for example. But nothing in the ideation phase has changed. That is the longest phase, and it is still done by hand. The result of using the hourglass framework is that no other body shapes are contemplated during ideation. So, of course you would infer that a large portion of the more than 80% of women who do not fit the hourglass shape would be dissatisfied. It’s no surprise there are high returns.”

According to Olson, “As much as the fashion industry has tried to eliminate returns, it still has a record amount of inventory, $550 billion, in warehouses. The problem it faces is how to use this returned inventory in a way that is both profitable and sustainable. Just dumping the returned inventory is not environmentally sustainable. However, reworking the goods and reselling them is time-consuming, costly and risky.”

She asks, “Can the fashion design-to-manufacturing-to-sales process be more circular to repurpose excess inventory and returns in the next season? More importantly, can the excess inventory be prevented from accumulating in the first place?”

These are important questions. They are questions that are being asked by executives in the fashion and apparel industry. In some instances — especially in Europe, local laws are being enacted to compel the industry to more quickly adapt its value chain and business models so that the industry causes much less harm to the environment.

“Savitude’s AI Design Intelligence is the work product of fashion designers and engineers working side by side building an extensive knowledge base, along with machine learning, leading-edge visual recognition and deep learning technologies,” Olson says.

Essentially, Savitude captures the thought processes of a fashion designer in code.

Currently, Savitude offers two products:

● AI Design Intelligence, which is the core technology behind both products. It includes a knowledge base with 40,000 rules and takes into account color, print and texture. It works with the results of Savitude’s image recognition technology.

● AI Curator, which makes personalized recommendations based on body shape to e-commerce shoppers with fully accessorized outfits — this is also available as a Shopify plugin. Savitude says this has led to an 11.1% increase in sales.

Design Studio, which is an easy way for fashion designers to incorporate visual inspirations and current trends into the design process, is in public beta. Fashion designers who want to test the product before it is made commercially available can sign up through Savitude’s website.

I asked Nick Clayton, CTO of Savitude: “What is the secret sauce that makes Savitude successful? What is unique about your approach? Deep learning seems to be all the rage these days? Does Savitude use a form of deep learning? Reinforcement learning? How do you handle the lack of high-quality data?”

Clayton responded: “We use both deep learning and reinforcement learning, deep learning in our image recognition stack and reinforcement learning on our knowledge base. The lack of high-quality data is handled by the deep domain expertise, allowing us to generate our own high-quality data where needed, but also allowing us to bake that expertise into our algorithms to draw reasonable conclusions from smaller amounts of data.”

He added, “Our core knowledge-based system is also largely explainable, which is to say we can dig into the data and figure out why it is making many of the judgements it makes, which helps as well.”

This should help Savitude as it seeks to gain acceptance from fashion designers at large fashion brands and the head merchandisers who make buying decisions at fashion retailers.

Lest you think this is “a bit far out there,” Amazon’s Lab 126, Google’s Project Muze with Zalando, Adobe in collaboration with the University of California at San Diego, and Microsoft, are some of the well-established companies working on ideas similar to what Savitude does. There are also a few less recognizable startups creating products to tackle aspects of the problem Savitude is solving.

Before founding Savitude, Olson founded two predictive modeling companies in the pharmaceuticals industry. She had successful exits with both. She went on to earn a master of fine arts degree in fashion design, then she ran her own label and e-commerce business for five years. There she lived the problem of “fit.” She began to relate her current circumstance to the modeling concepts used in her pharma companies, leading her to found Savitude with Clayton.

StyleSage powers speed to market in fashion retail

StyleSage is a software-as-a-service analytics platform that helps fashion retailers and brands make in-season and forward-looking merchandising decisions much more quickly to keep up with the accelerating pace of change in the fashion industry.

I asked Elizabeth Shobert, StyleSage’s VP of marketing, to explain the problem StyleSage solves for its customers.

“We’re providing insight into what’s happening outside a retailer’s (virtual) four walls,” Shobert said. “Specifically, the data we provide tells brands what consumers are looking for at the moment, which new products are being introduced into competitors’ online assortments, how much they’re being priced at, and when they’re getting marked down. In most cases, this information isn’t being gathered and digested at all or at any sustainable scale in organizations today. So we’re helping brands make both creative and analytical decisions, with this timely and accurate data. It goes without saying that both now and in the future, retail winners will be defined by their responsiveness.”

Shobert added, “Our typical customer is a retailer who’s focused on their digital sales channel, a retailer who wants a compelling and differentiated offering for their customer. Many of our clients have a global presence and are using our tool to get their local pricing and product mix right.”

Next I asked, What is the secret sauce that makes StyleSage successful? What is unique about your approach?

“What’s unique about us is that left- and right-brain idea mentioned previously,” Shobert replied. “We’ve taken the analytical rigor of a management consultant and paired that with easy-to-understand data visualizations for all types of roles in fashion retail organizations. In fact, we’re a strategic partner with BCG on fashion retail projects. The accuracy of the data and the flexibility of our platform is unparalleled in this space.”

Next, I observed to Alicia Perez, StyleSage’s data scientist: “Deep learning seems to be all the rage these days.”

She responded: “Deep learning indeed still represents the biggest trend in machine learning. The family of algorithms and techniques contained under the umbrella of the term ‘deep learning’ is actually not so recent. It has been around for almost a decade. It’s true that it was transferred from academia and adopted by the industry more recently, but it has become the de facto solution in the industry for dealing with unstructured data, such as text and images. The capability of these techniques to learn very good representations directly from raw data is still providing state-of-the-art results in problems related to the analysis of images, video, text and sound. Not only when we talk about industry use cases, deep learning is still the most addressed topic and solution adopted in top-tier research conferences.”

To explore in a bit more depth what AI and machine learning approaches StyleSage uses, I asked Perez, “Does StyleSage use a form of deep learning? Reinforcement learning?”

She said: “Deep learning is indeed key in our machine learning model zoo. It’s really difficult to extract good structured product data from fashion e-commerce sites. In retail, there is no requirement about how a product description should look or what information the e-shops should contain. The only pieces of information which are always available on an e-commerce site are images and some text, and deep learning is indisputably the best technique for modeling good representations directly from raw images and text.”

Going a little further to draw a distinction, Perez said, “Reinforcement learning on the other hand is focused on creating algorithms (usually software agents) by learning through trial and error in a controlled environment. Currently our business cases are not centered in learning behaviours but knowledge, so reinforcement learning is not really a strong part of our AI solutions.”

Finally, I asked Perez; “How do you handle the lack of high-quality data?”

“We consider ourselves really lucky because we have access to a mountain of historic data, spanning five years, from fashion e-commerce sites, which act as our data lake,” she said. “We’ve been working on such a huge amount of products for years, cleaning, curating and labeling the data. Based on that, in our team we have Quality Assurance experts and we have developed our own tools focused on our data needs and with specific key performance indicators that make the hard process of building a reliable data lake easier.”

Conclusion

Perhaps no industry has taken as big a hit to the underlying fundamentals of its business models as the global fashion industry has endured from COVID-19. According to “Fashion’s Big Reset,” a June report by the Boston Consulting Group, “Industry revenue this year could drop by more than one-third, the equivalent of up to $640 billion in lost sales.”

In a sign of the dire straits in which the the industry finds itself due to COVID-19, on Oct. 8, the Guardian reported that garment workers in countries where most apparel manufacturing takes place are being denied about $16 billion of payments because big fashion brands in the West have canceled orders that were placed before the pandemic or have simply refused to pay for orders that were already in production.

On Oct. 10, The Wall Street Journal reported (subscription required) that investors focused on sustainability are turning more of their attention to social issues. This is likely to affect the fashion industry significantly. On Sept. 22, The New York Times published “Can Luxury Fashion Ever Regain Its Luster?” — an article that questions if the business model that supports luxury fashion has a future in a post-pandemic world.

These are questions my partner and co-founder, Lisa Morales-Hellebo, and I grappled with in a series of articles in late 2018: “The Fashion Supply Chain Is Broken,” “Where Will Technological Disruption in The Fashion Supply Chain Come From?” and “What Are The Established and Emerging Business Models in The Global Fashion Industry Today?”

We argue that the failure of large fashion industry incumbents to adopt technology in strengthening their business models puts them at risk from: established technology companies with one eye fixed firmly on the margins that fashion brands command; emerging technology startups that develop products that change the value-extraction equation to the detriment of fashion industry incumbents; or digital-native fashion brands that more effectively combine technological know-how with expertise in fashion and apparel.

Alas, we did not predict that a pandemic would come along barely a year later and accelerate by about a decade the trends we have been observing.

The Boston Consulting Group states things most succinctly in its report noting that fashion companies must “Build a strong AI and technology backbone to digitize core processes.”

Going further, the authors of the report state that, “Adoption of advanced analytics and artificial intelligence (AI) will be among the fashion industry’s most significant changes in the coming decade, and one of its biggest challenges. Brands have been slow adopters for the most part, but laggards can take a page from companies that already use these technologies. Data will become an even more important competitive advantage: the brands with the most usable data will win. The biggest winners will be the brands that can codify data from all sales channels and consolidate it onto a single analytics platform to improve decision making on such core processes as planning, buying, promotions, markdowns, and in-season inventory management. Creating a market-leading e-commerce platform is another part of building out a more robust tech backbone.”

This graphic from the second article in the series by me and Lisa should make the risk that fashion brands face obvious. The data is from 2018, but the directional conclusions remain relevant, and in fact, may have only strengthened since then given how well technology companies have performed during this pandemic relative to fashion companies.

The bottom line is this: The outlook for startups like Savitude and StyleSage is positive. Very positive.

If you are a team working on innovations that you believe have the potential to significantly refashion global supply chains, we’d love to tell your story in FreightWaves. I am easy to reach on LinkedIn and Twitter. Alternatively, you can reach out to any member of the editorial team at FreightWaves at [email protected].

Dig deeper into the #AIinSupplyChain Series with FreightWaves.

●     Commentary: Optimal Dynamics — the decision layer of logistics?

●     Commentary: Combine optimization, machine learning and simulation to move freight

●     Commentary: SmartHop brings AI to owner-operators and brokers

●     Commentary: Optimizing a truck fleet using artificial intelligence

●     Commentary: FleetOps tries to solve data fragmentation issues in trucking

●     Commentary: Bulgaria’s Transmetrics uses augmented intelligence to help customers

●     Commentary: Applying AI to decision-making in shipping and commodities markets

●     Commentary: The enabling technologies for the factories of the future

●     Commentary: The enabling technologies for the networks of the future

●     Commentary: Understanding the data issues that slow adoption of industrial AI

●     Commentary: How AI and machine learning improve supply chain visibility, shipping insurance

●     Commentary: How AI, machine learning are streamlining workflows in freight forwarding, customs brokerage

●     Commentary: Can AI and machine learning improve the economy?

Author’s disclosure: I am not an investor in any early-stage startups mentioned in this article, either personally or through REFASHIOND Ventures. I have no other financial relationship with any entities mentioned in this article.