How Computer Science Can Make You More Fashionable

Using Google Trends & data analytics to discover the world of fashion.

Two summers ago, I came across and read one of my favorite books to date; Seth Stephens-Davidowitz’s Everybody Lies. In short, his book is a 300-paged thriller on what Google Search data can tell us about the world, & the secrets of the people around us. Davidowitz’s insight was one of the reasons I was so excited about the work I’d be doing at Google, as well as a catalyst for my interest in data analytics.

So, what does this have to do with being fashionable? The fashion industry is notably one of the verticals slowest to adapt new technology, where once-unstoppable giants (Forever 21, Barney’s NY) are approaching fateful horizons primarily due to an inability to integrate tech as quickly or as successfully as their competitors (Zara, Revolve).

As a result, data analytics and tech integration have become more important than ever in order for fashion brands to thrive and stay thriving. Pursuing fashion data analytics, in fact, is actually a very viable path to becoming the richest man in Spain. Nowadays, any “old school” approach only works if your technological tools are as advanced as ever.

So the solution is simple, right? Fashion companies just need to start acquiring more marketing data & running more data analytics! Not exactly…

Data Misunderstandings

Marketing data, recently reported to only be around 10% or 20% accurate, will no longer do. Humans are inherently risk-averse, and in a field like fashion, where preferences are personal and style acts as a form of otherwise shied identity expression, one can’t always listen to what people are saying they want. Instead, we need to examine what styles customers are actually, anonymously and honestly, searching for.

Let’s take a look at the ever so famous Chanel, for example. When I visit their website’s fashion homepage, I am presented with images of classic staples and models on recent runway shows. Google Search data, however, shows us what the people really want from Chanel right now: slides.

That’s right, this past spring the search term “Chanel slides” spiked by nearly 50%, higher than nearly any other Chanel-related term, and saw a 60% overall increase in the past year. The term “Chanel bag”, in comparison, sees a similar spike in late December (sale season?), and is otherwise around 10% lower in popularity than “slides” throughout the spring and summer. To conclude, Chanel’s slides are as desired in the summer season as much as their famous bags are sought for in the winter. Who would’ve guessed that one of the world’s biggest fashion houses is equally as wanted for its luxury items as its…poolside accessories?

So, Chanel website in July, why is it so hard to navigate to your shoes section? Why is there no image or direction as to where I can find these “slides” everybody is looking for? In fact, why is there not an entire section dedicated to them?

Big (and better) data

Although Chanel is actually doing pretty well tech-wise, the above is a highly applicable example of the difference between what fashion companies think customers want versus what customers are actually (secretly) looking for. So, while integrating more advanced methods of data analytics is certainly important, ensuring that the data is most accurately mirroring consumer preferences is just as monumental. Thanks, Google Trends.

Business of Fashion’s August 2nd podcast episode entitled ‘Inside H&Ms $4B Inventory Challenge’ highlights this misunderstanding between fashion companies and consumers; that what companies think people will want is not really what they end up buying. Thus, this problem extends far beyond whether or not something is displayed on a web page; it adds to the fashion industry’s massive sustainability issue.

New sustainable fashion startups such as Reformation boast ordering smaller amounts of inventory in an attempt to eliminate future waste. While this is an amazing first step, it is only scraping the surface. Producing large amounts of a product is fine, in fact, if everything ends up being bought. Therefore, if data analytics can help companies pinpoint what people will buy before even producing it, these small inventory trial-runs will no longer be necessary.

Time is of the Essence

Another detail our Chanel example highlights is that, regardless of what analytics are being performed, they become increasingly more valuable if they are reported in real-time. Trends and styles are consistently changing, and being unable to keep up with this information means fashion companies are wasting resources focusing on what has already run away.

Marketing and design are incredibly effective tools in selling & propelling products, with a significant example being Pepsi’s success in the 20th century. Today, these tools are simply not enough. Although technological advancement seems like an obvious investment, many fashion companies have yet to adapt, or have attempted to and are not doing so properly. Bye bye, Barney’s. Bye bye, Forever 21.

Okay, Google

When my Google internship project was broadly defined as creating a solution for businesses integrating G Suite and Google Cloud, I decided to zero-in on the fashion and luxury vertical. Not that I’m the most fashionable girl on the block, but I knew my engineering skills coupled with lots of fashion research would allow me to create a product that could propel the fashion-tech world to some degree. Thanks to my working at Google’s NYC location, conducting said fashion research has been widely accessible, and can be as easy as taking an afternoon stroll down to SoHo.

What I am working on — and I won’t make you sign an NDA before telling you — is exactly what I’ve been yapping on about: a machine learning tool to allow for better data analytics performed on fashion images. WhAt DoEs tHiS mEaN???!?!? It means I am developing a tool where a machine can tell you exactly what fashion trends are present in a photo, and later exports this information as a visual data chart in Google Sheets. These identified trends will, of course, be determined by Google Search data.

If you’re interested in the nerdy details, you can view my project’s code on my GitHub.

My model was developed using Google’s AutoML tool, and was trained on 500 images ranging from runway shows to celebrity appearances to street fashion shots. The labels I trained my model to recognize were based on Google Trends search data, where I could learn the current and up-and-coming fashion trends the world is searching for.

Machine learning models that detect fashion trends & style are tools that fashion companies need in order to keep up with technological advancements, run stronger and more accurate data analytics, and best understand what consumers are buying, wearing, and looking for.

There is still much progress to be made with regards to properly integrating technology into the fashion sphere, and it’s quite exciting that here, at Google, I can maybe be a small part of that.


Google Trends data is made available to the public — check it out by visiting trends.google.com. All of the data presented in this blog post was obtained from publicly accessible information.

Additionally, as a Developer Relations Engineer, it is actually in my job description to publicly talk about & share my work. Sorry, hackers — no official Google secrets were harmed in the making of this blog post!

If you are currently working in fashion-tech or related projects, I’d love to hear from you!