Fit Analytics: Solving sizing at scale with Google Cloud
About Fit Analytics
Fit Analytics' machine learning platform helps the world's best apparel companies solve sizing, sell smarter, and turn data into actionable insight.
Tell us your challenge. We're here to help.
Contact usFit Analytics uses Google BigQuery, TensorFlow, and Google Kubernetes Engine to streamline and scale the machine learning behind delivering 250 million size recommendations a month with 99.99% uptime.
Google Cloud results
- Reduces data processing times by a factor of ten with Google BigQuery
- Taps into a rapidly expanding library of knowledge through the global TensorFlow community
- Supports 99.99% uptime with microservices on Google Kubernetes Engine
Data processing in minutes, not hours
The Fit Analytics data platform powers more than 250 million size recommendations every month for over 150 of the world’s top apparel and footwear companies. Brands such as ASOS, The North Face, Tommy Hilfiger, JD Sports, and Calvin Klein use the company’s Fit Finder size advisor to help boost conversions, reduce returns, and help shoppers find clothes that fit.
“Sizing and fit are the biggest consumer pain points when it comes to buying clothes online,” says Andrew Delany, Head of Marketing at Fit Analytics. “The traditional approach of using a static size chart results in a poor user experience for shoppers as there’s simply no consistency in sizing between different brands and retailers. We use a combination of user input, machine learning, body modeling expertise, and the industry’s biggest data set to generate accurate sizing recommendations. Shoppers get a better user experience and our clients benefit from better margins; we typically see conversion lifts of up to 11.5% and returns reduced by 4.4%.”
As the company signed up larger retailers, scalability became a key concern. “Processing times started to increase,” says Gunnar Kedenburg, Head of Data Science at Fit Analytics. “That’s something we couldn’t allow to happen, as it threatened our ability to work effectively with the data. Google Cloud drew our attention because it has great support for machine learning and data processing, including access to the state of the art in scalable systems that would allow us to keep processing times down.”
Availability is vital to the success of Fit Finder and other Fit Analytics products. Delivering 99.99% uptime demands a strong technical setup, particularly with more than 300,000 requests per minute passing through the system during peak periods. Security and data privacy were also key concerns. Careful anonymization helps ensure that the privacy of Fit Analytics users’ data would not be compromised. With Google Cloud, all data stored by Fit Analytics is additionally protected using best in-class measures.
“Our clients’ service level expectations are high, and the 99.99% availability Google Cloud helps us provide means we match them. The Google Cloud infrastructure helps enormously with capacity planning, and lets us deliver rock-solid service to clients throughout peak periods, such as online sales or the holiday season.”
—Terese Haimberger, Team Lead Backend Development, Fit AnalyticsAfter surveying the provider landscape, Fit Analytics selected Google Cloud as an optimal machine learning and data processing environment with impressive scalability.
Engineering optimal environments for scaling and stability
The initial integration project focused on using Google BigQuery to optimize processing of purchase and return records, which are central to the Fit Analytics data platform. As Gunnar explains, transitioning to the new environment proved to be straightforward: “We expected to have to change a lot of our code, but that wasn't the case. It was minimal. We were using relational databases before, so we were really happy to see that everything we do runs well even in Google BigQuery, which has a different design.”
The improvements in processing times were impressive. After implementing Google BigQuery in its data processing pipelines, Fit Analytics found processing times reduced by a factor of 10. Pipelines that ran for hours would now finish in minutes.
Following the success of the Google BigQuery deployment, Fit Analytics looked to enhance the scalability of its platform with a microservices architecture on Google Kubernetes Engine. “We have a lot of different services that we run in Google Kubernetes Engine,” says Terese Haimberger, Team Lead of Backend Development at Fit Analytics. “We can create and resize clusters very easily and if something breaks, it is automatically restarted.” In addition to providing enhanced system-wide reliability and scalability, Google Kubernetes Engine also makes day-to-day life easier for Fit Analytics’ developers and data scientists as they create and refine the company’s suite of products.
“By running Google Kubernetes Engine clusters instead of provisioning our own, we can help ensure a quick turnaround time,” says Terese. “We can run a big cluster for a short period of time with a very low engineering overhead, while safeguarding against future scalability concerns. Another major advantage is the easy separation of staging and production the platform enables; that’s a big plus for our development team.”
Of course, the benefits aren't just internal, as Terese points out: “Our clients’ service level expectations are high, and the 99.99% availability Google Cloud helps us provide means we match them. The Google Cloud infrastructure helps enormously with capacity planning, and lets us deliver rock-solid service to clients throughout peak periods, such as online sales or the holiday season.”
Staying close to machine learning’s cutting edge
Machine learning is integral to what we do,” says Gunnar. “It’s how we tie our predictions to real-world results, such as whether a similar recommendation was successful in the past. It allows us to uncover how different products relate to each other, and to users’ body measurements.”
“Google Cloud lets us run our fit prediction system in an almost entirely serverless way. Workloads run transparently on large Google managed clusters, which are more powerful than anything we could build ourselves, and require zero maintenance. The rapidly evolving nature of the platform also helps maintain our competitive edge.”
—Gunnar Kedenburg, Head of Data Science, Fit AnalyticsWith machine learning so central to Fit Analytics’ business model, staying up to speed with best practices and developments in the wider AI community is essential. “We see Google’s intense engagement with technologies such as TensorFlow and consider it a competitive advantage to be close to the source of these developments,” says Gunnar.
For the company, it was important to build a system that would allow it to concentrate on its core strength, machine learning, without creating unnecessary engineering overhead.
“Google Cloud lets us run our fit prediction system in an almost entirely serverless way,” says Gunnar. “Workloads run transparently on large Google managed clusters, which are more powerful than anything we could build ourselves, and require zero maintenance. The rapidly evolving nature of the platform also helps maintain our competitive edge.”
Rapid product development while keeping clients satisfied
Having a managed environment means development and product teams can focus on delivering value to clients, without worrying about additional scalability or maintenance concerns. “Our range of solutions is rapidly expanding to include style as well as fit information,” notes Chief Technology Officer Tom Shenhav. “Our ongoing aim is to make it as simple as possible for apparel companies to connect customers with clothes they’ll love, and new personalization features such as the ability to provide fit-aware product assortments on the fly are a big part of that. Having Google Cloud in the background means we can rapidly develop these types of solutions with confidence.”
Tell us your challenge. We're here to help.
Contact us
About Fit Analytics
Fit Analytics' machine learning platform helps the world's best apparel companies solve sizing, sell smarter, and turn data into actionable insight.