Google Cloud and MediaAgility are helping Chicory solve for challenges in the fast-growing online food and grocery market.
This story is originally published on Google Cloud blog.
Google Cloud results
- Decreases server provisioning time from 10 minutes to 45 seconds for high scalability
- Consumes billions of analytic events every month quickly and efficiently
- Trains advanced machine learning models in less time
- Achieves fast migration onto Google Cloud by relying on expertise and best practices from MediaAgility
Achieved 7x year-over-year increase in views
Over the past two decades, ecommerce has transformed industries, from retail and travel to banking and hospitality. In recent years, the shift can be felt in everyday tasks such as grocery shopping. The increased reliance on digital shopping has many major grocery chains in the U.S. and around the world investing in online grocery shopping infrastructure to meet current and future demands.
And it’s not just the act of buying groceries that’s gone digital. Cooks of all levels now go online for inspiration, with tens of thousands of food sites, blogs, video channels, and apps inspiring people to try new recipes. Many people even plan their grocery lists around a specific meal they’ve discovered online.
New York-based tech startup Chicory saw these two growing trends and came up with a simple idea: make online recipe content shoppable. The company connects thousands of food brands, online grocery stores, independent food blogs, and major food sites such as Delish, Taste of Home, Kitchn, and Betty Crocker. Today Chicory reaches about 100 million consumers across more than 1,350 websites with its signature “Get Ingredients” button.
Improving agility through microservices
Chicory creates digital tools to connect touchpoints and analyze recipe traffic data to influence shoppers. By connecting more partners into its ecosystem—such as recipe blogs, food brands, and online grocery retailers—Chicory can grow the number of customers it serves and deliver more value for its partners. That means that it is essential for Chicory to scale quickly, onboard new partners fast, and maintain the agility needed to change course in a rapidly evolving market.
“We had one huge code base running the entire network, and we wanted to break it into microservices to be more agile,” says Asaf Klibansky, Director of Engineering at Chicory. “We started looking at Kubernetes for container orchestration, so it made sense to also look at Google Cloud due to the great support that it has for the Kubernetes platform.”
Chicory reached out to Google and experienced partner MediaAgility with questions about best practices, architecture, and migrating to Google Cloud. MediaAgility used its expertise to help Chicory come up with a plan to not only migrate onto Google Cloud, but also improve the architecture to help Chicory achieve even greater efficiency and performance.
“MediaAgility came in with recommendations and best practices that help us get the most out of Google Cloud,” says Eric Chamberlain, Engineering Manager at Chicory. “With their help, we’ve started moving into production quickly.”
“With GKE, we can provision a new server in 45 seconds instead of 10 minutes. We’re scaling our network exactly when and where we need it.” —Asaf Klibansky, Director of Engineering, Chicory
Scaling from one million to seven million views
Google Kubernetes Engine (GKE) is one of the cornerstones of Chicory’s upgraded Google Cloud environment. Chicory runs almost all its microservices on GKE, taking advantage of the industry-first four-way autoscaling in the container orchestration service to dramatically simplify provisioning.
“With GKE, we can provision a new server in 45 seconds instead of 10 minutes,” says Asaf. “We’re scaling our network exactly when and where we need it.”
“Google Cloud played a major role in helping us achieve a seven-fold increase in views over Thanksgiving.” —Joey Petracca, COO and co-founder, Chicory
For Chicory and its partners, one of the biggest days for food in the United States is Thanksgiving. People are creating menus, making lists, buying food, and checking recipes to make an incredible feast. The year before implementing Google Cloud, Chicory delivered one million views of its technology and media placements on Thanksgiving. Under GKE, Chicory scaled out massively for the big day, supporting a 7x increase in views the following November.
“Google Cloud played a major role in helping us achieve a seven-fold increase in views over Thanksgiving,” says Joey Petracca, COO and co-founder at Chicory. “GKE allows us to scale out quickly, be more agile, and serve more customers without crashing or interruptions. From a business standpoint, it was a pretty incredible accomplishment.”
Analyzing billions of events quickly
The Chicory analytics stack consumes billions of events every month to help connect partners with customers more effectively. Chicory now relies upon BigQuery to act as the data warehouse for this core service. Unlike other solutions, BigQuery separates compute and storage so that when Chicory needs to add more data, it doesn’t need to spin up additional clusters. This makes BigQuery more economical and much easier to run. Dataflow adds even more efficiency to the analytics pipeline as a highly scalable and automated means of preprocessing data.
“BigQuery runs much faster than our previous solution,” says Asaf. “MediaAgility worked with us to rearchitect our analytics stack and optimize it for BigQuery. Thanks to their help, we could make a big move without stopping work on our own suite of products.”
Building advanced machine learning models
Chicory is also seeing great results by working with the learning models in AutoML. Compared to Chicory’s previous mix of in-house and open-source machine learning models, Cloud AutoML is much more powerful and easy to work with. One of the first models Chicory has been working on is text classification. This model combs through Chicory’s entire database of ingredients to identify, remove, or better classify ingredients that might be confused with other ingredients.
For example, a recipe might refer to “aubergine” whereas the grocer would understand that as “eggplant” or confuse a chicken breast for a chicken tender. The text classification model intelligently matches the recipe’s ingredients with the correct product in a grocer’s database.
“Working with Google was such a great experience. The AutoML team worked closely with us to identify our needs and deliver on a product in just six months.” —Eric Chamberlain, engineering manager, Chicory
“AutoML significantly reduces our time to iterate and improve on machine learning models,” says Eric. “We can build advanced machine learning models without an army of data scientists.”
MediaAgility also introduced Chicory to the AutoML team to further improve its models. Originally, AutoML limited the text classification model to 100 classifiers—too small a number to properly identify all the possible ambiguities in ingredients. After talking with Chicory, the AutoML team at Google created an update to increase the number of classifiers to 5,000.
“Working with Google was such a great experience,” says Eric. “The AutoML team worked closely with us to identify our needs and deliver on a product in just six months.”
Increasing partner diversity and reach
Chicory has recently partnered with several big brand sites. With the improved scalability, flexibility, and automation gained through Google Cloud, Chicory quickly onboarded all relevant customers to make their content more shoppable.
“Every partner that we onboard increases value to all other partners on our network,” says Joey. “Google Cloud has helped us take on a more diverse range of partners so that we can continue to improve experiences for all of our customers.”