The wearable health and fitness technology vendor has already put advanced applications into production for its workers, with plans to dramatically expand over the next year.
SAN FRANCISCO -- Whoop is unleashing AI on its workers, with Snowflake's burgeoning development capabilities providing the wherewithal to do so.
Based in Boston, Whoop is a wearable technology vendor whose products track health and fitness through a device worn around the wrist. The device monitors metrics such as heart rate, recovery, sleep patterns and stress to help improve well-being and performance. While the wearable has no screen, Whoop's app is a hub for logging data and connecting with other users.
Famed athletes using Whoop, according to the company, include Olympic gold medalists Michael Phelps and Gabby Thomas, veteran basketball superstars LeBron James and Sue Bird, and soccer legend Cristiano Ronaldo.
Snowflake, meanwhile, is a data management vendor that has expanded into AI development over the past two years. It provides a comprehensive platform for modeling and storing data, and using that data to train and maintain AI models and applications.
After struggling to manage its growing data volume in 2022, Whoop adopted Snowflake to store and access its data. Then, as Snowflake grew its scope to include AI development -- including generative AI, and now the burgeoning realm of autonomous AI agents -- Whoop broadened its use of Snowflake to develop AI tools.
Whoop is just beginning to put some of those AI tools into production to help inform its employees and make them more efficient, but has plans for more widespread use over the next year, according to Matt Luizzi, Whoop's senior director of business analytics.
At the current pace of development, I believe that within the next 12 months we're going to be at a point where I can come up with use cases for every single department.
Matt LuizziSenior director of business analytics, Whoop
"At the current pace of development, I believe that within the next 12 months we're going to be at a point where I can come up with use cases for every single department," he said.
As little as three years ago, such development seemed impossible. But before Whoop could develop cutting-edge AI tools and put in place a plan to empower its employees with emerging technologies, the company first had to find a way to manage a morass of ever-increasing data.
Prelude to AI
Whoop's data volume has exploded since its founding in 2012.
Until switching to Snowflake in 2022, Whoop was using a combination of Amazon Redshift and Dremio's data lake platform for its data management. Redshift, while capable of handling Whoop's growing data volume, proved burdensome to oversee, according to Luizzi.
In particular, balancing cost and performance required constant oversight.
If Whoop wanted to increase Redshift's performance when running large workloads, it had to make adjustments that led to downtime. Conversely, if it wanted to reduce Redshift's performance for smaller workloads, that also led to downtime. And that was long before Whoop began building AI tools, which place greater demands on data management systems than traditional analytics workloads, given that AI applications require large amounts of data to avoid potential hallucinations.
"We were running into situations where it would be 7 p.m. and we'd have a deadline, and there would be 15 minutes of downtime that we couldn't query data," Luizzi said.
When such situations arose that delayed workloads into the night, employees made decisions to simply shelve the work until the next day, putting projects behind schedule. With too much downtime resulting in too many delays, Whoop decided to change its data management system.
Luizzi was familiar with Snowflake from previous jobs, so Whoop didn't do a full evaluation of other data management providers such as Databricks, Google Cloud, Microsoft and Oracle, and quickly adopted Snowflake.
"I fell in love with Snowflake when I started using it," Luizzi said.
Despite the difficulties that sometimes arise during data migration from one data management system to another, moving from Redshift and Dremio to Snowflake was relatively simple for a team of five people, according to Luizzi.
Fortunately for Whoop, its data wasn't as complex in late 2021 and early 2022 as it is now, consisting mostly of data ingested from SaaS applications, he said. And because Snowflake is designed to be interoperable with other systems, little of Whoop's data needed to be transformed to make it compatible with Snowflake.
"It was surprisingly fast," Luizzi said. "The hard part was the volume of data being ingested into Snowflake for the first time, but it only took a week."
The biggest change that Whoop saw after its data migration was improved performance, Luizzi continued. Queries ran much faster than they did on Redshift. In addition, because Snowflake separates compute and storage, when someone was running a query, others still had access to the data being queried.
Meanwhile, with Snowflake automatically turning on and off based on demand -- unlike Redshift, which was always running even when there was no workload demand -- Whoop now saves tens of thousands of dollars per month, according to Luizzi.
Whoop, a vendor of wearable technology to track health and fitness, is using Snowflake's platform to develop and deploy AI applications aimed at making the company's staff better informed and more efficient.
Snowflake's evolution
In early 2022, when Whoop migrated off Redshift and Dremio, Snowflake's primary focus -- as it was for Snowflake's competitors -- was data management and analytics.
The vendor had executed the largest tech IPO in history in September 2020. And its cloud-based data management capabilities, enabling customers to use BI platforms to analyze data within Snowflake rather than forcing them to move data into another tool for analysis, were highly regarded.
That focus changed for Snowflake and its peers following OpenAI's November 2022 launch of ChatGPT, which marked a significant improvement in generative AI model technology.
With GenAI's potential to make workers better informed and more efficient, many enterprises suddenly wanted to develop and deploy GenAI tools. And with data the foundation of AI, many data management and analytics vendors saw an opportunity to expand by creating AI development environments where customers can securely build pipelines that connect their proprietary data to GenAI models.
Snowflake was slower to respond than Databricks and other competitors, but since Sridhar Ramaswamy was named CEO in February 2024, Snowflake has created a vibrant suite for AI development.
"A lot of what Snowflake is doing is making [AI accessible] for enterprises with the data that matters to them," Ramaswamy said on June 2 during a press conference at Snowflake Summit, the vendor's annual user conference. "Enterprises store their most valuable information with us, and we're helping them unlock value from that ... to rethink how businesses should operate with data augmented by AI."
Toward that end, Snowflake created Cortex, a full-featured environment for AI development. Recently, on June 3, during the conference, Snowflake unveiled plans to add agentic AI-powered capabilities aimed at making it easier for users to interact with data and execute routine AI development tasks.
Beyond building Cortex, Snowflake has made it a priority to add new governance capabilities to ensure the safe and secure use of data as it's used to inform AI applications and better enable customers to make use of unstructured data when developing AI tools, according to Christian Kleinerman, Snowflake's executive vice president of product.
"Because of AI, unstructured data has gone -- almost overnight -- to being dramatically useful, dramatically more able to provide context," he said during a media conference.
Customers, some of whom were taking a wait-and-see approach to Snowflake's AI development capabilities after its slow start, were pleased with what the vendor revealed on June 3. Luizzi was among them.
"I'm very impressed by the direction that they're going," he said. "Last year's summit was a lot about laying the foundation. Now, they're really bringing it all together, and seeing that wrapped in an agent that allows LLMs to orchestrate your tasks is where the market is going. Snowflake, I find, to be right on the cutting edge."
Now, Luizzi continued, based on the tools that Snowflake has already developed and those now in the preview process, Whoop is in a position to execute its plan for widespread use of AI.
Whoop and AI
Like many enterprises, Whoop's interest in developing AI tools was sparked by ChatGPT's launch.
ChatGPT, which is trained on public data, can't, on its own, be used to aid an organization with its operations. Not having access to an individual organization's proprietary data -- its financial records, sales figures, customer data and other data -- ChatGPT had no understanding of that individual organization.
But GenAI's potential to play a significant role within Whoop, both as an aid to Whoop's employees and a means for customers to interact with their fitness data, was immediately recognizable, according to Luizzi.
In 2023, after configuring his own AI pipeline with Streamlit and other tools, Luizzi developed a crude application that enabled him to use natural language to prompt the application to write code based on a specific question. It didn't work particularly well, he noted, but it showed enough promise to serve as a proof of concept for what was possible.
"I envisioned a world where we would be able to access all of our structured and unstructured data ... using a natural language interface," Luizzi said.
All of Whoop's structured data is stored in Snowflake, while the company's other data is stored in Apache Iceberg tables that can be queried through Snowflake. However, to develop functional AI tools, Whoop needed to make sure that its data was in order, that it could be discovered when needed and that it was appropriately formatted.
Over the latter half of 2023 and into 2024, as Snowflake began to build Cortex for customers to develop their own AI applications, Whoop transformed data to make it AI-ready -- a translation process that is now automated -- and reworked metadata to make data discoverable. When a private preview of Cortex Analyst became available to Whoop in June 2024, the company was ready and able to develop more advanced AI applications than the one Luizzi built on his own.
Still, Luizzi and his team chose not to push applications into production.
Text-to-SQL translation capabilities are key when using a natural language interface to interact with data. Before launching AI-powered applications across Whoop's operations -- and for its customers -- it was imperative to Luizzi that the applications reach a certain accuracy threshold. For that to happen, although Cortex Analyst improved the accuracy of application outputs, Whoop needed the reasoning and context awareness capabilities of agentic AI.
"The larger vision was still unattainable at that time because of the lack of an agentic system," Luizzi said. "I've been aspirational throughout this process, but not pushed things out prematurely because, as with any tool that's cutting-edge, if people have a bad first experience, they're far less likely to come back to it."
Snowflake unveiled Cortex Agents, a managed service for agentic AI development, in preview in February. The vendor introduced more tools aimed at powering agentic AI workloads in preview during the conference. Among them are Snowflake Intelligence, an agentic AI-powered interface for interacting with data that Whoop has been helping Snowflake refine since January, and Snowflake Openflow, an ingestion and integration tool that joins disparate data types.
With agentic reasoning and context awareness now available through Snowflake, Whoop is at the point where it is now putting into production the AI tools that it's been refining for two years, according to Luizzi.
"Now, it's about how we can enable people," he said. "Making AI part of our DNA is where we're headed."
Among the applications now in production are those that do sentiment analysis to glean insights about customers, summarize surveys and look for trends within Whoop's data.
Looking ahead
With Snowflake's capabilities, Whoop has developed custom GPTs. The language models are trained using Whoop's proprietary data so that the models have the context awareness to respond to queries about subjects such as marketing and sales.
Those custom GPTs are the foundation for natural language interfaces that enable Whoop's employees to engage with data.
"AI is deeply at the center of our culture at Whoop, not just in the data team, but across the board," Luizzi said.
Despite how Whoop has been able to evolve along with Snowflake, and how the company is now using Snowflake to fuel its AI initiatives, there's always more that an enterprise wants to do than it can now.
Snowflake's AI development capabilities are not quite complete, Luizzi said. But he expects that within a year, they will enable him to come up with use cases one minute and build AI applications for those use cases minutes later.
Examples include translation, given that Whoop is an international company, as well as image generation and modification of content for marketers. However, Snowflake's agentic reasoning capabilities currently result in about 90% accuracy. Whether that is good enough for a particular application or Whoop needs to wait for that percentage to improve further depends on the use case and the technical skills of the potential user, according to Luizzi.
"I can envision coming up with an idea list and passing through as we figure out new use cases," he said. "I've found that Snowflake is listening and developing really fast. That's what gets me excited. I can push the boundaries of what's possible and get these [applications] running really fast."
Over the next 12 months, Snowflake's ability to deliver on the vision of the capabilities it revealed on June 3 will be key for Whoop to develop and deploy all the AI applications that Luizzi envisions. In addition, he noted, as foundation models continue to improve, not only will it be imperative for Snowflake to partner with AI developers such as OpenAI, Anthropic, DeepSeek, Meta and Mistral, but it will also be important for Snowflake to make the models built by those developers available in a timely fashion.
"The rest will take care of itself," Luizzi said.
Eric Avidon is a senior news writer for Informa TechTarget and a journalist with more than 25 years of experience. He covers analytics and data management.
