A major bottleneck for AI projects is the data preparation. The process is often expensive, time-consuming, tedious and prone to error. Keep in mind that the data is usually trapped across multiple silos and as for the tools for storing and processing the information, they are often legacy on-premise data warehouses that can take weeks to produce meaningful results.
Despite all this, there is some good news: A fast-growing company, called Snowflake, is helping to break down the walls. Founded in 2012, the company has built a native cloud data warehouse that now has about 2,000 customers like Slack, Adobe, Square, Capital One and A&E. Snowflake has also raised a hefty $928.9 million from tier-1 VCs, such as Sequoia Capital, Altimeter Capital, ICONIQ Capital, Redpoint and Sutter Hill Ventures.
Oh, and there is a tier-1 CEO, Frank Slootman, who came on board in early May. He has a sterling resume in the enterprise software world, having led Data Domain from the early phases to an IPO and then to a sale for $2.4 billion to EMC. After this, he took the CEO spot at ServiceNow when it had less than $100 million in revenues and took it to $1.4 billion. The company currently has a market cap of $48 billion.
OK then, what makes Snowflake so important? And why has it proven to be critical for AI and ML (machine learning)? Well, there are many reasons. But let's look at some of the notable ones, like the following:
- Easy Setup: By filling out a simple form, you can immediately provision the service. There's no need to wait for IT to buy servers, for example.
- Zero Planned Downtime: Simply put, Snowflake gets upgraded without interruption.
- Architecture: The storage and compute capabilities are separated. Because of this, it is easy to scale up or down the service because the data is stored only once. There is also a separation of resources, which means that there is no competition for fixed IT assets within an organization.
- AI/ML Applications: Snowflake integrates with the entire ecosystem, including connectors to Python and Spark (a majority of the customers are doing modeling and predictive analytics).
- Business Model: Snowflake goes against the grain of the SaaS subscription approach. Consider that the company instead charges by usage, which harkens back to the old-school mainframe model of time sharing. But ironically enough, it could be highly disruptive to the cloud industry since usage pricing is often much cheaper. Let's face it, there are difficulties with estimating the number of users for subscriptions as well as how much of the service may be consumed. "I think SaaS is transitional," said Frank, "just like client-server was."
- Multi-Cloud: Snowflake is available on AWS and Microsoft Azure (and it is currently being rolled out on the Google Cloud Platform). This is another potential cost benefit as customers can have a bake-off between the different platforms.
Quite a bit, right? Definitely. Although, the broad range of features can actually be overwhelming.
So to get a clearer understanding of how Snowflake works, let's take a look at some use cases. One is Weather Source, which operates a massive historical database of detailed weather information. This has proven to be useful in creating AI/ML models for forecasting and optimization across industries like energy, retail, financial services and agriculture.
With Snowflake, Weather Source can now share its data with its customers without the need for setting up complex FTP sites or other traditional delivery systems. The reason is that the data is in one location and is linked to customer databases (the information seamlessly shows up in the tables). This not only results in steep reductions in heat, electricity and storage, but the data is live. In other words, all changes are instantly updated across the tables.
Now another interesting use case of Snowflake is Sigma, which has developed a cloud-based BI (Business Intelligence) tool with a one-click integration. This makes it possible for anyone in an organization to easily explore and gain insights from data, which is likely to unleash significant value. After all, according to Gartner, only about 35% of employees use traditional BI systems because of the challenging setups, poor user experiences and complex modeling.
Bottom Line On Data & AI
"Data is a strategic asset," said Frank. "This has been true for the past 30 years."
But to get real value from data requires taking unconventional approaches. For example, Snowflake recently launched its Data Exchange, which is essentially an appstore to allow third parties to gain access to data sources. But this is not just a "data dump." The Data Exchange also allows for personalized data shares.
"One of our core values at Snowflake is to think big," said Frank. "And this requires that we continue to reimagine the business."
Tom (@ttaulli) is the author of the book, Artificial Intelligence Basics: A Non-Technical Introduction.
