Snowflake’s CEO Sridhar Ramaswamy & Axios’ Ina Fried

Thank you.

Hey, great to be here. So when it comes to AI, Snowflake is a data platform. So people use Snowflake to bring data from disparate systems. We have a partnership with Workday to share data between Workday and Snowflake without having to copy. So in that sense, it becomes a little bit of a 360 over all of the most important data. When we think about data, we think about things like Google Drive or Office where documents are stored. But often the most important information data that a company has is around things like its performance. How well is the sales team doing? And where customers are seeing a lot of ROI is in reducing friction into how their teams are getting at this data. And then as May was saying, if you can begin to then orchestrate actions around the data, magic happens. Absolutely. In areas like customer support, those are the easy ones. But we see people doing this on things like observability data. We acquired a company called Observe. What they do is they collect telemetry information from various applications, web services that are running out there and make it super easy for the on-call person to get at the right data. And agent AI is great at orchestrating access to this. And so a slew of customers, whether it's AstraZeneca or a small asset provider like TS Imagine, they all are taking this basic formula of reimagining various kinds of work as here's a smart model that can help you think through how to access a set of really important data sources that are often sitting on Snowflake.

Well, we still have to work through in all of these situations what the ROI is. And we also have to acknowledge that diffuse impact is hard to measure. What I mean by that is let's say you made getting to sales information 10% faster. The immediate question that any reasonable CEO is going to have after that is so what does it mean that I get to have 10% fewer people? And so making sure that every project has a clear notion of what's the value that it's going to be delivering as it scales and also what it takes. Many companies end up forgetting things like governance. It's actually quite terrifying to put up a data set into an AI agent and give it to a whole slew of folks if you're not sure that you're managing the permissioning underneath correctly. For us, for example, we launched a sales agent within Snowflake and it was a replacement for our dashboarding solution. The clear ROI that my data team gave me was instead of spending 5 million a year on licenses, we could be spending internally roughly a million and a half on providing the same data to the entirety of our sales teams. It was a replacement. That was a clear ROI for me. And now they tell me additional things like, oh, here is how much time we're saving for the sales team. And I kind of, that's very nice, but you know, there's the hard ROI that comes with it. One of our customers, TS Imagine, they basically augmented this. They had a small customer support team, but they augmented it using an agentic solution that was built with Snowflake Intelligence. What they got, they didn't let any of the folks in that team go, they kept them. On the other hand, they concretely said, listen, now support for complex asset managers that we supply data to is more of a 24/7 thing. I used to run support from 9:00 to 5:00 on weekdays. This thing can now run every single hour, every single day. That's a measurable step forward for me as a business. And so these are the things that one needs to work through. There's also, you know, worries about things like cost. Our customers are asking for, first they love the consumption model which is pay as you go. But on the other hand, they want to make sure that there's no runaway cost. That one extra ambitious person that May was talking about wasn't burning through $5,000 of AI every single day. And so these are some of the things that you need to drive impact at scale. And finally, there's also very much the people change that you have to grind through in every group where we want people to use different kinds of AI tools. Sometimes it's coding agents, sometimes it's data agents like the ones that we have or it's even the tools that are provided by the likes of Google and Microsoft Gemini. I've pretty much started publishing reports of how my exec staff is using these tools. It's a little bit of sometimes naming and shaming. But that's sometimes what it takes to drive actual usage because people are afraid of these systems. And so you kind of have to do a lot of work to work through the psychology of what does it even mean for people to adopt a product that is ostensibly really great for them.