AI is not magic

I recently sat down with Christian Kleinerman, Executive Vice President of Product at Snowflake. With over 15 years of product experience, Christian previously served as General Manager of the data warehousing product unit at Microsoft and led YouTube's infrastructure and data systems at Google before joining Snowflake.

Our conversation explores how Snowflake balances customer needs with forward-looking innovation, the critical role of data governance in AI implementation, and why organizations need to maximize their data strategy if they want any success with AI. Christian shares insights on how advanced companies are leveraging AI to democratize data access, the productivity gains possible through AI adoption, and why getting your "data story in order" is essential for any successful AI implementation.

‍

‍

Transcript

This transcript has been edited for clarity.

‍

Balancing feedback vs. innovation

Jake: For those of you who don't know Christian, he's the EVP of product at Snowflake. He has more than 15 years of product experience. Prior to Snowflake, he was at Microsoft and served as general manager of the data warehousing product unit. At Google, he led YouTube’s infrastructure and data systems. 

Christian: Sounds right.

Jake: We read LinkedIn correctly, then. Alright, let's start off with continuing down the road of AI and we talk about the role of AI and Snowflake's part in it. If you pay attention to a lot of the recent announcements of Snowflake, they've talked about being flexible, meeting the customers where they are, etc.

But we're no longer in an era of optimization. We're more in an era of innovation at this point. How do you think about balancing both customer feedback, which may be backward-looking, and your forward-looking thoughts on the product or where the future is headed?

Christian: I would say for the longest time at Snowflake, we've had a clear thesis on how we want to help organizations, how we want to add value. How you get there is very much based on partnering and listening to our customers. 

We're not in this ivory tower, building technology. We want to have an ear to the ground. We feed off of opportunities and listening to what works and what doesn't work. But it's all informed towards a higher-level goal.

For me the obsession is how do we eliminate data silos? How do we help organizations be able to reason across their entire data states? AI has made it even more important. We talk about how there is no AI strategy without a data strategy. That has continued to be the case.

I was chatting with a bank a couple of days ago, and they're like, can AI help you with this? And the question was, can you find someone in your company that, if you give 'em the question, we'll find the answer. If the answer to that question is no, then I don't know that AI is magic.

AI is just a continuation of that example of how we listen to our customers innovate together, but it's also with the directional vector of how we help organizations get value out of data.

Jake: That's interesting. It starts bleeding into the discussion of how much do we stick to our old business, like tried and true, versus how much time do we spend on new, more innovative products. I think that one thing that's fascinating, going back to the different trends, is what we've noticed in cloud and later stages of the internet, how a lot of the incumbents have done a phenomenal job at protecting their moat.

So, how do you think about, for Snowflake to become a hundred billion dollar company, what does that road look like, and what's the split between those two?

Christian: Something that we did well in the early years of the company was to establish ourselves as the data warehousing company.

We had this slogan: Data warehousing built for the cloud. One of the most painful things is that we're known as the data warehousing company. And that's true of every successful initial positioning of a company once you expand. It is so difficult to get out of that initial positioning.

It's been years now. We do tons of data engineering. If you wanna do Python, you're gonna do Java. We have plenty of traditional predictive machine learning. We have AI, but many of the conversations we have are still: Hey, can I use you for data warehousing? The way I think about it is we want to help organizations with the entire lifecycle of data.

In the early years, maybe four or five, six years ago, it was very exciting, but painful. There were organizations telling us, if you do this, we'll pay you. So we had product market fit on things that we were not even dreaming of building, and they were all adjacencies to what we had built, the philosophy by which we built the original Snowflake product, which we still carry forward. 

Everything we do is focused on simplicity, ease of use, and abstracting complexity. We have broadened dramatically what we do, but always with an agency to what we want to do. I think we have no appetite to go into applications or horizontals or verticals, but it's how we help in the broader journey of data.

‍

AI's data problem

Jake: When we talk about this broader journey of data, what I think is really interesting is we’re  talking about the innovation and what's required. The AI problem is actually a data problem, for example, right? What are some of the problems that you guys are proactively solving to embrace innovation and enable customers to embrace that innovation?

Christian: Many of you in here are responsible for security and even if not, you have a role. And I say on data management and data governance. Governance is one of these fuzzy words. Many people equate it with security. But an important aspect of governance is, do I even know what data I have?

It is not uncommon to talk to larger organizations, many Fortune 500, Global 2000 companies, where they start not quite knowing what data they have, let alone how to secure it. So we think of the journey of helping organizations discover data, enumerating data, cataloging data, and classify data. 

Where do I have PII data? It's a simple question. It's a hard process. It's not just technology. Some of it is a human business processes to identify that. The answer to your question is governance of data needs to be understood. Data needs to have clear policies, clear access controls, and then I can go leverage it for AI and plenty of other use cases.

Many organizations use Snowflake primarily for the governance capabilities. Everything else, the price and the performance, is second order. I'm not diminishing it, but helping organizations. Governing data is very important.

‍

Examples of AI's potential

Jake: It’s like talking about the data vegetables to Paul's point. I'm sure you have several customers that are much more advanced than others. There's the crawl, walk, run analogy we talked about earlier. One thing we're always trying to figure out is what does the future hold and what is coming next in, let's say, the next 12 months, what are some things more advanced companies are doing in their data journey that the laggards or people that are like earlier in their journey, that you believe will become commonplace?

Christian: It's an interesting question. I think there are. Different ways to think of advanced. One vector is not necessarily more complex, but it's just forward thinking. We're working with a car manufacturer in Japan, and their insight was AI is the next level of democratization of access to data.

I've been doing data and data technology long enough that I remember the big promise of BI— business intelligence—was data for the masses. And it is true, it changed the accessibility of data from a central team with days or weeks of process to now many people in an organization can go into a BI tool and click on some dashboards and do some drill-downs and get some answers.

What this car manufacturer said is they have many people in the factory floors that don't know how to use a BI tool, but they can speak, they can ask questions, and then they built with Snowflake, a chat bot with a voice interface, and they're asking questions, is this part available? How many more? It's simple from what was done. I'm not trivializing it, it's hard work to get it to give the right answer. But it's advanced from the thinking and the power of AI. So that's an interesting use case. 

Then there are other ones which are technically advanced. I’ve been chatting with a healthcare organization that is trying to take all sorts of interesting data and train it into a custom LLM. The goal of the project is to identify patterns predictive of medical conditions. The numbers they were giving were that if they improved just one or two percent the chances of readmission into a hospital than you're creating a lot of benefit for humanity. 

The numbers we're studying were not only economically how much you spend and the cost to people, but also how you're improving lives. And this is not the future. This is actual projects.

‍

What enterprises want

Jake: That's amazing. The comment on democratization of data is really interesting because part of it is product, part of it is cost, and maybe there's some combination of the two. In the survey that we ran, we asked what is the largest barrier to adopting AI or new technologies.

What I found was fascinating, and this was a complete surprise for me. I thought security actually would be much higher. There's only 7%. The largest barrier to the adoption of AI and new technologies in general was almost half. It was at 44%. And it was cost or lack of ROI. And so as you think about, it could be AI, it could be new products, what do you feel is the ideal split between driving additional revenue and eliminating costs for customers?

Christian: I think eliminating costs sounds a little bit too narrow and transactional. I like to think more of improving productivity, which at the end of the day means costs changes. So no doubt there. But most of the use cases I see right now just take processes that today have a heavy human element and turn them into a lower human element.

I don't yet subscribe to the idea that the machines are gonna take over. I don't believe in that. But it has been for the last many years, decades, a combination of how much computers and technology can help me and how much humans do.

I think what we're seeing with AI is a shift where a lot more can be done by the machine, but still humans are validating and verifying. I think that's the bulk of the use cases. At the same time, there are things that are truly mind-blowing on what you can do with some of this technology, but some of those are more bleeding edge, and it's going to take more time.

The most common conversation I have with CIOs, CTOs, CEOs, is about “how do I improve the productivity of my organization.” We, ourselves, Snowflake, have deployed AI in lots of areas, both on our engineering team, but also our support team, which is what I think many organizations started and are starting to see the benefits.

So the productivity gains are great, I don't hear it as much when you're benchmarking against headcount and people, so that's more manageable. I would say that the folks that are saying security is not a priority or is not the biggest inhibitor may not all be as far along in the journey. I heard now several AI projects are being shut down because the chatbot leaked information or didn't honor some calls and permissions. I think that we're still in the early stages of adoption, and my sense is that we're going to see more and more security and permissioning, and privacy as an important factor. 

‍

Closing thoughts

Jake: I remember 3, 4, 5 months after ChatGPT launched, there were a handful of Fortune 500s that outlawed any AI in the enterprise. That was short-lived, but still fascinating. We are in early days, and a lot of the use cases, again, going back to the survey, the primary use cases are around developers right now, coding. And that's certainly gone mainstream. And the second, of course, to your point was customer service or customer success.

You did bring up an interesting point, and I want to end on this. Which is, and I'll ask you two questions. You can take your pick if you had to send one message to every CIO/CTO/CSO, in the world or in this room, what would that be? And what is one data-related lie that you think people shouldn't fall for? What's one common misconception?

Christian: Those two questions are related. My advice to every organization in the room, but also what I do on a day-to-day basis, is get your data story in order if you want to leverage AI for data.

Most organizations want to leverage AI for talking to the data, whether it's structured data or unstructured data, and I see so many instances where it gets into the live part of it, and there's this belief that AI is magic. And the company realizes I have not reconciled my data models for years. I do not have clear permissions. I have three copies of the same data set, and believe that somehow AI is gonna give me a single correct answer. AI is amazing technology. I'm on the bullish camp here. I think it will drive dramatic economic change. It will democratize data. It will make all of us more productive. It will create disruption.

It will create economic opportunity, but there's no magic. There are some things that are fundamental. That's the lie. And the fix to the lie is get data in order, eliminate silos where possible, eliminate copies, and that's the path to productivity.

Jake: Fantastic. Thanks so much for joining.