Transcript#
This transcript was generated automatically and may contain errors.
Hello everybody. I'm very, very happy to be here today. I'll be joined by four guests and it's my absolute, absolute pleasure. We're going to talk about productivity, the myth around it in AI and customer stories, real customer stories. You've seen this metric before, 4% of initiatives are actually showing ROI and everybody's fighting on this, trying to make this as high as possible. This is what I call the illusion of productivity because everybody feels that everything is going faster with AI but there is actually very, very little companies that are getting a real ROI from it.
So what do we do with all this productivity? If it's not reinvested in ROI? Well, the reality check is that if we're not redesigning the workflows, if the business metrics don't move and if shadow AI is growing and things are not governed, then yes, of course, there will be no ROI.
Well, the reality check is that if we're not redesigning the workflows, if the business metrics don't move and if shadow AI is growing and things are not governed, then yes, of course, there will be no ROI.
Three traps killing AI ROI
There are three traps that we've seen across all our customers and this is something that we're all helping with. The first thing is just measuring activity instead of ROI. You can have anybody in your team going five times faster but if nothing else is happening, this is not value, right? Real ROI is in spendings being down, costs being down, lower loss, working capital optimization and obviously more and faster revenue and this is where we want the impact.
The second area where we've seen like a second trap, it's attribution issues. It's easy to attribute in the front office but there's a lot to harvest in the back office of your companies and this is where we see a lot of ROI. So look at like agencies, BPO's, anything that SaaS where you can potentially cut, like those are areas where you can really harvest a lot of ROI. And the third trap is considering governance as just something heavy. It's actually a speed layer. It's something that enables you to scale safely.
Where to invest for ROI
So, okay, we've seen some traps. Where are the best areas to invest? Number one, revenue. So anything where days equal money and can be accelerated with AI. Quote to cash, claims to pay, release to market, all those things in product organization, in the back office, everywhere. You can find those areas where days equals money. This is where to go. The second part is spend compression. I talked about like the outsourcing costs, the agencies, the SaaS costs. These are areas where we see a lot of return on investment. And the last part is risk, right? Anything related to risk reduction. A lot of you here are in regulated companies and even in other, in all kinds of companies like fraud leakage, compliance failures, and so on. Those are areas where you can really harvest ROI.
What's actually very interesting in the recent studies is that you can go on your own, you can do the build, or you can go with partners. And what has been shown by research is that when you go in with partners, you have 2x likelihood to get to reach deployment for your initiative. So this is how you get, if you have the right ROI focus and the right partners, this is how you get the results. Why? It's actually logical, right? Partners have repeatable playbooks. You're not reinventing the wheel. You benefit from production-grade architecture from the get-go. They help you redesign the workflow and it also brings a discipline around measurement.
Snowflake is a partner, you can imagine, right? We're working with over 13,000 companies and providing value with AI and with the data platform for all those companies. But we're not on our own. We actually have a whole ecosystem of partners where we've been investing in creating deep integrations so that the experience that you have as a customer and the value that we provide together is the strongest in the market. And four of them are actually in the room today. And this is what you're gonna hear about. So the next step now, I'm gonna have Kelly from Fivetran join. And Kelly is gonna tell about one customer story of Fivetran and so will Jessica from Glean, Chetan from Posit, and David from Elementum. All right.
Posit and Snowflake: bringing data science to where the data lives
All right. Good morning, everyone. I'm Chetan Thapar. I lead product for Posit's Snowflake integration and I'm here to talk about what happens when you bring your data science to where your data lives inside Snowflake. So Posit is the company behind some of the most widely used open source data science tools. Thank you, Remy. So we are the creators of RStudio, Positron, Shiny, Quarto, tidymodels, and so many more. And more importantly, we have millions of data scientists who use our tools every week to be able to accelerate their productivity. And about 10,000 commercial customers. And these customers are spread across different industries. We have 52 of the Fortune 100, all top 20 global pharmas, and out of these, three joint customers, which are Posit and Snowflake joint customers, recently published their success stories publicly.
Number one, NMDP. They are scoring 42 million donors daily so that they can match patients with life-saving transplants. Dairy New Zealand, they are able to deliver economic forecasts for pharmas 4x faster with Posit and Snowflake. And Pinterest, I'll talk about that next. But first, what does Posit and Snowflake enable?
So at the core, you have these open source tools, VS Code, RStudio, Streamlit, Shiny, Quarto, Dash, that your data scientists already use, know, and love. Wrapping around that, we have our Posit team platform, which is made up of three services, Workbench, Connect, and Package Manager. And with this, in Snowflake, you get essentially an enterprise-grade platform out of the box. You know, centralized administration, automatic credentials, one-click deployment, inherited Snowflake governance across all of the stack. And now with governed AI agents, we have agents for exploratory data analysis, for data science coding debugging, for cell service analytics, all of them powered by Cortex-hosted LLMs. So these are governed AI agents. All of this runs on Snowpark container service in your account, and we provide a fully managed service. So everything from data scientists to decision makers, from exploring, building, iterating, developing, deploying, to interacting as a business user, all of that happens in the Snowflake boundary.
Pinterest customer story
Pinterest, 5,000 employees, a billion pins, and a people analytics team, which was sitting on some of the most sensitive data in the company. Every service cycle, they had about 15,000 employee comments, and they wanted to do some serious data science with it, but they were blocked. Three challenges. Number one, security. This was deeply personal data about strategy sentiment, about management sentiment, which could not leave Snowflake. This was PII. Second, a workflow problem. The current process to get to the data was five-step process, and each step was friction and potential security vulnerability. Third, a standardization problem. Each of the analysts was using a different tool, so the output, even on the same data, was different. So the confidence of management was eroded in the actual output that they were getting from the data science team.
They solved this by deploying Posit Workbench as a native app on Snowpark container service directly through the Snowflake marketplace. How does this look in practice? It's like your data scientist goes, opens a browser, which is hosted on Snowflake, picks their IDE, and that's it. They just get started. They are seamlessly connected to Snowflake. Now they are seamlessly connected to the AI agents, and they can just be productive.
So for Pinterest, it meant there was no data movement. Everything was out of box. The analysts were productive on day two, and zero ops in terms of any management overhead. But more importantly, it was the capabilities that unlocked for them. So before we talked about a workflow being fragmented, manual, and non-secure, after they could do serious statistical analysis, they started doing driver analysis, specifically relative weights analysis, to see what are the real factors, what are the real levers, which are defining and which are moving the employee sentiment. They could do structural equation modeling to see what is the causality chain. Is it the management sentiment which is driving the strategy sentiment, or is it the other way around? And all of this, they had 30,000 survey comments for which they had an NLP preprocessing pipeline, all within the product. More importantly, all of this was auditable, reproducible, and validated. You take the same models on the same data pipeline, you run it, you get the same result. And that matters when you're doing quarter to quarter analysis and when you want management to make decisions in which they trust.
More importantly, all of this was auditable, reproducible, and validated. You take the same models on the same data pipeline, you run it, you get the same result. And that matters when you're doing quarter to quarter analysis and when you want management to make decisions in which they trust.
And I was just talking with a couple of people in the audience. This is a bit of a side, but now we are taking the same deterministic models and putting it in our AI agents. So you have LLMs orchestrate the whole workflow, but you have validated methods, deterministic skills that we are adding. From IT perspective, they basically just said, it is set it and forget it. They didn't have to babysit infrastructure, manage upgrade cycles, and so on. So they loved it.
Insurance customer story
Okay, so I talked about Pinterest, but this is not just for that use case. Here's another customer in insurance space. They had 28 propensity models for marketing. And before Posit and Snowflake, they were running on a legacy platform. And it took them, for each model, about 24 hours to do build and scoring. After Posit and Snowflake, two hours. That's an order of magnitude improvement for the model pipeline. Even more interesting for the actual inference for the scoring, they went from two hours for each model scoring 11 million rows to 24 seconds. And that was using our orbital open source packages. For the POC, once they identified the workload, it took them three days to see the results because it was just so apparent. All of this had operational savings, but the most important thing was, now earlier they were doing annual model updates when it took 24 hours for one model. Now they could do monthly model refreshes. So for 11 million of their members, they could do fresher targeting, and that meant better ROI on the same marketing spend.
All right, I think we have a quick demo here. Let's see if this works. Actually, I don't think it works. So I'll have to show it to you later. Please come to me. I think there's some access permission here. So I'll just leave you with three things here.
I think the demo basically goes, you go into Posit team, you have an AI agent that's running. It explores data. It builds visualization. It builds a machine learning model. You go and ask the coding assistant to build an application based on that. You deploy it. There are two people who interact with the application using natural language, and governance walks through all of the stack. But I can show that in detail later. But anyways, Posit and Snowflake Advantage, three key things. Faster time to value, where our insurance customers saw the results in three days. No ops headache. The platform admin called it set it and forget it. And improved productivity. That's with AI-powered data science tools that just work out of the box, powered with cortex alleliums, and a seamless development to deployment loop. So if any of this resonates, please find me. And I also have my colleague Greg there. And my information is on the screen. Thank you so much.
