Snowflake on Snowflake: How and Why We Built Our Marketing Analytics Ecosystem

Lourenco Mello: Hi everyone, welcome to today's webinar. We're gonna go ahead and give people a minute or so to log on and then we'll go ahead and kick things off. So hang tight and we'll be right back with the content. Okay, looks like we got a good number here, so we're ready to get started. First and foremost, welcome everyone to today's webinar. We're really excited that all of you have chosen to join us today for this exciting session. We have really interesting content to share with all of you that will cover how our own team here at Snowflake has used Snowflake technology to build our own marketing analytics ecosystem as the foundation of many of the marketing use cases that we've deployed internally.

um i'm thrilled to be sharing this content with all of you today my name is lorenzo mello i'm a product marketer here at snowflake and i'm fortunate to be joined by two awesome speakers today from the snowflake business intelligence team and they'll go into details of their implementation and as well as some of the impact that they've been able to drive from from their implementation and the approach they've taken before going into the content itself just to touch on a couple logistics housekeeping items Number one, this call will be recorded, and you'll have a copy of this deck and recording shared with you upon its completion. So feel free to just enjoy the presentation today and just know that you'll receive a copy of the session after the session. And then number two, on the ON24 console, you'll see a Q&A box, and we highly encourage you to please go ahead and ask some questions.

At the end of the call, we'll cover as many questions as we can. We have a team of people standing by, Guan and CJ here, who I'll introduce in just a second, that are happy to answer your questions. So please ask away if you feel compelled to do so. And so without further ado, I'd love to introduce our two speakers today. We're excited to be joined by CJ and Guan from the business intelligence team here at Snowflake. So CJ and Guan would love a very brief introduction and what your role here is at Snowflake. CJ, let's start with you and then go to Guan.

Carl-Johan Wehtje: Sure. Thank you very much, Lorenzo. I'm CJ. I lead the business intelligence side for our marketing team here at Snowflake. I've been here for about four and a half years and very happy to be here today and talk a bit about what we've done and what we've achieved so far. But before getting into that, over to you, Guan.

Guan Wang: Thank you, CJ and Lorenzo. My name is Guan Wang. I'm the Global Director of Marketing Intelligence, Snowflake. I've been with the company for about two years. Prior to Snowflake, I had over 10 years experience in revenue strategy operations and analytics across a full customer lifecycle, marketing sales and customer success in high growth staff, as well as Fortune 500 companies. Next, Lorenzo will introduce agenda for our webinar today.

Lourenco Mello: Thank you, Guan. Thank you, CJ. We're excited for this session. Very briefly on the agenda, we have a few different sections of content. First, I'll start by highlighting a brief point of view of Snowflake for marketing. Next, I'll hand it off to Guan, who will cover a bit about the business intelligence team and the team's mandate. And then CJ will come in and present the bulk of the session, where he'll cover kind of three major areas, creating that data foundation and some of the steps that the team took to get there, how they've enabled collaboration through the right tools and dashboards. And then lastly, CJ will share a preview of what he and the team have in store for the future. And finally, as we discussed, we'll wrap up the session with live Q&A from CJ and Guan. So feel free to ask those questions in the Q&A box.

But first, you know, I like to start with this pretty simple slide that covers a bit about the term data driven marketing. We hear that term quite a bit. It's becoming kind of interchangeable with this modern marketing era and generally where marketing is headed. But at the same time, because of the somewhat at times over usage of the term, it's also become kind of a catchphrase at times. So I think it's important to explicitly tie it back to what marketers and businesses really care about. And it's important to note that when we refer to the marketing persona in an organization, we're actually working with and speaking about one of the most data-driven personas in any organization, one that really has seen the value of data and worked with data and really depended on data for quite some time.

And so this is a pretty plain view, but one that I think is helpful to kind of re-anchor everyone on some of the really tangible things that marketers are aiming to achieve with data-driven marketing. And it's essentially a means to an end for marketers to do things like Getting deeper customer insights from their data, so leveraging BI and analytics tools for insights and visualization. Ultimately, being able to combine that with machine learning to help deliver the most relevant and timely campaigns for their audiences. The middle column there, getting timely measurements on how to evaluate the impact of their campaigns and marketing spend. Marketing spend is one of the most inefficient slices of the budget. I think it's an estimated 30% to 40% of marketing budgets are wasted due to ineffectiveness of campaigns.

So that measurement and visibility really becomes critical to optimizing marketing. And then they're looking to effectively activate their marketing data to power things like segmentation models and recommendation engines. And this is as a means to an end to improve the timing and the impact of those customer engagements. And so these are just some examples of concrete things that marketers are doing and that are aiming to do every day with data as that foundational centerpiece of their marketing strategies. There are many, many more obviously, but these things I think are really core to what marketers are aiming to do. But the reality is, and this slide is an eye chart by design, is that doing these things can be complex, especially if, and nearly impossible, honestly, if you have the wrong technology foundation.

And the reason why it's becoming increasingly hard to accomplish this is because of the explosion of data, which in turn is generating an explosion of data silos. And so this graphic that you see here, which you may be familiar with, I think provides a good representation of really the breadth of options that exist today and the fragmentation that currently exists in the MarTech space. So in a little bit more detail, this is a graphic from the marketing technology landscape, so from Chief MarTech. which they update on an annual basis. And this version that you see here is released just from a few months ago. And you can see each of those dots that you see on the slide there, or maybe I should say that you can't see on the screen, it's an individual tool, technology, or application that is available today as part of the MarTech stack and creating its own data stream.

so no need to really explain why this is so challenging to navigate for marketers um but in 2011 there were 150 logos on as part of this study and in 2022 just a bit over a decade later we're talking about around 10 000 logos that's just an unprecedented growth in the ecosystem that on one hand shows that deep fragmentation and complexity but on the other hand it shows the incredible innovation that this space has has experienced throughout the last decade But ultimately, each of these applications is generating new data that needs to be stitched together in order for marketers to deliver that consistent experience that customers expect today and really demand today. And to help show this, I saw this stat from one of our partners, Fivetran, that I put at the bottom there. On average, an organization uses 110 SaaS apps.

So that's just a pretty amazing stat that stuck with me around, again, the complexity of stitching all of that together and creating a unified data strategy. and in a world prior to snowflake having to bring all of your customer data together was was really difficult and honestly a very cumbersome task the consequence of this is that critical data about customers sat in various silos and it actually sits in various silos today if you have the wrong technology stack it may live in crm it could be inventory data it could be your point of sales data service data product data, and this fragmentation of customer data within an organization makes it really, really difficult to implement a full view, a unified view of your customer data, which is what we commonly refer to as customer 360. And that's a foundational step on your way to data-driven marketing.

And the impact of that is a very heavy lift for your engineering resources that have to bring all of these data sources together in a single location. and really maintain that data as it evolves. So you can just imagine how complex, how expensive and honestly how daunting that task really is for organizations. And so the goal of this is actually, the goal is to get to the exact opposite state, right? It's to build that customer 360 or this complete integration and unification of your customer data where all of that data sits together in a single source and is made available and readily accessible, I should say to your teams.

uh i think one key word to mention here is enterprise-wide data strategy which spans beyond just customer data and into things that that impact the customer experience so for example if you're able to adopt an enterprise-wide data strategy you're able to implement advanced marketing use cases that take into account inventory for example so if you have surplus inventory you might be able to push out a promotion or an ad campaign to push out that inventory that's really key really making sure you have that enterprise wide view of data But in short, the high-level message here is data silos make it extremely challenging for an organization to adopt a single view of the customer and therefore truly deliver on that potential of what we refer to as those data-driven marketing outcomes that ultimately drive your business forward. But the good news is we have Snowflake to address this, right?

And I'll just take a few minutes to introduce Snowflake in the data cloud. At the highest level, Snowflake is exactly the platform that allows organizations to break down those data silos and really enable marketers and really any persona across the enterprise to tap into that enterprise level single source of truth. I'll cover kind of three pillars that I think are worth mentioning here to paint a more vivid picture of the data cloud and how it can help marketers. The first one on the left there is what we've been driving home, which is this fact that Snowflake is one platform for all your data, right? That means you can use data from any source, whether it's internal or external. That includes first-party data, second-party data, third-party data sources, or even different data formats, whether it be structured, semi-structured, or unstructured.

And it could really ultimately be streamed data, batch data, or shared data. All of that data can be harnessed in a single location. The second pillar that you see there in the middle is around instant governed access. So what that means is that you can work with fresh data and reduce data latency with data sharing and really get that instant access to data as it's updated. So to provide a more concrete example of where data latency exists in this scenario, is one that many of us have experienced, which is when we receive ads for an item that we've already purchased. That means that somewhere in the back end, the data is not unified, and there's data latency between those two systems.

And then also in the middle here, critically, I think it's important to mention to this audience is on privacy, which is, I think, top of mind for all marketers, particularly with the existing and upcoming regulatory changes. With the data cloud, just the simple fact that you're storing all of your data in a single place, It means that there are less copies of your personal identifiable information sitting in different silos. And so by definition, you're actually increasing the control and security and really in governed access to that data within that data set. So now the right side there, just to touch on activation, you can collaborate in the data cloud by enabling activation and analytics platforms to access your data. It's an elastic platform. So think of that in the sense that you can scale up or down to meet the needs of your peak intervals and really only pay for what you use.

And then lastly, we depend on our ecosystem to make these marketing use cases come to life. And Juan and CJ will cover that in just a second as well. And with Snowflake, you can seamlessly integrate with the best of breed engagement engines, marketing clouds, great partners across data integration, data modeling and transformation, BI, activation engines. really this is you know a very important point is the data cloud as a platform really depends on that best best of breed performance technology from our ecosystem but the real power of snowflake isn't limited to just first-party data the real magic i think comes from the ability to seamlessly collaborate across your ecosystems in really a seamless and privacy compliant way so to break this down a bit from a business lens this means you can start by having You have seamless access to all your organization's data.

This is what we talked about, that single source of truth. It means eliminating separate data platforms or really eliminating the need for any data marts that may exist by department, by subsidiary, by geography. All of that is brought together. But then beyond your four walls, you can actually have instant access to data from your suppliers to drive supply chain efficiencies, for example. or your partners to analyze joint go-to-market campaigns, or your customers really to help your company deliver exceptional experiences to them. And the ability to have access to that data without any movement, while it's all sitting in the data cloud, while maintaining that kind of highest privacy standard is a really powerful concept, and really one that can generate tremendous value for you.

And then finally, on the outer ring you see there, data cloud gives you access to Snowflake Marketplace, there are hundreds of companies that offer their data or data services again directly in the data cloud without ever changing hands so think of data sets that can be used to enrich your own data whether this be shopper data that you can query trends or household data weather data really any kind of data set the diversity of the data set that's available is really unique that can help kind of complement your existing data and really extract the maximum value for your data so this is one of the true benefits of working with snowflake in the data cloud And so with that, I will hand it off to our speakers. I'll wrap up my part and who will cover the main event for today. So, Guan, I believe this is over to you.

Guan Wang: Thank you, Lorenzo. Hello, everyone. I want to introduce quickly about the marketing intelligence function. Our team is a global diverse team with a bold vision and mission. Our vision is to predict real-time ROI to dynamically optimize all Snowflake marketing programs. And ultimately, we want to disrupt the aged B2B marketing analytics practices. Our mission today is to build scalable solutions models to enable our Snowflake marketing to become the industry's most insight driven team. And this one is also a core of Snowflake marketing pillars. I also want to really quickly introduce about our market analytics ecosystem, which is fully powered by the data cloud. We use all Snowflake major workloads, including collaboration, data engineering, data warehouse, data science, and machine learning, and applications as well as data lake.

All the data is ingested into Snowhouse with over 1,000 tables for marketing. We pull data from ZoomInfo outreach heap intricately via data sharing, you know, and with a few button clicks, and then that data magically appears in the Snowhouse. For digital ads platforms, including LinkedIn, Facebook, and Google, we use both five chain as well as some custom scripts to ingest data into Snowhouse. Before we make use of any of data, we did a lot of data transformation to make sure the data is in the right format. And after we've done a lot of data modeling, we then built advanced data science and machine learning models, including attributions, scoring, forecasting, and segmentation. And with those advanced data science models, we build our own marketing-specific feature store that can be shared across departments.

Based on those predictive models, we send the results, business results to and consumption platforms, including Tableau, Southspot, Snowsight, as well as Salesforce. With this data-rich ecosystem, we have driven our revenue growth and optimized our marketing investments. Next, CJ will share Snowflake's story, how and why we built our marketing analytics ecosystem.

Carl-Johan Wehtje: Thank you, Guan. So I want to start by touching a bit on the second to last pillar from Guan's last slide, which is essentially the consumption layer or the consumption use case, because this is essentially where we enable collaboration and invite our business partners to start their interactions with the data. And typically I split these types of interactions into three different categories based on their terms of structure and format. In terms of the business visualizations, we use ad hoc analytics typically for when I or one of our analysts on the team do one-off reporting or when I dive deeper into some specific data sets or analytics work jobs. And we typically do this by just writing some pure SQL or using simple tools like Snowsight to quickly and simply visualize and see the clear patterns and trends of the data.

However, when we want to do other things like do more formal reporting on standardized metrics or invite the wider audiences, that's typically when we move over and use more dashboards in tools like Tableau. These dashboards make it a lot easier for, for example, business leaders to, at a quick glance, see their performance at any point in time or how we're tracking quarter of a quarter or on a year-over-year basis. Dashboards are also a great way of framing a clear story, which is even more important when you're sharing with wider audiences. where not everyone may have the full context or knowledge of what are the key metrics or data points that are of importance.

And then lastly, because people interact with the data very differently, we also want to enable a good self service reporting environment so that we can give the data to the end users and give them the power to investigate and look at the data wherever they are and by whatever needs they have. Dashboards are great at telling a clear story, but we also need some flexibility and really empowering our end users. And so with that, I then move on and I want to start by next going through essentially what are the steps we're doing to take the data from all of our different applications and tools, putting it into Snowflake, how we then process, clean up the data, combine different data sources, and then how we on the data team side, collaborate on the code, how we version control everything that we do so that we can enable good documentation and tracking of how our models evolve over time.

So first step and anything that, or in some ways the most important step to happen is actually getting access to the data. And we do this, we ingest as much as we can of our data into Snowhouse, which is just our internal instance of Snowflake. Snowflake supports multiple ways of bringing the data in. My, by far, favorite method of bringing in data is to use the data marketplace or direct data shares. Just because it's so easy, fast, and doesn't require any work with fragile APIs or reliance on our IT team or engineering data pipelines.

However, when this is not available, we typically use a combination of custom scripts on Snowpipe, Or if one of our partners like Fivetran have established connectors for specific tools or platforms, then we'll tend to use them as it's a really quick and easy way to, with a few clicks, authenticate against any of the platforms and then seamlessly and almost instantly the data appears within Snowflake. However, Just want to take a little bit of a closer look on the data sharing as I think this is a really powerful tool. One example on our side is that we use heap analytics for our website tracking, basically seeing who comes to our website, what do they do, how do they go between different pages, and what do they sign up for, for example, on the website. operates on the Snowflake Data Cloud.

And with that, they offer the ability to access all of our heap data in Snowflake directly through direct share. And getting access to all of our data means that we can incorporate it with our data sets to build better real-time targeting and insights. And with a direct share, only the recipient account, so in our case, only us can discover and see our data. And it means we always have access to the most current data. Updates are made automatically without any of the data actually having to move. And basically getting this set up and going was literally a matter of filling out our account name within the heap platform. Uh, and then all of our data just appeared within Snowflake. No need to set up any engineering pipelines or engage IT or even reaching out to heap to get any specific API access or anything. It's just super smooth and fast.

And then next step after obtaining the data, we want to get the model right. And while one can expose many of the raw data tables directly, we always perform some operations to clean up the raw data and join in additional data sources so that we have captured a more complete picture. And also, this makes it easier whenever we run queries, whether that's ad hoc or through dashboards, we already have everything cleaned up and aggregated to an appropriate level, providing much better performance whenever we run those dashboards. And having these good efficient models also means that everyone in the organization, regardless from what tool you're working or looking at the data, we're all using the same models and hence can expect to get the same type of outputs and results. And in order to achieve this, we are heavy users of an open source technology called dbt Core.

DBT core is used by our data teams for all of our models and essentially is where we define what a table and view should look like. So write the SQL statement, make all of the joins and define how do we want our end table to look like. And with this, we can define the relationship between all of these different tables and views. And if there's cases where we have multiple tables of different aggregation levels, we can also define how they should later on be connected and joined later on in a reporting tool like ThoughtSpot or Tableau. And we use Airflow essentially to manage these dbt jobs so that we can define however often we want to update these models and tables so that we manage the balance between always having fresh data and only running the models as often as we need for performance and cost considerations.

And another thing that I think is really cool with this modern data stack and finding nice shortcuts provided by partners in dbt is something called dbt packages, which is essentially prebuilt models, which can really help with cleaning up and preparing raw data sources. One example of this is we do a lot of reporting on our digital advertising data and where we use the usual suspects of Google, Facebook and LinkedIn to run digital ads. We ingest this data with Fivetran. And Fivetran has actually built dbt packages, which automatically cleans up all of those raw data sources and outputs them into one model, which means I, as a data analyst, don't really have to figure out and really understand all of the raw data sources and the nuances between the different platforms.

and instead can immediately get a cleaned up, normalized reporting set with all of the advertising data and can move on to the more complex sides of calculations of either joining in other more proprietary data sets or get quicker to expose this data to our end users so that they can start drawing insights from it. And then next thing I want to touch on is that with so many different data sources and models, it becomes really important to understand the data lineage. As a data team, we need transparency on where data is being pulled from, what systems are ingesting them, and how it's being transformed along the way before we actually put it in the hand of analysts or business folks. And this is really important in order to scale an organization and to be able to maintain trust between grown teams and departments.

Hence, all of our models are versioned, controlled and peer reviewed so that we have safeguards against introducing production breaking changes, writing bad code or implementing incorrect or bad assumptions. And I also think this adds a layer of documentation. Because while I consider documentation hugely important, it can also be extremely time consuming, hard to formulate, and inaccessible and hard to keep up to date. As an analyst, I therefore really appreciate having direct access to Snowflake's entire data team's code base so that I can, through GitHub and dbt, go in and see what kind of assumptions, transformation and logic our teams are using, when they made those assumptions or when they changed the code and how that's been evolving over time.

And building on this, we're also starting or introducing more use of a tool called Alation that takes a lot of the metadata from Snowflake to automatically outline data lineage, the content of different tables and data sets, And making this searchable both to data teams and the wider business teams so that anyone can go in and find where a data lives, see who's the owner of those data sets and who's been working on it. So that they know if they have any deeper questions or seek more help, they know who to reach out to. But with all of this essentially foundation laid out, I now would like to move on and talk a little bit about how we actually make this data interactive for the end users and how they draw insights from it. First off, I'd like to cover a little bit about Snowsight because as an analyst, I really enjoy using our own Snowsight tool.

It's a very quick and simple querying tool where it's easy to just explore data, see the simple trends and patterns, and even set up some reports and very easy dashboards that can give direct access to either raw data or simple data reports for also our end business users. It's a great starting point when developing more advanced reporting because it's directing SQL and I have full control on everything. I can very quickly try things out, iterate and evolve on it. And it's also easy for business users to go in and see whatever they need if it's just a simple request. And they can even apply some simple filters or change time and date ranges for what data they're looking at, for example. However, if we have some more formalized reporting or it's a wider audience, then we typically move on to other areas, such as using dashboards instead.

Because when you have wider audiences or talking more with executive teams, that brings up a slightly different set of requirements. I personally really enjoy Tableau and think it's a great way to create dashboards with a clear narrative that in a formalized way brings up the key features and metrics that I know are relevant whenever looking at these datasets. It helps in making it crystal clear for whoever uses the dashboard, what we're tracking, how we're tracking it, what are the key KPIs, So that when we share it with a wider audience and not everyone will have the same contextual knowledge of what we're working on or what we're optimizing for. But we can really bring this up and make sure that everybody looks at the data in the same way and gets the same kind of interpretation from it.

And as mentioned earlier, this is also a great way to bring in different data sources or Essentially what I wanna say is we have, I said, hundreds of different tools and applications, and many of these different applications, they only have point to point or no integration between each other. But with the Snowflake data models, we have the ability to basically present all of these different data sets in one report or one dashboard here, making it much easier for people to get a complete picture of what's going on. And because any of those source applications often have more narrow restrictions or access permissions, we can also expand the group of people that have access to the insights from the datasets whenever we expose it in, for example, Tableau and use the dashboards. But then everybody's needs are different.

In order to allow for more individualized reporting and allowing for people to really drill down multiple layers into the data or just creating their own report, we use a tool called Fortspot. And I think Fortspot is really cool because it's a really easy and intuitive way for users to simply search for whatever data points or metrics they're interested in, And that way they can create their own reports. They can, by any metric, click on anything and drill down and slice and dice the data by whatever parameters they are interested in. And then they can also save these searches and create their own reports and share it with wider teams.

And because I said before, we have built clear models And we essentially use the same model for, regardless of whether it's the ad hoc analytics, Tableau dashboards or full-spot dashboards, everyone is still looking at the same data and getting the same results and outputs. But we have a very flexible model where we can bring the data to end users in the way that meet their needs. And We know that we have a huge potential to democratize data, and that's why we really wanted to build out this ecosystem so that our team can become the most data driven in the industry.

And it makes me really happy every time I see how many of our marketing team on a daily basis use our dashboards, reports and data science models to track their performance and create strategies for how we should run our business in the future and really contribute to more sales pipeline and in the end, more revenue for our company. But With that, I thought I'd also like to go through some of the key challenges and learnings that I've had during my time at Snowflake. Obviously, everything isn't always simple and straight sailing, and I just thought I'd share some of what I've taken away from my time here. And I think the first one, which should be a fairly obvious one, is that data quality is always a hard thing, and there really isn't any such thing as perfect data. But having access and the ownership of all of your data is a really powerful way of actually dealing with this.

When you have access to all of your data, it's much easier to actually identify the data quality issues that you may have, And because we have the ability to operate on that data, we can much easier fix any problems that we discover by either bringing in related data sources to replace whatever data that wasn't good enough, or create calculations and transformations to proxy out any of those issues. Secondarily, I think on a more behavioral level, we also really noticed that there's many different ways that people want to interact with the data and people have different levels of data literacy. And this one really calls for a flexible data approach where we can expose the data in whatever way makes most sense for the end users to really help them drive their business.

And this is again where I think creating our models really makes it easier to serve those very wide and different types of data exposure models. And then lastly, our third point is that regardless of how well people know the data or how much context they have, documentation is still a really important thing. As I said, in order to scale and maintain trust between different teams. However, it is hard to keep it up to date and there's a lot of different concepts. So it's really important to keep it simple. But this is again where I think having access to the data and having a clear tracking of the transformation you do and the data lineage, being able to leverage metadata in Snowflake and tool speculation makes all of this a lot easier.

And this brings me up to my last point, and essentially that what I think has really helped us solve almost all of these issues and enable our full ecosystem that we've created is really the way in which we model the data in the Snowflake layer. And I think one of the examples here is my key takeaways from my time at Snowflake. A couple of years ago, we had a different BI tool where we did Most of our data modeling and data transformation kept our business logic within that BI tool. But this had a few downsides because it meant every time we ran dashboards, it took some time to basically do those calculations. And especially it was hard because all of the logic lived in the BI tool.

If we wanted to run any models anywhere else, or just do ad hoc analytics, we every time had to recreate those models, which makes it hard to follow basically what logic is correct one, what happens when things changes. But when we instead moved out, started modeling the data with dbt and had better tracking, It really enabled us to get a much more flexible approach. We got much better performance because most of the heavy and time and compute intensive calculations were done much earlier, so created an easier environment for us data analysts and data scientists to work with the data and also created a much better environment for the end users because they got their results quicker. and had a more flexibility in where they actually want to see that data. And I think that's essentially the key takeaways that I wanted to cover.

Looking ahead, this is really just the beginning of what we're doing. We're working on many other things, such as creating even better ways of, or even more flexible ways of people to interact with the data, wherever it makes sense for them, and building out integrations between the tools so that make it so that we make it easier for people to start off some data exploration in, for example, Tableau, and then allow them to switch over to Fault Spot to drill down into the data deeper. And then we want to also make it possible to push this data into other tools like Salesforce and Outreach so that, for example, the Fault Team can take action on their data there. So that's just an example. But with that, I think that's the end of the part I wanted to go through. And I'll hand it back over to Lorenzo to take us into the questions and then close us out.

Lourenco Mello: Hey, awesome, CJ. Thank you very much. And thank you, Guan, as well, for the great overview and a great presentation. Before we get to questions, and I see that there's some great energy in the question, the Q&A box, so keep them coming. To our attendees, please take a look at the resource box. We have lots more Snowflake on Snowflake content and sessions that highlight some of the work that, some of the great work that this team and other teams have done in implementing our own technology within our walls. Please also be on the lookout for an upcoming session that will likely be in the fall for how this team has implemented Snowflake for the marketing attribution use case. So I think there's tons of interesting content coming down the funnel there. And then CJ also mentioned the work he's done with ThoughtSpot and how impactful it's been.

We do have an upcoming demo with ThoughtSpot and Supermetrics that looks at how organizations can use Snowflake and their technologies to optimize marketing spend. So lots of great momentum and upcoming goodness for you to keep an eye out for. And fortunately, we do have plenty of time for some questions. And so I will go ahead and open the Q&A box here. And I'll start with one that goes for CJ here. And it's related to handling disparate data granularity. And the question is explicitly, how do you handle this disparate data granularity in the data cleaning process?

Carl-Johan Wehtje: So I think that's really where generally when we build our models, we keep it at the lowest level of granularity that we need. But it's also depending on it's very use case dependent. And it dbt allows for an easy way where we can actually switch out and use different cables for, for example, if we need different levels of granularity. So Basically, we have a lot of flexibility on that. I'm not really sure how else to address it. Guan, do you have anything you want to add to it?

Guan Wang: I think if we want to track the change, I believe... ...CJ will use... hybrid approach, how we track all the changes across our foundational layer of the data.

Carl-Johan Wehtje: Yeah, I think that another key point is also to bring up in, but Snowflake is really good at handling a lot of data and essentially because of that, generally we find that it works really well to keep the data as long as we have it clean and don't need to we typically like to create large monster tables where we have all of the data that we need and that makes it really quick and easy whenever whatever granularity level we want to look at it because Snowflake purely does a quick query maybe it's a simple aggregation of counting by a parameter or something like that but because we're because it's just a simple table, it all runs really quick and easy. Not to say that Snowflake don't handle many different joins and complexities. That also works great, but I just find it easier if we put it all in a format that works and has everything that we need.

Lourenco Mello: Perfect. Thank you, CJ. Thanks, Guan. Go ahead, Guan. You want to add something?

Guan Wang: Another quick add here is if we want to really check all the detailed data points or fields, Snowflake has a great functionality. You can snapshot all the fields in your Snowflake instance. And basically, if anything changed, then it will be captured in that snapshot table.

Lourenco Mello: Perfect, thank you both. We have another question. Juan, this one I'll ask you. I didn't see Google Ads or GA or TikTok on the ingestion portion. Does Snowflake integrate with these platforms?

Guan Wang: Yes, because the space is limited on the slides, so we didn't show all the data sources. But we do ingest hundreds of data sources into marketing database, basically marketing schemas. And CJ has a lot of insights, maybe not in this webinar, but we can share with you all the data ingestions we have done at Snowflake. Awesome.

Lourenco Mello: CJ, this one you mentioned you really like working with Tableau, for example. Do certain BI tools work better with Snowflake than others?

Carl-Johan Wehtje: I wouldn't really say so. And I think it really comes down to what you want. In many ways, I would say that the BI tool is really only the layer in which you expose it. All of them have certain characteristics and some pros and cons. what I think is really important is getting access to all of the data and scanning the data into Snowflake and then creating those efficient models on top of it. That's really what in my mind, what is important. And then you can essentially just expose that data in whatever BI tool, and they have slightly different visualizations. But that's more a matter of taste and how do you want the end users to look at the data and interact with it.

Lourenco Mello: Perfect. I agree with that answer for sure. One here on DBT Core, CJ. This one you covered, I think. DBT Core already contains DAG functionality. What does Airflow allow on top of that?

Carl-Johan Wehtje: So I think that DBT has one offering where they have a UI, and that's essentially the DBT offering or product. and that's where you can run the dbt models natively within dbt. We just use the dbt core open source code and then functionality, and we define the DAGs within dbt core, but then essentially the scheduling and running of those DAGs and the jobs, that's what we schedule within Airflow. Hopefully that makes sense.

Lourenco Mello: Makes sense. And Guan, I think this is an important one because we hear the question all the time around how the teams that are closer to the data work together with marketers. I'd like to just kind of ask you from your experience also across the different use cases your team has built. Do you build, excuse me, do you build data models based on business metrics that marketing team wants to visualize in Tableau? Or do you build the model first and then have the marketing team build their metrics in Tableau? And TJ, feel free to chime in as well.

Guan Wang: Yeah, so our team has a business requirement intake process. We have a framework we want our stakeholders to fill before they come to us with a project request. And that basically set up the foundation between requesters and our team. And then requesters usually come to us with a set of, here are the key metrics they want to check. and then our team internally will discuss is this something doable or this we need to come back with our own with our recommendations so usually there's some kind of negotiation and some kind of you know recommendation there's a definite collaboration process in between and you know for our end products whatever the data science machine learning model recommendation engine or even attribution model and then all the dashboards CJ's team has built.

It's all based on the collaboration and we want to make sure the end product is actionable and usable by business stakeholders. So hopefully that answers this question.

Lourenco Mello: Perfect. CJ, one question here I think is relevant for the audience around elaborating on the documentation piece you touched on. This person, and I'm sure it's common to others listening, would like to share a catalog of what's available with executives. They're looking for a different and better option as a tool than a simple Google Docs. Do you want to talk about your experience as CJ in terms of the documentation piece?

Carl-Johan Wehtje: Yeah, sure. And I think this is something that we're actively working on. But as I mentioned before, we're rolling out the use of elation, which I think is actually a very good tool for doing this in that It very seamlessly reads all of the metadata from Snowflake and then brings in all of the different databases, schemas, tables, and views, and shows what are the data sets or what are the columns and fields that exist within that. then whether you want to expose that to data teams or your executives, that's very easy within Elation. They can search for it. It's possible to make relationships between the fields of different tables. So if you, for example, have the same measure in multiple different tables or the same dimension or column in multiple different ones. You can also show the data lineage. So how does that flow through? And is it the same definition for all of them?

So there are many different partners that do this, but I've had a very good experience with the nation.

Lourenco Mello: Perfect. It's great. In Guam, I think this one is kind of a, higher level question that's relevant for everybody here, and it's around kind of the return on investment. You've done several of these projects internally at Snowflake. The question is, how have you shown your return on investment for the new ecosystem to justify the changes that you have made for this enterprise wide approach?

Guan Wang: For us, to be honest, the cost is all the data. If we're talking about marketing investment, I think that's a broader question. We do have our marketing attribution model that can answer how much investment we made in marketing and how much sales pipeline we generated and how much you know, sales bookings we influence, et cetera. So that's definitely advanced attribution models. I think we will feature it in about three months, two or three months. And in terms of our team, so, you know, our team definitely invests in all those tourings. And the end goal is not to, you know, generate the revenue directly. The end goal is to make sure every single marketer, including Lorenzo here, to be data-driven. Remember that our mission is to enable the Snowflake marketing team to become the most insight driven team.

And with that mission, we're going to build every possible solution that's cutting edge, that makes everyone data driven. So that's our return, I think, for how we measure our return on investment in my team.

Lourenco Mello: And Guan, just to pile on what you said, one of the things that we hear customers that are doing this effectively talk about over time is, it takes some time to get that customer 360 to build that unification view of all of your data at some point. And then you really start seeing the use cases that are able almost to plug out of that data, start generating that ROI. So it's a journey to get to that full ROI. And most customers that we talk to, because it's an ongoing iteration and continuous improvement, are on that process now. And so I think that first step is to get that full view of your data unified, as CJ covered.

Guan Wang: and then build that foundation and then start adopting the use cases that are most relevant to your specific business needs and you're going to start seeing that roi come forward yeah and lorenzo looks like while you and i was speaking there's a live question coming from an audience um that person asked whether we use a tool to build our roi logic um so we actually build our own marketing attribution model inside snowflake In the coming about two or three months, we will do a live webinar to feature our Snowflake-specific marketing attribution model. We definitely want to push our own marketing attribution model to the frontier of the machine learning and data science deep learning. Hopefully, some of the audience here today can join us in that live webinar.

Lourenco Mello: Perfect. Thanks, Juan. Let me take one here. um cj regarding using the data do you have an audience manager or some other service that you have hooked to snow house um no so we don't have that right now um well so i'm just gonna

Carl-Johan Wehtje: start off not exactly care what's meant with the question, but I'll interpret this audience in terms of digital advertising or creating basically our different segmentations of the database. And with that interpretation, that's something that we have been building and working on right now.

defining all of our audience and segmentations of the database within snowflake and then from there pushing it out to whether it's our external digital advertising vendors or our marketing automation systems like marketo or just putting it in salesforce to easily see the size of our database the engagement of people that we interact with um and and how we perform with them and i think We typically, I keep coming back to this, but I think it's really the core piece or one of the more important ones is to really do this within Snowflake because that gives you one place where you can handle it all and then much easier from there, push it out to any of the other end systems.

Lourenco Mello: Great. Thanks, CJ. Juan, I'm going to go to you on this one, but I know CJ probably has some opinions here, and we got time, I think, for this one and then one more afterwards. If marketing rules are constantly changing, like lead routing, is it best practice to keep up with those changes in Snowflake or the BI tool itself?

Carl-Johan Wehtje: I think I can step in on this because I think it comes back to that. Sorry, Juan.

Guan Wang: Oh, yeah. So we do have a lead scoring process and lead routing process. So, you know, a lot of the logic actually lives within Snowflake. Snowflake is basically we have our lead scoring engine that happens every every um you know two hours and we push data directly from snowflake into salesforce and other other tools we do have another b routing tool um you know but overall i think the main logic lives from the top lead all the way to the bottom lead um so that's how we prioritize for our sales team

Carl-Johan Wehtje: Yeah, and I think I would add again, so especially when you have things that are constantly changing. I think that's that's one of the cases where it's even more important or even more powerful to basically track that within Snowflake, One, you only need to do it there and then you can expose it into whatever BI tool or other system that you're pushing the data to. And by having it in Snowflake, it also allows for generally a much better and easier way of tracking all of those changes and making sure that you have a latest version, whereas for example, creating a calculation or some logic in, for example, Tableau, that's just a table calculations. There's generally no way of tracking when that last changed and when it was last updated or what was the old definition. So with that, I think is really something that's very powerful to have within Snowflake instead.

Lourenco Mello: That's great. Thank you, Juan. Thank you, CJ. We are at time. This has been extremely helpful to understand how Snowflake is adopting our technology with some really positive impact, some really practical guidance, CJ, that you shared today. So thank you very much for that. I'm sure everyone found it extremely valuable to hear kind of the perspective of a team that's walked the walk and continues to do that. And as CJ mentioned, this is just the beginning. There's going to be more coming on this and more momentum built by this team and others. So thanks, everyone, for tuning in to today's webinar. We really appreciate your time and hope you enjoyed it. We hope everyone has a great rest of your day and we look forward to seeing you next time. Thank you.