The Data Stack Show - 118: Bringing Powerful Business Intelligence to Mobile with Zack Hendlin of Zing Data

Episode Date: December 21, 2022

Highlights from this week’s conversation include:Zack’s extensive background in the world of data and the genesis of Zing (3:02)Working on relevance, feeds, and ads at Facebook & LinkedIn (9:20)Exploring BI and queries on mobile devices (16:48)Reliance of input quality in data (23:28)Delivering a mobile-first experience in BI (30:11)Limitations of visualization on mobile devices (34:00)How BI tools interact with one another in Zing (45:21)The future of user-experience in consuming data (49:19)Final thoughts and takeaways (59:56)The Data Stack Show is a weekly podcast powered by RudderStack, the CDP for developers. Each week we’ll talk to data engineers, analysts, and data scientists about their experience around building and maintaining data infrastructure, delivering data and data products, and driving better outcomes across their businesses with data.RudderStack helps businesses make the most out of their customer data while ensuring data privacy and security. To learn more about RudderStack visit rudderstack.com.

Welcome to the Data Stack Show. Each week we explore the world of data by talking to the people shaping its future. You'll learn about new data technology and trends and how data teams and processes are run at top companies. The Data Stack Show is brought to you by Rudderstack, the CDP for developers. You can learn more at rudderstack.com. Welcome to the Data Stack Show. Today, Costas, this is going to be a really interesting one. We're going to talk with Zach Henlin from Zing.

And Zing brings BI and analytics to mobile devices, which is a really interesting concept. It sounds really simple, but anyone familiar with the world of BI, especially traditional BI, has access to probably exclusively to a desktop environment. And Zing is trying to bring that to mobile devices, which is fascinating.

I have so many questions. I think my main one is really simply why mobile? Do you really need to access these numbers on the go? What are the big trends that are driving the need for this? I mean, I understand getting notifications or looking at Salesforce pipeline generated for a B2B company, etc. Or sales for an e-commerce company. But actual BI with the ability to do some querying on the mobile devices is a pretty provocative concept, at least as far as my experience with BI. So that's what I'm going to ask. How about you? Yeah, it's very interesting to hear, you know, in 2022 about the need of BI to the mobile experience. We would all expect that this is like a solved problem already, right?

Because BI has been around forever, like mobile devices have been around for a long time already. But it seems like that's not the case. And we have someone here who's really passionate about doing these writes. I'm pretty sure that there are like many difficulties in delivering like the writes kind of experience on a mobile device for something that is so complex in terms of interacting with the visual side of data. So, first of all, I really want to chat with humans, see what are the difficulties of doing that. What it means to take dashboards and interactive dashboards and strip them down like to the size of a mobile phone,

right? And do that like effectively. Let's see what he has to say. I think that's going to be like super, super interesting. I can't wait. Let's dig in. Zach, welcome to the Data Sack Show. Thanks for having me. Great to be here. All right. Give us your background. I started a company called Zing Data, where I'm co-founder and CEO, and we make it so you can do BI on your phone. So we usually query data, visualize it, set up alerts, save questions, and collaborate with colleagues. And before that, I was VP of product and built out the data infrastructure at a company called OneSignal that powers about 10 billion notifications a day. If you get news or sports scores, it's probably coming through them. And then before that, I shipped books first work in speech recognition, actually their first mobile ads format, which was a big motivation for trying to make experiences with data better on mobile. And also worked on translation and newsfeed ranking and relevance at LinkedIn.

Very cool. Man, what a, you've done a huge amount. Well, give us the quick background on, so you've obviously had a ton of product experience and in doing that, I'm sure you've been exposed to, you know, all the different kinds of needs for data doing your job. Is that where Zing, the need for Zing came from? What's the Genesis story?

Or maybe tell us about when you sort of had the first idea that led you eventually to Zing. Yes. So the Genesis was I was working at this kind of Series B company at the time. And we had sales folks out in the field. And the cool thing was we had gone from literally one production database where if you wanted to run a BI or analytics query, you were logging into the server, running it on the production database and, you know, production traffic rates load up. Sure. And I was like, Hey, we want to know how many users are taking which actions, who we want to upsell to. Just like all this stuff, like a company, once you have users and some of them

are willing to pay you, you want to figure out how to optimize that. And so built out sort of the basic infrastructure with one of the engineers on the team and got super basic BI in place. And then the great thing was like salespeople were like, oh, great. This tells me who I should be targeting, who signed up for free, that we could be upselling. Marketing was like, this is great. This tells us which features people are using and what resonates with them. Product loved it because they could see, oh, here's the feature I built, but people aren't using the sub part of it.

And so it was great. But then we realized once we like open this box of insight that it was actually really constrained as to how people could use it and who could use it. And if someone didn't know SQL, if someone was on the go, all those questions ended up basically getting funneled to one or two people who could go write some SQL for them. And you'd bump up against the limits of like simple counts and sums really quickly with the drag and drop WYSIWYG type tools.

Yep. And so the genesis was we now had all these people wanting to use data. They saw how useful it was. And the capacity constraint was still like the data team having one person who was available and had a queue of a ton of stuff. We said, well, what is it that makes it so these people can't do it on their own? And it was, it didn't work where they needed it to the tools weren't flexible enough and it was pretty much single player. So it was really hard to start from somewhere.

Someone else had done most of what you wanted and just tweak it. And so that was kind of the genesis. Plus the other sort of inspiration was a really bad meeting where it was sitting with other folks and we were trying to decide where we like hire salespeople or make certain investments and during this meeting for like an hour and we were kind of guessing at where are we growing free users fastest, where are the biggest new sources of revenue? Where's the biggest churn risk? Just like all these kinds of questions and everybody had their phones. Nobody had their laptops.

And so we ended up having this like whole hour long meeting where we're making like important decisions. And I knew all the data existed because I built this data infrastructure. And we were just, well, three months ago, I heard that it was the UK or whatever it was, and it was just like, there has to, this sh I, it was that like born out of that frustration was this should be a 10 second question on your phone yeah that was the that would have made the meeting 10 minutes instead of an hour that would have gotten us to a faster better decision and oh by the way if we did that

and that was available to everybody sales folks could know it for the customers that they were going to meet execs could know it for the partners they were going to go meet. Even engineers who were on call could, when they get paged, instead of having to run home to their laptop, they could dig into, hey, is this peak of the spike in bug reports due to not being able to connect to a database or having something else fail or having billing fail? And they could triage and do all that without having to be too stressed or run back to their computer and so that was the genesis really realizing that out of all the tools out there and we evaluated a ton of from free open source stuff

all the way to you know the offensive enterprise type offerings that none of them had really nailed easy to use mobile experiences not just doing but asking from scratch and none of them had really nailed easy to use mobile experiences, not just doing, but asking from scratch and not a lot of them had really nailed collaboration and a lot of that was because they were building for a small number of data people. So collaboration didn't matter as much. You didn't need it to let you set up real time alert because you weren't on your phone and only a small number of people had access to whatever tableau or whatever because it was expensive and so the whole model of the

products that were built were focused on selling a ton of features to people whose primary job was doing stuff with data business analyst data scientist and if you really wanted to make it open and really make it easy to use, you had to kind of go back and say, well, what is a less feature rich, but radically easier experience that can work anywhere? Okay. I have so many questions about that, but I want to look back into your history a little bit and ask a question about your work on relevance and feeds, right? Or even like ads for Facebook. So that's fascinating to me because there seems to be on the surface an interesting corollary between figuring out what's important to people and trying to surface that. And when you think about the world of BI being compressed, you know, the world of BI, as we traditionally know it, which is, you know, a big monitor, a SQL editor,

you know, sort of unlimited real estate. Like when you think about condensing that down into a mobile experience, you have to answer this question of what do you actually expose to the end user, right? And so maybe, you know, tell me if this is not true, but it seems I'd be interested to know, did you draw on any learnings from the LinkedIn and Facebook experience in terms of determining, I mean, those are obviously consumer-focused businesses, but really with data, you're talking about consumers. That's who you seem to be serving. So any learnings that influence the way that you thought about bringing, curating BI in a way into a mobile experience, which is a pretty difficult challenge?

Yeah. So the way, the commonality, we wanted to build BI that you didn't need to read a manual for or undergo a long training for, and it's kind of the same way if you open up the Facebook app or the LinkedIn app, at least in an ideal state where maybe they say, Hey, here are people that went to your school. Here are things that are probably relevant for you. If there's an article, they're not showing you to your school. Here are things that are probably relevant for you. If there's an article, they're not showing you the whole article.

They're showing you a preview of that link. In fact, one of the things I worked on when I was at Facebook, which I think is still there, is if you go through your newsfeed and someone has posted like 50 pictures. I was looking at the data when I was a PM on News Feed there, and I realized if something had one, two, three, four, five pictures, it had a really high click-through rate. People were likely to click into it or like it or comment it at a really high rate. But if it had six or more pictures, the likes, comments, shares on that post were way, way lower. That didn't make much sense to me initially.

So I dug into it. And it turns out, even though there was text that said, hey, John had shared, you know, 45 photos. The link, the preview of those photos was just like the first five photos. And so if it was a wedding, it's five photos of shoes or flowers. It's actually not the photo with the bride and groom. Right. And so what we did and we did all these fancy experiments, we could automatically re-rank the photos in clever ways based on comment count on an individual photo level, but basically what we ended up realizing was people just weren't aware that there was more stuff they weren't reading, they were just looking at the

graphics. And so in the lower right corner, we put a light gray wash and said, plus 10 plus 45 or whatever. And that dramatically increased the likelihood of someone to go and actually see all those photos and then they were much more likely, you know, take actions. And so that was kind of like a lesson in how do you just make a thing make sense visually without necessarily requiring someone to understand when we tried this thing where we automatically re-ranked photos based on like and comic it was just super confusing to people they had no idea how it worked stuff

got out of order and it was like that was a really complex system that we had tested and an easy thing that just did the stuff that makes sense by default was the right answer. Another example is like translation of posts. If we know from your LinkedIn newsfeed that you don't have this language on your profile and you've never posted content in that language, but maybe one of your contacts has, we'd automatically translate it. And then if you want, you could change your translation set.

But by default, we had a sense of what languages you spoke and, you know, what language the content was in would automatically translate. And so all these things were like big engagement wins and they're great metrics that were sad sake on the improvements that they had. But the underlying idea was how do you build a product that someone doesn't really need to read a manual to use and it does a lot of the harder or more complex stuff kind of automatic. And if they want to change it, they can.

So when I was translating that over to building Zane, an example of a thing that is a big pain with SQL is dates. I could probably ask, you know, even the most seasoned data folks, you know, maybe know all those like dates and truncation functions and interval functions to say, hey, I want the last seven days of data. And I want that by minute or by hour. And we said, well, like, you shouldn't need to know that. If you have a timestamp or a date field, we'll automatically cast it to a date and we'll show you, we did that, so if you want to tap, you know, down arrow and configure it, you can. But instead of it like failing by default or showing you millisecond timestamps

by default with like different time zones, let's just do what you or I would do if we were, you know, smart data scientists trying to create a useful graph or another example of a thing we did is maybe you end up querying like a really big data set and in the background, you know, that might be a billion rows that are returned. You don't want to return a billion rows to a phone. So by default, we say, hey, we've limited it. And we don't just limit. We don't just add like a limit 100 or limit 1000 or whatever.

We actually do a windowing and ranking function in the background to show the biggest contributors based on the group buys you specified and the metric that you're looking for and that's basically what a data scientist would do if you said hey show me the biggest revenue products by region right they would say okay i'm showing you the first 10 contributors based on the metric you care about and then if you want you can get everything else from there and so it took a lot of thought around what the right experience is. And I think this is something that like, you know, Airbnb does well. I think Figma does well is make the simple stuff simple. And then, you know, we actually have a full custom SQL editor, if you want it that you can do from your phone and saying, and we have type of heads and all

that stuff, that's not what you see when you go to query something. If you want, that's Barry Bum and three dots and all that power is there. But by default, we want to make the simple stuff, hey, how many of my users did this over the last week? Split by day and type, right? That that is a 10 second query where you literally just tap and drag for your phone.

And you don't need to know time operators. You don't need to know relative dates. You don't need to know exactly what the syntax needs to be. We just handle that. And then if you want, you can change it. Okay, so I want to dig into the mobile focus a little bit. So, you know, BI on mobile is a fascinating concept, right? And so you sort of have, I would say, generally, you think about this, and you think about, you know, BI sort of on the spectrum of the person who's building the BI, they are in data tooling on a desktop, hammering on SQL, you know, in the guts of the BI tool, sort of like building BI, right? Building intelligence, etc. consuming that on mobile, the ways that you generally think about that are sort of an executive who can look at basic Google Analytics stats or use Tableau's mobile phone to look at their KPI, mobile app to look at their KPI dashboard or use the Salesforce app and they favorite a sales pipeline report and refresh it when they're bored in a meeting or whatever. And so it's like highly consumptive.

And so it seems like there's this, at least practically in my experience, which is limited, but this huge gap between I'm producing BI in a desktop environment with pretty heavy duty tooling, or I'm consuming and generally what's a pretty bad like mobile experience if you think about, you know, I mean, Tableau on the phone just isn't great, right? I mean, I don't want to be too specific about competitors.

I think they each have values that they offer. Tableau is super flexible and stuff like that. But Tableau, Power BI, even ThoughtSpot, which is one of the somewhat newer companies they are view only on your phone some of them have very limited nlp but you need to pre-specify all these aliases and they don't work very well or like light drill down maybe right but for power bi and tableau like if you click on a number it just makes the bigger. It doesn't actually, there's no interactive querying there. Aside from this like very limited, what I've heard, they're probably going to deprecate from someone who works at Tableau, like NLP type thing that doesn't work.

And ThoughtSpot, you can't ask a question from scratch on your phone. Same with sort of creating dashboards in Power BI or Sigma or any of those. Built to, in fact, for Tableau, their web version is much more limited than their Windows and Mac version. So if you actually want to use the full power of Tableau, you need to run it as a desktop application. Yeah, on the local machine, sure, yep. And on Power BI, it's even worse. You have to write on a Windows machine.

You actually, there's not a Power BI Mac. And so I think that's what was built roughly 15 years ago because that was the world. And now the world has shifted. So let's look at Slack, which is workplace collaboration. More than 75% of Slack's weekly active users

use Slack on their phone in a given week. Or let's take something that's even not a great experience on your phone. Google Sheets or Google Docs is okay. Those have more than a billion downloads on Android. Each of those have more than a billion downloads on Android. And if you look at you know how people send emails right it used to be you'd see sent from my iphone and that was like a novel thing

now more than 50 of emails are read on your phone for the first time and people actually send a lot of emails from mobile and you know if you think about that initially, like shot on my iPhone as a video or set from my iPhone for email, even video editing like TikTok, you literally can edit a video with iMovie or TikTok on your phone and do something like reasonably good, reasonably quickly. Editing video, editing photos with Instagram, some adobe's tools there canva for design right like all these other parts of getting work done have built good mobile experiences you can trade stocks on your phone you can buy a car on your phone and data because it was built largely by legacy companies and sold to data teams where you're at a decimal time, that's basically where the feature set has remained, even of the newer engine.

And so our view was you have to start from a place that is different, which is, you know, if you think about it this way, if you're TikTok and you're trying to take every feature that's in Premiere Pro or Final Cut Pro and make that work on a phone, that's not going to work well. So you're like, oh, you just view stuff. That's the easy way to handle it. But if you say, we're actually going to make this a lot simpler so way more people can do it, and we're going to use these interactions that make sense on a phone. Like being able to send you a push notification when your data hits a certain value or changes by a certain amount, making charts interactive, all these things that natively make sense when you tap a screen, but don't make sense when you're kind of at a big, big computer.

Those are the things we started from. So we actually built mobile before we ever built web. Why? Because it's way harder. For instance, if you have a huge result set, we don't show you a billion rows. We do windowing and ranking function in the background to limit

and show the biggest contributors based on the metric you specify. Or if you have an iffy internet connection, we will run the query. When it's ready, send down a push notification to background load that query result when you have an internet connection everything else will just time out and so what that means is the experiences that everybody else built on mobile didn't work well so people didn't use them and so the the perception was, oh, if I want to do something with that, I can't do it on my phone because it just didn't work well there.

And that's because you were trying to take Final Cut Pro and jam it onto a phone instead of saying, let me build the right 10 things you want to do on your phone. We talk about like BI for the 80% of the company or 90% of the rest of the company. Right. The folks whose primary job is not creating dashboards, but who still want to use data to get decisions made. And so that was the starting point. I love it. Okay.

I have one more question, but I've been, I could go on for hours. I think every episode, but I know Kostas has a ton of questions. One thing I'm really curious about is if you you know like what one thing that i think is interesting about let's use the you know final cut pro versus tiktok example right one interesting thing is that you know you could argue about the quality of the input but basically you have like raw video footage. Right. And so that's the input into either either program. Right. Oh, you know, you have a less sort of fully featured. You're not going to make, you know, a movie that goes into IMAX theater on your phone necessarily.

I mean, you'll be able to get there one day. So I totally get that. But what's interesting about about BI and analytics is that the inputs vary widely, right? And so, I mean, from the quality of the underlying data to the modeling that's done on that data by a team that makes it sort of possible to even do some of those queries or see charts and stuff like that. So how do you think about the input, right? Because your success, it seems to me would be highly dependent on the input, whereas with the Final Cut versus TikTok, you really just sort of need like video footage. Yeah, I mean, I look, I think the input matters even in the Final Cut Pro TikTok example. So if it's super shaky, and there's a lot of wind noise, that's going to be

not a great video either way. If I put my phone on tripod or it's stabilized in some way, maybe I have a gimbal. I actually can make probably a pretty cool skating video, even with my iPhone or a skiing video with, I have a gimbal or something. And so the inputs do matter. I, the nice thing about data though, is if your data is in Trino, Starburst, Snowflake, right, you're going to want some data quality checks there regardless, right? Yeah. Whether you're putting that video in the Final Cut Pro or TikTok, like you want it to be good at video.

And I think the same thing is true with data. Whether you're going to analyze it in Zing or whether you're going to analyze it in Tableau or Power BI or wherever it is, you want to know that it's up to date. You probably want some aggregates or views of those tables that give you like rollups that are more useful. And so you're absolutely right. Like the input quality does matter. The nice thing though, is you can have that same underlying kind of data store data warehouse, lake house, depending on what you're using and hook zing to it. But also maybe hook your Python notebook where you're doing

and building some machine learning model. So usually the infrastructure is already there and what zing is doing isn't necessarily trying to replace finance team using Tableau that has 50 different numbers on it, but it's trying to radically extend the value of the investment that you've already made in getting your data warehouse up and running, in getting your real-time data streaming in so you can update user attributes. And taking that and saying, well, that's a thing that doesn't just need to be visible at your computer.

Or if you want to query it, you have to kind of wait for someone to create a dashboard for you. And they have 10 other things on their to-do list. But rather is a thing where they can then use that on their own. And so that investment that you've made is, if you think about it, amortized over a much greater number of people. So we don't expect that for most companies. In fact, some research from Gartner last year said they expect most enterprises are going to have multiple BI tools. And the reason is they serve like different needs. Tableau is 740 bucks a user a year for an enterprise license.

You're probably not going to give that to everybody in enterprise and their global form factor is not good at all. It's like, this is view only. And so that solves a certain set of needs for maybe creating very specific charts or certain exact presentation. That doesn't solve the need though. A salesperson who's on the go wants to know if a customer is using the product. It doesn't serve the need of a PM who wants to know real time product analytics

and cut it by something for a new product that they launched this morning, but they're flying today. And so we think that there's going to be a set of tools that serve different needs. And where we see us kind of really tapping into a thing that we had no idea. When we built this, we had no idea these folks would sign up. We had, when we launched on product, we had a trucking company sign up. We had one of the biggest event companies in the world sign up. We had a retailer sign up. And you might say like, why are they signing up for this relatively young company when they, actually a lot of them already had Power BI.

When we hopped on phone calls later, had power BI or had Tableau. And they said, yeah, but it's not useful to people in the field. Or it's always working on an older extract of data. Or I can't get like real-time alerts in the way I want them. Or I have a set of dashboards, but it's, I have a pipeline of, you know, two weeks of dashboards that I want to get created. And we said, well, what if you didn't need to create a dashboard? What if it was literally a quick question you could ask, tap and drag, count of people grouped by this over this time period.

And that was a 10 second thing. And it radically opens up, you know, for an energy company, they can know what wells are pumping right now for a agricultural company that recently signed up. They can know, you know, what sales they have lined up. So if they should be picking fruit, that might be better if it was on the trees for another week. And they want to know that when they're figuring out literally in the field with fruit to pick or a retailer wants to know where they're getting low on inventory. And they're not Walmart scale. They don't have these huge systems. And they just have a Postgres database and they hook us into it

and they can know when inventory is running low on fast selling product. I love it. So fascinating. Costas, please jump in here. I've been monopolizing the conversation as I always do, but that's the show, right?

I monopolize, then you monopolize and then Brooks tells us you're done. Yeah, absolutely. No, it has been great so far. So Zach, I have like a clarification, let's say question. We are talking all this time about like mobile, but there are different flavors of mobile, right? It's a different thing to have phone.

It's a different, completely different experience to have a tablet, for example. Right. And without even like going to, let's say, more different sizes of phones and all these things. So when you say that you want to deliver a mobile first experience, do you focus on a specific device? And does the device make that much of a difference when we are talking about BI? Because just to give like an example, like I've seen like design tools like on an iPad. It's a completely different experience compared to something like an iPhone, right? Like you have brand input tools to use, mechanisms.

The real estate is like much bigger. So tell me a little bit more about that. Do you focus on a specific device type or not? Yeah, we think about, the primary thing we think about is like, how do we have these bridges between mobile and web just like work, literally a conversation I had with some engineers and designer on my team earlier this morning, where if you have a dashboard, how do we reflow it in a way that makes sense?

So let's say on desktop, I have four charts at the top, one quarter wide. If I try and squeeze that down to your phone, it's going to be probably illegible. And that's basically what most of the legacy players. And so we reflow it uh where we maintain the top left to right top to bottom order but where we have rules and logic that

makes more sense so on mobile we'll show max two things side by side and we'll reflow it maintaining that order in a way that makes sense. That's like a small example. And that kind of extends across different screen sizes. What we do think about, though, is kind of where are users primarily? And so what we've seen is it's actually primarily mobile phones and then web. So we also have a web client. And we haven't seen a ton on iPad. We were client on iPods and Android tablets.

But, and this is actually very similar to what we saw when I was at LinkedIn and Facebook and all the other places. It's like tablets definitely have a place. But really, if you've optimized for a really small screen and you have a way that it can scale up all the way up to desktop and you squeeze that down and make it super small none of that is legible and that's not very usable and so what we've done is said well okay how should that scale down well you know maybe you sample it but if you sample it then you need to tell someone you're sampling it right and you're kind of downscaling it so they really want every point that they have you know one tap to get and so a lot of downscaling it. So they really want every point that they have, you know, one tap to get. And so a lot of it, it comes down to

what I would call is progressive disclosure. We show you sort of the windowed version, the aggregated version, whatever it is. And then if you have the space, we automatically show you the whole thing. And if we're showing you a subset, we tell you that very clearly so you know exactly what's happening. And if you want, you can say, we tell you that very clearly. So you know exactly what's happening.

And if you want, you can say, hey, always show me the whole thing, even though it might visually look suboptimal. But that's just what I want. So, for instance, if you always want a table of the 100 most recent user actions, not a sample of that, not just the most 10 recent. You could say that, say, hey, always show me the full result. And we'll do that because we recognize that ultimately a user needs to be in full control. And then we just try to do sort of things that you would probably do if you were building this as a data scientist on the project. Stas Pilsen- Yeah, that makes sense.

So, okay. What are the limits on what can be done on a form screen right because okay when we're talking about BI we're talking a lot about

visualization right like people I mean naturally they want to visualize things right okay like tables

and we just scroll like through tables which on its own it can be like a pretty tough thing to do like on a phone screen. Right.

But what are like the limitations that you have so far? Alex Williams- Yeah. So there's a bunch of limitations and a bunch of unique things that are great. Limitations are really big data sizes. You need to be clever about how you handle. So if you have a result set that's really huge, by default, we actually limit that in clever ways. But if you want the whole thing, you can get that.

It's just what we do is we progressively load that result set. So instead of sending a billion rows to your phone, we see that there's only 10 rows left. And when you get to row 90, we load the next hundred rows, right? We just do that stuff to like make it make sense. Or if you want to fill, you know, you can just tap the column header and filter it. But for a big, quick, for like a big result set where there's, you know, a million rows, you don't want to do all that filtering on a phone necessarily.

So over a certain size, that filter should be applied server side and we should send the results. So to make all that stuff work like around those limitations in a way that doesn't feel limited, we spend a lot of a lot of time on. That said, the realistic limits are you're probably not going to build a machine learning model on your phone. Maybe you want trend lines or something simple like that. But you know, if you really want to run scikit learn and develop a new Python package, like that's not a thing you're going to do on your phone, at least at this point, we think that if you have like really complex joints, simple joints are fine, but if you, if you literally joins simple joins are fine but if you if you

literally have you know 10 different tables and you're trying to munch them together in a very nuanced way where you're parsing some json from one and then using that to join to something else and all of that that that's probably not the best use case for doing it on a mobile device. But frankly, the same would be true if it was Tableau, ThoughtSpot, Power BI. You're still basically going to want to do the really heavy lifting stuff in a pre-computed table or in a view or something like that. And that's where oftentimes in Trino or Starburst or Snowflake or BigQuery, someone has set up pipelines or data flows that make it more usable. The fact is, if you're frequently querying that stuff, that probably should live in a layer before it gets to BI.

So hardcore machine learning, huge complex joins. And then I think the third area where I would steer away from mobile is if you really want to do very unique bespoke kind of graph types. So I don't know if you've ever seen the New York Times where they have these really cool

custom infographics. And as you scroll, there's these diagrams that will show like a company's earnings, and then it will show, goes into a narrower funnel and then splits in various ways. That's like a thing where, you know, there's a lot of nuance that goes into setting those up. They're somewhat non-standard. You probably want a lot of configuration options.

And that's probably not best done on a device where you have a lot of stuff you want to set up. So that's probably where I would sort of not steer someone towards Zing or kind of a mobile UI. We think about kind of the problem we solve is like the easy 80% of data questions. We've heard from data teams that that's at least half their time. Hey, can you cut this by this?

Can you graph this? Can you set up an alert that emailed me when this thing happens? And we want to make that stuff easy. And then we think data scientists should, right, be figuring out the right way to structure the data, the right way to get real-time streaming data in. So then, you know, users in the field can set up alerts. So, right, maybe they, you know, maybe they pump their data through RudderStack into Trino or Starburst or BigQuery or whatever it is,

and then all the people in the field can do this really cool stuff on it. So we think about it as letting data scientists do more of the cool, interesting work of building machine learning models, building training models, building retention models, all that kind of more interesting stuff and spending less of their time like, hey, can you regenerate this dashboard with this one other filter?

Yeah, yeah. So, okay, let's see if I understand correctly. Okay, someone can do, first of all,, we did many different things with data in general. It's from writing and building complex pipelines in Python, training models, and of course, doing stuff like exploratory analytics and dashboarding and all these things. From all of the stuff that someone can can do with data, like from what I understand, when we're talking about like a mobile experience, we are talking more

about having, consuming process data. It's like we have some kind of like report already made by an analyst, right? And then you want to have, let's say, some level of control over that, build down or metarize it somehow. Like it's not just a passive consumption of PDF file with graphs, right? It's much more interactive. But still, we are not talking about putting someone on his phone or her phone and ask them to go and do data data motor leak, right?

David Pérez- Right. You though can ask a question from any raw database. Ding is hooked up to. So I literally can go into a database of my user events and say, Hey, I want to know the number of events split by type that happened over the last day or the last week split by minute and set up an alert when that happened and I can do all of that from a raw table or raw view. Now, what I'm not doing is, and I can even define calculated fields and metrics that

I want to reuse from mobile. What I'm not doing is saying, Hey, I have these 10 different data sources that I need to munch together and create a table that's going to update every day. I'm not doing that on my phone. It's still going to be your sort of traditional kind of data engineering. Yeah. team and function. But that's frankly what you'd be doing kind of to prep your data, even if you were going to do analysis in, you know, a Python notebook or any other kind of BI tool, and we think that's actually a really good use of a data team. Why? Because that's the stuff that's reusable pretty broadly.

The granular element for Zing is not a dashboard. It's actually a question. And the questions, if you share them through your organization are visible to anybody who you've given access and you can search across any of them. So what that means is you're not starting off with, Hey, what is this dashboard that needs to have been built to answer every question specifically for this function, it's Hey, someone has already looked at this. I can click that.

I tap three dots to say, make a copy and I cut it any way I want. So I'm not limited to the filters someone has put on that. I'm not limited to how they decided to visualize it. I literally could change that in tap. We still need good input data as does any tool, but we think about it as you actually should be able to ask any question from any of your data that's sort of relatively straightforward and i would think about that in a more quantitative sense it's like probably 30 or 40 lines of SQL tops. It's more than that.

You're probably doing something pretty fancy, right? If it's 20 lines of SQL, maybe you're defining some calculated fields. You're doing some group buys. You're applying some conditions. Maybe you're doing a couple joins. Totally fair game to do in Zane. And we make that pretty easy. But if you're, if you literally have 10 different tables, you're munging together and you're doing lots of complex casting. And one thing is a timestamp.

And another thing is a date and you need to harmonize those match the names, do joins, there's lots of incomplete stuff, right? That, that you probably want to have handled kind of here on in the process. Yeah. It makes sense. probably want to have handled kind of here on in the process. Yeah. It makes sense. Does the user have to write to actually write Python?

It's already equal. So, so, so in fact, the primary interface is you see, you open up the home screen, you see a list of all the questions your colleagues have asked, um, and dashboards they've created if they have, and you can search through any of them. Also see a list of all the tables at your organization that you got access to. And so you can just tap on one of those tables and query those tables and start asking a question from scratch. So your starting point is not one of necessarily just viewing it. Hey, I want to see what a colleague did and build on top of that. Or it's, I want to see a table and go interrogate something there.

And so it's very much focused on showing you what is possible and what others have done instead of just empty, like starting you off in a empty SQL editor window and then the whole UI is you know, top and drag you top of field and drag down to get the summer top to drag to get the, you know, to add a filter condition or exclude something or whatever. So you can do a lot of stuff. You can do rejects and all that kind of fun stuff, but we make it that level of complexity buried kind of one, one level deeper. You could run full SQL if you want.

There's actually a full SQL editor. But we bury that behind the three dots for sort of, you know, the DevOps use case or the engineer who gets pinged and wants to go check something in the server logs and they want to parse some JSON. And so by default, like we don't show parsing JSON as one of the kind of top level things when you're using the visual editor, because that tends to be in the

bucket of one level more hardcore, one level more technical, like the typical business user, but if you want to do that, you type the three dots, you go to customs equal and you can, you know, parse your JSON fields and pull up the stuff you want from that. Here's some. Yeah. Um, that's interesting. And you can, you know, parse your JSON fields and pull up the stuff you want. Yeah, that's interesting. And, okay, you mentioned at some point, like, earlier that we are, I think you said, like, Gartner was saying that organizations in the enterprise are going to be using, like, multiple different BI tools, right? Yep.

So how did you see this happening? Like how, like, first of all, can like BI tools, do they operate between them? Can I have like an analyst who creates, let's say, a report on, I don't know, using Tableau, for example, or Looker, and somehow this can be exposed like through Zynq or like vice versa, because I'm not aware of like how this is happening. David Pérez- So right now we have it indexed on that. And you know, if it comes up as a more frequent customer ask, we will. We're not sure that the right answer though is saying, hey, go create a complex dashboard somewhere else and then try to jam it on a phone.

We actually think there's value to saying, hey, what does someone in the field need and how is that different from what someone on the finance team with 50 different metrics needs? So we could build a way to pull that in if there was demand there. But we actually think there's a lot of value in saying, hey, we're going to make it so easy to create something that you actually don't need that in some sense, it's more work to connect these things

up than just to say, Hey, how many of you make people did this kind of this group by this, Oh, I want to cut it some other way. Let me view that. And we focused a lot on making creation easy. And the reason is, if you look at even in the US, a lot of people, more Americans, according to Pew, 85% of American adults have a smartphone. 77% of American adults have a desktop or laptop. So more people have a smartphone than have a computer. And it's kind of like saying hey let's

try to bring everything in final cut pro and jam it onto a phone you still probably want to think through well if this is going to be vertical video if this is going to be a thing that's supposed to be snackable and more lightweight instead of a feature film like what do we want to do that's a little bit different about that so it kind of makes makes sense. I'll give a quick example. On Zane, you're on a phone. You could do it on the web. We've really optimized the mobile experience. And so what do we know about what exists on your phone?

GPS, right? So you could get a location if you gave permission. I'd say, hey, instead of let me query by warehouse and have a dropdown, which is maybe what I would do in tablet show me inventory for the warehouse that I'm in. That's running low. And so that's actually a way better experience that uses the fact that there's information, there's sensors on your phone that you don't have on your computer and other use cases. Hey, send me a push notification. And we've actually already built something that lets you do this.

Send me a push notification when I tap on this chart, any value that's above X of 10 units or, you know, when it changes more than Y percent or whatever, minute by minute or day on day. And send me a push notification. And so those types of experiences are ones that really, really make sense to do from your phone using the sensors and the unique attributes of it being with you all the time and real time stuff making more sense than it might on a desktop. And so we focused really on making that a great experience rather than necessarily kind of vacuuming up everything from every other tool. Because a lot of dashboards don't get used that much. And so we want to reduce the friction of asking a question instead of a dashboard being a

week-long project or longer at some organizations to create. We want it to be a thing you can kind of quickly ask a question instead of needing to, you know, lay out and design a dashboard of 50 questions. Yeah. Makes sense. All right. One last question from me and then I'll give the microphone back to Eric. So there are, there are many things that are happening lately, like in

technology that have to do with like how like a user can interact with technology, right? We have all these AI things like generative models, blah, blah, blah, all that stuff. You, you are building the product again, like trying to utilize, let's say, unique and new ways of interacting with information, assuming also that like we are talking about someone who is in motion, right? Like they are out there and that creates like a, you know, like unique environments to ask questions. So as someone who's I'm pretty sure you're spending like a lot of time,

like thinking about that stuff. Is there something in the near future that you're excited about? Something that you feel like might change the way that we consume data or we work with data? And I'm talking about like, from the perspective of like the data consumer, the main user of let's say BI2, right? So where do you think that the next big thing will come from? Yeah, I think there's two really exciting areas.

The first is some of the sort of large language models, so Stable Diffusion and OpenAI. They both have kind of ways to query across all of a huge number of data points on how people use language so what does that mean it means that if you say hey i want revenue by country but your database only has sales and it's by region. Maybe you don't want to directly resolve that query, but you might say, hey, when you say sales,

you may mean revenue. And when you mean country, but we only have region, we're going to, we think that's the linkage you want. Does that make sense to you? Historically, you had to manually create all those aliases.

And then there was stuff with word to VAC and vector embeddings and kind of an understanding of how words were related with like word to VAC from Google and TensorFlow and stuff like that. And I think it's getting good enough where you're going to be able to, and this is actually something we're working on as well, say, well, you're asking for sales by region, but we have TXN underscore amount, maybe for transaction amount. And, oh, by the way, we have all these other questions that people have saved and tagged or TXN underscore amount. When they save the name of the question, they actually say sales. And so we can build those linkages and build that understanding.

And we think that some of what is happening with open AI and stable diffusion and some of these like large language models will make that radically better than it has been historically. That's one area I'm super excited about. It's a little bit less like purely generative. And I'm a little bit skeptical in the, Hey, we will magically show you every interesting insight about all your data. I have never seen that work well. Maybe it will get to the point where that is. And there are a bunch of companies that try to do that. I just haven't ever gotten everything I needed from that.

And I think it's hard to build trust in those systems. I think a much better way and almost the way you think about building newsfeed is, let me take some understanding of language. Let me take an understanding of your social graph within a company and use all that context to show you stuff. Here's a question your colleague recently asked. And oh, by the way, you comment on a lot of stuff they comment on. And it's related to topics that you frequently query. That is a great thing to proactively show you. So I'm not trying to do it in kind of a vacuum.

I'm trying to do it on top of these sort of large language models and on top of kind of the knowledge graph or social graph that you might have. That's one really interesting area. And then the second, I think, really interesting area is a more localized context. And that's where you are. That's what you're doing. So, for instance, if I'm an engineer and I have alerts set up, maybe I only want alerts when I'm on call for this given set of data. Or if I'm an inventory manager at a warehouse and I frequently visit multiple warehouses, show me the stuff for the warehouse that I'm in right now, or the

really high priority stuff for a nearby warehouse, right? So your location there, alerting there, who you're connected to, all of that context, I think is going to make querying data, using data much more natural, much more organic. You could imagine, hey, sales of a fast-selling item are going so quickly and you don't have any more inventory coming in until a month from now. I probably want to let you know about that so you take it off your website or so you can try and find more supply more proactively. So I think those are two really interesting areas, large language models, and then this move towards all the cool things you can do by connecting with the sensors in your phone and the kind of social network, if you will,

of the people you work with and the questions they ask and the questions you ask to show you stuff that's way more relevant. This is great. That's all from my side, Eric. I'll give the microphone back to you. So you can analyze this again. I love it. Well, we're close to the buzzer.

So I only have one more question, but, you know, and this will veer more into maybe I could call it like the world of ethics than, statistics about screen time and, you know, how, you know, whatever, you know, engagement in these different apps, you know, spending too much time in them is like harmful, all that sort of stuff. I'm really interested in, I mean, you can have kind of, I at least have two different reactions when I think of driving more experience as on mobile. My first one is, is more screen time on mobile a good thing, right? I'm sure there's all sorts of statistics and arguments. I don't know. But then I think, well, actually, like,

that's probably a much more effective way to use my time on my phone than, you know, scrolling through a Reddit feed about, you know, a stock crisis that I'm completely disconnected from, you know, just to get the spicy takes from people who I don't know if have any authority on the subject. But anyways, I'm just interested to know, like, how do you think about that? Do you think about that? Yeah. So, so I think what we're trying to do is much more like high utility right we don't think that i mean if you're a data nerd and you love to i mean i'll give an example i used to travel with

my laptop and even when i was on vacation when i was in management consulting and even when i was at facebook because it was like hey maybe i want know, or maybe something breaks or maybe I need to do something I can't do on my computer. And we view kind of what we're doing is freeing people from, you know, if you're a dev ops engineer and you're on call, you actually can dig into the data. You can run the query, you can figure out what's going on without needing to run. You as a sales person can go, you know, to that client meeting and look at how many, you know, users are using your product or whatever, and be prepared without necessarily needing to like tether your laptop to your phone and

try to figure it out in a harder way. So we think about it as like freeing. I think it's probably more conflicted when, you know, at Facebook or LinkedIn or, you know, be true with YouTube or TikTok. Then you're like, people are spending hours a day on that stuff collectively. And is that a good use of their time? And I think that's a fair and it's a much deeper question. And frankly, I think the answer is that a lot of those systems are engineered to capture as much of your time as possible. Yeah.

I don't think that sales and marketing data is likely to be something you're going to be spending three hours a day on your phone looking at. And I'd way rather that, you know, if you're at your kid's soccer game and you need to check something to reply to a client, that you can do that in 30 seconds from wherever you are instead of having to log on and take 10 minutes to do it there so the way work has evolved is such that people want to be able to get stuff done from wherever they are and if you do that's actually very freeing so i've been able to go skiing and say, hey, wait, something is broken with the site. Look at these errors.

Let's fix this. Without having to literally get off the mountain, go grab a computer because I saw some planning via email, I'm actually able to go do that in the field. And I think that actually is very freeing. So the macro question, are people spending too much time on their phones? I think it's a question of like, how is that time spent? And if that time is spent by giving them more flexibility, helping them learn, helping them make better decisions, which is what we think Zing is doing, that's great.

And if that's spent, you know, responding to every push notification that comes in on Wall Street bets on Reddit, I think that's less unequivocally clear. Yeah, yeah, totally. Well, no, I think, you know, I one of the things that I would love to see result from from Zing is actually saving a bunch of time on like executive meetings or even, you know, you're on Slack on your phone. And one of the challenges with decision-making is everyone has like a hypothesis or an opinion, and it's actually hard to access that data, right? So I definitely see a world where Zing can create that freedom by helping you be decisive and accurate, you know, in decision-making even when on the go. So very cool. Zach, this has been such a fun show. We learned a ton and you're doing great work.

So, yeah, we'll have you back on soon. Thanks for having me. It was great talking to you guys. Thank you, Zach. What an interesting conversation, Costas. My big takeaway is that I think Zach is onto something in that he recognizes the sort of undeniable trend of accessing business information on mobile devices, especially relative to the rise in remote work as a result of COVID, general trends towards mobile usage. I mean, are people going to start developing

complex software and writing Clojure on their iPhone? Probably not. Because the ergonomics are just so difficult to distill. But consuming and filtering data really does make a lot of sense in a lot of ways for the use cases that he talked about. And it is surprising to your point in the introduction that this really hasn't been addressed well before. So I'm excited. I think that, I think they have a huge potential for success just based on the macro trends.

Yeah. And I will add to that, that it's also like his passion to make this happen, that is also important. There's definitely like a lot of opportunity out there, but there are opportunities that require like a lot of focus, stamina, like the will to make it happen, and he seems to be like the person that can do that, so yeah, I feel like we are going, they are going to be part of, let's say the next wave of BI tools out there. And as you said, like, that was like, like a surprising part of conversation with him.

There were many use cases of having the data while learning, like is quite important. Which is an indication of like also how big this market is, right? Yeah. So yeah, let's see. I think I'll go and try an indication. I'm very curious. Yeah, I need to as well, actually. I really do.

All right, well, thanks for listening in. Subscribe if you haven't. Tell a friend if you enjoy the show. And we'll catch you on the next one. We hope you enjoyed this episode of the Data Stack Show. Be sure to subscribe on your favorite podcast app to get notified about new episodes every week. We'd also love your feedback.

You can email me, ericdodds, at eric at datastackshow.com. That's E-R-I-C at datastackshow.com. The show is brought to you by Rudderstack, the CDP for developers. Learn how to build a CDP on your data warehouse at rudderstack.com.