š§š Why AI Needs Real-Time Data Streaming | Jacqueline Cheong, Artie
Why 95% of streaming projects fail, building AI BDRs in-house, how to acquire enterprise customers with cold emails, raising $12M to fix data streaming, and what its like working with Standard Capital
In this conversation with Jacqueline Cheong at Artie, we learn why 95% of real-time streaming projects fail despite its importance in the age of AI.
Jaqueline also shares the sales playbook she learned during YC, how they got all their early enterprise customers through cold emails, building their own AI sales automations (Artie has no BDRs!), and getting customers to switch to Artie even after spending millions building the same product in-house.
We also go inside Artieās recent $12 million Series A and what itās like working with Standard Capital, a new firm started by three ex-YC partners.
I talked to a dozen people to prepare for this conversation, including Jared Friedman at YC, Jaquelineās sales coach Rasanath Das, and numerous Artie employees like Anirudh Sriram, Ryan Choi, Sarah Berkin, MBA, Shangbing Jiang, and Jacquelineās co-founder Robin Tang.
Support this Episodeās Sponsors
Numeral: The end-to-end platform for sales tax and compliance.
Flex: Sign-up for Flex Elite with code TURNER, get $1,000. Apply here.
To inquire about sponsoring future episodes, click here.
š Stream on Spotify and Apple
Timestamps to jump in:
4:08 Artie: Real-time data streaming
5:13 Why moving data is so hard
9:14 Evolution of data warehouses
12:47 AI needs real-time data
18:44 Build vs buy in data streaming
22:51 How to build in a crowded market
26:26 Early focus on a specific hard problem
30:33 Acquiring enterprise customers from cold emails
32:51 Onboarding their first customer with no UI
35:46 Solving compliance and implementation
38:50 How to automate internal engineering, marketing, and ops
44:01 Building an AI-powered GTM pipeline and motion
53:00 Starting Artie to solve their own problem
59:25 Discovering YC through a friend
1:02:20 Everything Jacqueline learned about sales
1:06:29 How to improve your sales discovery calls
1:10:08 Inside Artieās $12m Series A
1:16:44 What its like working with Standard Capital
1:22:59 Jacquelineās favorite book
Referenced:
Founding Sales (Book)
Find Jacqueline on X / Twitter and LinkedIn
Related Episodes
š Stream on YouTube, Spotify, and Apple
Transcript
Find transcripts of all prior episodes here.
Turner Novak:
Jacqueline, welcome to the show.
Jacqueline Cheong:
Yeah, thanks for having me.
Turner Novak:
Real quick, for people who donāt know, on your shirt, it says Artie, but what is Artie?
Jacqueline Cheong:
Yeah, so Artie is a real-time data streaming platform. So in a nutshell, what we do is we help companies move data across their systems in real time. And a very simple example is moving data from Postgres to Snowflake.
Turner Novak:
And why is that a big deal? That seems like not, something that seems like that consequential. Why is it such a big deal?
Jacqueline Cheong:
Yeah. So I mean, companies have different data stores. Thereās operational data stores and then thereās analytical data stores and they were built for very different use cases. So you have all your transactional data that lands in your operational database, but maybe you want to join it with other data, you want to analyze it, run machine learning models on it, build customer-facing analytical products and query that data that belongs in a analytical data store. So how do you get data into the analytical data store? Thatās the problem that we solve.
Turner Novak:
Canāt you just switch up how the dataās stored initially? And again, it seems like maybe... Why is this such a big deal?
Jacqueline Cheong:
Itās like a-
Turner Novak:
So no oneās built this or?
Jacqueline Cheong:
It feels like a deceptively simple thing.
Turner Novak:
Yeah.
Jacqueline Cheong:
Can I just copy and paste files into-
Turner Novak:
Yeah. It seems like it. Itās just data. Copy and paste it.
Jacqueline Cheong:
Yeah. I think why itās so hard is fundamentally, youāre moving data between heterogeneous systems. So a Postgres database is built differently than a Snowflake.
Turner Novak:
So you cannot just copy and paste it?
Jacqueline Cheong:
No, no. Just the different data types that they allow. There are restrictions on character length. And then so when youāre moving that data, thereās very small conversions that actually have to happen such that you can land that data into Snowflake and have it still be usable. So thereās a lot of complications. And then as you build this system out itās, how do I make sure the data thatās copied over is accurate? How do I make sure thereās no missing data that was dropped along the way?
Turner Novak:
How do you get missing data? Itās just data that gets dropped in a copy paste or?
Jacqueline Cheong:
Yeah, you skipped it. You skipped it because itās not a data type that Snowflake allows. You donāt know what to do with it and the system drops it or you didnāt pick it up properly or it was written out of order. And then as you scale it out in production, the edge cases continue to build on each other.
Turner Novak:
So how big is this system that someone might be using? Is there 100 rows in a spreadsheet and-
Jacqueline Cheong:
No, no.
Turner Novak:
... itās like how big do these-
Jacqueline Cheong:
I mean, transactional systems can be tens and hundreds of terabytes, can be petabytes big. And when we talk about the data thatās moving, weāre only moving the data that has changed. So itās not the entire database or the entire tables, but we, of course, do a original backfill, but afterwards, weāre just taking the data that has changed. And even then, weāre talking about a billion, multi-billion rows every single month. Thatās not uncommon for transactional systems.
Turner Novak:
Oh, wow.
Jacqueline Cheong:
Think about every time you call an Uber ride, every time you check into a hotel, every time you buy a flight. Every time-
Turner Novak:
Even when you open an app, isnāt that getting logged?
Jacqueline Cheong:
Yeah. Yep. Everything is getting logged, every email you send, every message you send, everything is logged. And so, think about how big these transactional systems can be.
Turner Novak:
So if Iām Uber, I probably have how many different databases and how many different rows might I have if Iām that scale?
Jacqueline Cheong:
Yeah. Imagine Uberās rides table.
Turner Novak:
Okay.
Jacqueline Cheong:
Across the US globally, the rides table, each ride that exists in its system.
Turner Novak:
Okay.
Jacqueline Cheong:
Thatās the rides table. That must be massive.
Turner Novak:
And then would they have a different Uber Eats table?
Jacqueline Cheong:
Yeah.
Turner Novak:
Okay. So itās a different database.
Jacqueline Cheong:
Mm-hmm.
Turner Novak:
Is it all rides globally in one spreadsheet or in one database or is it different-
Jacqueline Cheong:
Depends how they architected it. Yeah. It can be split up. As your data gets bigger and bigger, some people start sharding their databases. So rides can be sharded into every single month per region is a different shard. And then you can see how this gets even more complicated when youāre talking about streaming that data into a different store for a different use case, let alone in real time.
Turner Novak:
So there may be some cases where you have hundreds of different databases that are all dumping into your central data warehouse?
Jacqueline Cheong:
Yeah.
Turner Novak:
Do people ever have more than one data warehouse?
Jacqueline Cheong:
Yes.
Turner Novak:
Oh, wow.
Jacqueline Cheong:
It is not that common because by default, itās supposed to be your one centralized data store, but you might have Snowflake for running some AI or ML models, customer facing, maybe not customer facing dashboard, maybe itās AIML models, your BI reports, your ops team can query that data, marketing, sales will run off of that data. And then you might also use a ClickHouse for observability and logs and hosting your customer facing analytical dashboard. So even within warehouses, there can be different warehouses for different use cases that theyāre better fit for. We also see customers, they might have a Snowflake for their more like BI reporting system, and then they also have a Databricks because their data science team is running ML models there.
Turner Novak:
So how did this market kind of evolve to the point of you have all these different types of databases and data warehouses and also to the point where you guys do real time data streaming. I remember that was kind of a big thing when you started it. How did it kind of get to that point up to when you... I think when you did YC, it was summer of ā23?
Jacqueline Cheong:
Yeah. Yeah.
Turner Novak:
... you started the company. So how did the market kind of evolve and play out?
Jacqueline Cheong:
So this is a very old market. Letās say some of these timelines are a little bit more made up, but 30 years ago, when it was mostly people would move data between their transactional databases and then move it into Teradata once a month, like a really old warehouse.
Turner Novak:
This is pre-internet, right?
Jacqueline Cheong:
Yes.
Turner Novak:
So youāre probably taking a floppy disk and putting in a different computer and dumping it in or something?
Jacqueline Cheong:
Honestly, I donāt even know the mechanics of that, but it was physically moving data from one system, downloading it and then moving it over once a month and then maybe your executives would look at it every quarter or reporting would be done there. And then when warehouses went into the cloud, basically technology got better. You could do more things with the data and then you could also store more data, now that it was in the cloud. And so, because you could do more things, there was more use cases and people started moving that data into, letās say, Redshift.
Turner Novak:
Redshift. Is that Amazonās?
Jacqueline Cheong:
Amazonās warehouse was the first cloud data warehouse. And letās say it was 10, 15 years ago, it was once a week, thatās probably good enough. Fast forward to today, itās not uncommon for a company to say, āHey, weāre moving data from all these different systems into our warehouse once every hour.ā Thatās not so surprising anymore because again, technology got better, there were more use cases. People have more data to work with as well and the use cases have also become more common. So itās once an hour, but then the trend has been very consistent. Itās consistently going down. And I think whatās happened over the last couple of years is with this, everyoneās trying to figure out what to do with AI.
And when youāre building AI or agentic systems basically, systems where software is making decisions on their own autonomously or automating workflows or maybe even interacting with your users without a human in the loop, a lot of companies are now realizing, āOh wow, I need to be able to feed live data into these systems because without a human in the loop, thereās no human to be like, āOh, I know this. Something happened in the last 30 minutes that you donāt have context to. Letās not push this decision out yet. Itās just going to make the wrong decisions.ā
Turner Novak:
So you basically, in order to build good AI products, you need instant data being collected and used.
Jacqueline Cheong:
100%. If you want it to be pushed beyond a demo or a pilot, you want it to be pushed into production and for it to be running mostly autonomously, it needs to have live context into everything that happened as quickly as possible.
Turner Novak:
And why didnāt it... When you think about how the market kind of evolved over time, I remember when I met you, I honestly didnāt believe you when you told me real-time data isnāt actually really a thing. Why had it not happened? Was it because... Actually, thinking back, we did have LLMs being pretty predominant. It was like the summer of 2023, right? Or was it ā22? Iām trying to remember.
Jacqueline Cheong:
Yeah, thatās when everything really started going.
Turner Novak:
Yeah.
Jacqueline Cheong:
I donāt think AI was pushing any of this real-time stuff then. I think it was more companies that have fraud models.
Turner Novak:
So you had a lot of fintech use cases.
Jacqueline Cheong:
Fintech use cases or customer facing dashboards. Imagine you as a customer using a product and thereās an analytical dashboard that you can see based off of previous actions. Youāve done a bunch of stuff, reconfigured things, youāve launched a new campaign, and then this dashboard shows you nothing. Itās an hour old, bad user experience and not just bad user experience, but usually in those cases, the customer thinks the product is broken. And so, those use cases were the ones that were dominant. And I think what was... And the AI stuff is more in the recent, Iād say 12 to 18 months, but why itās not democratized basically, is itās a really hard problem to solve. Itās not like it didnāt exist. So there are companies like Netflix, like DoorDash and Instacart, theyāve basically spent many years building out this system in-house.
Turner Novak:
Oh, so they have this technology?
Jacqueline Cheong:
Yes.
Turner Novak:
Okay.
Jacqueline Cheong:
Yeah. Itās like you could have streamed data 10 years ago. Itās just really hard. And theyāve... Usually have a team, maybe five to 10 engineers. Some of them have distributed systems, knowledge. Maybe theyāve done it before. They really understand how to scale this in production. And then you also, after you build the actual data movement pipeline, you have to build the observability, the alerting systems, the monitoring to make sure itās actually robust. And then failures always happen, right? This is software. How do you recover from failure seamlessly?
So all of these little things, theyāve built out over many, many years, but this is a very painful path. And that was the realization. For someone thatās starting today, thatās like, āI need to get real-time data into whatever downstream system.ā Maybe itās a warehouse, maybe itās a data lake, maybe itās another database, whatever it is, if I want to do that, do I really need to spend two years... Well, before two years, hire a team, at least a few with distributed systems experience, and then spend two years building. And by the way, the success rate is actually not that high.
Turner Novak:
Yeah, whatās the success rate? And I remember you told me this.
Jacqueline Cheong:
There was a research paper that actually Confluent put out and it was like the success rate of streaming projects is under 5%.
Turner Novak:
Howās it so low? Itās just you should just be able to write some code and connect the connectors and it works.
Jacqueline Cheong:
Yeah. Thatās what everybody thinks. Thatās literally what my co-founder experienced back at his previous companies. It was like, you make a plan, youāre like, āOh, itās going to take two to three months because we just need these three or four core components and we have a plan and weāre just going to get it set up.ā
Turner Novak:
Yeah.
Jacqueline Cheong:
Yeah, it might take you a month or two to get a demo, like a pilot setup and it might work. And then you try to go to production and itās a whole different animal where all those edge cases I talked about in the beginning, āOh no, 1% of the data is dropping. Oh, we read them out of order. Oh, now we have shards. Schema changes have happened to half the shards, but not the other half yet, but we need to reconcile that because weāre merging into one unified table in Snowflake.ā Itās just really complicated stuff and then it never ends. Itās been almost three years and we are still finding edge cases in different systems as we onboard different customers and everybody has messy data. How can a month be 49?
Turner Novak:
You mentioned that youāve had some teams that have built something in house that have switched over. So I mean, thatās kind of an interesting, I guess, development, but why... People have already built this, why would they switch over?
Jacqueline Cheong:
Because it doesnāt end. The build is not the end of the project. You have to actively maintain it. Youāll find even more edge cases, and then youāll have to patch it, and then youāll have to build the monitors to prevent or observe those edge cases, and then you have to build the failure recovery modes.
Turner Novak:
Yep.
Jacqueline Cheong:
Itās unending, and then your use cases will change. And by the way, as your data scales and increases by 10X, the same system that worked, may not work anymore because scale will also break the system. And so, what tends to happen with those teams is theyāve built it and theyāre like, āOkay, maybe I can sit back.ā And theyāre realizing thereās someone constantly maintaining and running these pipelines. Maybe then they onboard another database. They basically have to rebuild it to support that new database. Even adding new tables within the same database is not always automatic.
There are manual scripts that companies have to run and then schema changes. So a lot of this stuff is not automated and those companies that have switched over, itās basically at that point, my team is spending 30, 40% of their time on maintaining and running this pipeline. And we have customer requests, like features, products that we need to build thatās being pushed out. And so, theyāre just trying to take their time back. This is not their core product and it is actually something that should be commoditized.
Turner Novak:
Should be commoditized because everyone ultimately is building and using the same-
Jacqueline Cheong:
Yeah.
Turner Novak:
... connections and data sources that theyāre using?
Jacqueline Cheong:
If youāre moving data from, again, Postgres to Snowflake, just using that just because so many people are using Postgres and Snowflake, does every single company need to figure out that and build that same pipeline? It just seems a little bit backwards.
Turner Novak:
But you still get people that hesitate on, should I actually use a third party provider for this? Whatās kind of the biggest hesitations that you get?
Jacqueline Cheong:
Yeah, thatās a big one. I mean, with any system thatās real time, by definition, itās mission critical. And so, the typical thought is, āWow, this is mission critical, especially in the early days when it was like two of us, no employees.ā
Turner Novak:
Are we going to trust this start-up?
Jacqueline Cheong:
Yeah. How can we trust this start-up to power our foundational data infrastructure?
Turner Novak:
Because if it goes down, their product canāt work, right?
Jacqueline Cheong:
Yeah. Yeah. Yeah. And so, that is a very rational fear that people have. And I think, but when you think about it is what we benefit from is one, this is our full-time, this is our product, this is our core product, and we benefit from scale. So working with many different customers with seeing more edge cases than any singular company could, and then dealing with a lot of data complexities in the day-to-day. So itās basically weāve seen more and weāre able, our software has accommodated for likely more edge cases than any one company can solve themselves. And thatās why itās actually a safer option and a faster option to get up and running.
Turner Novak:
I feel like some people might argue that itās a crowded market. Thereās just a lot of different data tools and pipeline connectors, et cetera, out there, but you still did it anyways, so why?
Jacqueline Cheong:
I think on the surface, it looks like a very crowded market. Thereās so many data integration tools, but when weāre talking about the streaming space, and my co-founder actually went through this, when he wanted real-time data in Snowflake, he actually went out to try to buy something first. He initially tried a bunch of the batch players, but obviously, it didnāt meet his latency requirements. And the other thing is, he was working at Opendoor and Zendesk, so the scale of data was really big and batch systems just canāt keep up in those situations.
Turner Novak:
This is where you dump the entire database in and it syncs and -
Jacqueline Cheong:
No, this is even like, āHey, every hour weāre going to go and find any changes that have happened in the last hour and then move it over.ā But imagine now that your company is so big that the amount of transactions that have happened in the last hour is massive. The time it takes to move that data continues to increase at every interval. So thatās what I mean by batch systems just canāt keep up at scale. But then heās like, āOkay, let me buy a streaming product.ā And you know what exists in the market with streaming in particular, is a lot of raw infrastructure. And-
Turner Novak:
So what does that mean?
Jacqueline Cheong:
Yeah. The analogy I like to use is, imagine you wanted a wedding cake and what exists in the streaming market is thereās different types of flour, thereās different types of salt and thereās vanilla extract and icing, different types of icing.
Turner Novak:
Okay.
Jacqueline Cheong:
And then based off, and theyāre like, āOkay, now you go, based off of how many team members you have, do you have a baker amongst your team? Have they baked a wedding cake before?ā And then across a period of time, whatever you guys have baked, the quality of that can really vary. Or maybe the wedding cake doesnāt even, it completely topples over because you donāt have the right team or you donāt have the right experience. And thatās the streaming market, a lot of raw ingredients.
Turner Novak:
So you still have to build it.
Jacqueline Cheong:
Exactly. You have raw ingredients, but you have to build it yourself. And so, we kind of approached it in a very different place. Our assumption is that companies want the outcome that a streaming system provides, which is, I want data faster to pump data into a risk system. I want data faster so my AI agents can get the context they need, but they donāt actually care about the infrastructure. So our whole thing is like weāre coming in and weāre like, āForget all the raw ingredients. Hereās the cake. This is a cake thatās completely done. Itās already baked. Just go and eat it.ā
Turner Novak:
That is actually what most people would prefer on their wedding day is just, āHey, baker, bake me a cake.ā
Jacqueline Cheong:
I donāt think anyone bakes-
Turner Novak:
Nobody bakes their own wedding cake.
Jacqueline Cheong:
Yeah. So it was taking a very different angle to this space and it seemed to make sense at the time. We were like, āWhy wouldnāt people want this?ā
Turner Novak:
Yeah. What were some of the biggest product decisions that youāve made over time, like early or even more recently?
Jacqueline Cheong:
I think one very opinionated thing that we did in the beginning was we only focused on databases as a source. Weāve since branched out. We have an events, we can ingest events now and stream it into downstream systems.
Turner Novak:
So this is when it happens in the product, it doesnāt go to the database, it goes straight to the data warehouse?
Jacqueline Cheong:
Yeah. It hits our events API and this is usually more like web events click streams, this type of data that doesnāt need to hit a transactional database. But in the beginning it was like, āWeāre not going to touch any other source. Weāre only going to do databases.ā
Turner Novak:
Yeah. Whyād you do that?
Jacqueline Cheong:
So that we could be really, really focused because there are maybe 10 really important transactional databases that exist today. And we were like, if we only focus on these very few sources, we can build really, really in depth, focus a ton of time on dealing with these edge cases that typically break the pipelines, and then we can build the best product, the most reliable product. But if we get spread too thin, by definition, youāre trading off on quality. So we were like, āWeāre not going to do anything else.ā And we did have prospects that come in, theyāre like, āHey, if you build an integration into Salesforce or Workday, weāll buy your product.ā And that was... When you had zero revenue, itās hard to say no.
Turner Novak:
Did you chase any of those or do any of those?
Jacqueline Cheong:
No, we said no.
Turner Novak:
Oh, wow.
Jacqueline Cheong:
And it was painful, but I think it was the right decision because for the first two, two and a half years, that really was our strength. It was our secret sauce. And for people who tried our product, they were like, āWhoa.ā And this is kind of funny because you would think products would work, but theyāre like, āWhoa, what you said in the demo, it actually works.ā
Turner Novak:
Yeah. Yeah, I remember when I invested, when I first met you guys, some of the diligence from engineers was, they had very strong feelings about other products, strong negative feelings, specific words to describe them, but yeah, Artie just worked really well and I actually liked using it. So itās interesting that thatās kind of the NPS score of the market, is these visceral reactions from the engineers of not wanting to use the product.
Jacqueline Cheong:
Yeah. I mean, itās actually, I didnāt even know this before because I wasnāt in the data space before this, but itās crazy the amount of skepticism that comes in because theyāre like, and this is-
Turner Novak:
It just works. Come on, that canāt be possible that this thing just works.
Jacqueline Cheong:
Yeah. Yeah. And actually, theyāll come in, theyāre like, āBy the way, this is the beginning of a discovery call.ā Theyāre like, āHey, I just want to let you know Iām very skeptical because Iāve tried tool one, two, and three, and Iāve been burned by all of them and so, excuse me if Iām a little skeptical on what you can do.ā And then you show them the demo and theyāre impressed, but thereās still a lot of skepticism. When they try the product, and I always tell people, all those things that broke from before, just give us your hardest tables and your most problematic data. Try to break Artie with it.ā And thatās when they get that reaction like, āWow, you guys werenāt kidding. It actually just works.ā
Turner Novak:
Yeah. Itās kind of like the do hard things, solve hard problems. So basically, you tell your prospects, youāre like, āTry to break it, try to stump us, try to do something that you donāt think weāll be able to handle.ā
Jacqueline Cheong:
Yeah.
Turner Novak:
I think another super interesting on the customer front, when you were doing YC, I think thereās a common YC playbook is sell to other YC companies, people in your batch, whatever.
Jacqueline Cheong:
Yes.
Turner Novak:
Thatās kind of the most common YC playbook. Thatās how you grow really quickly. What you guys did was the complete opposite of that. So how did you get the first couple customers?
Jacqueline Cheong:
Yeah. And by the way, we tend to like to follow YC advice because itās very practical and very smart. Our problem was, you need to have a decent amount of data for streaming to make sense. And so, most YC companies, especially in our batch, when youāre just starting out, you donāt have data problems. And so, we wouldāve loved to sell to our batch, but no one had data problems. And so we couldnāt. And so, we had to rely on just cold emails. And so I was just sending, I think during YC, they had this like... Theyāre like, āJust write personalized emails.ā This was in the very beginning of AI, so GPT 3.5 wasnāt even that good. And so I was just handwriting emails and I had the target to write 20 emails a day.
Turner Novak:
And it was just Netflix, head of data, whatās their email?
Jacqueline Cheong:
Yep.
Turner Novak:
Be like, āHey, James.ā You just slide in and just say whatever needed to be said?
Jacqueline Cheong:
Yep, exactly-
Turner Novak:
Okay.
Jacqueline Cheong:
... like, āHey.ā The email to Substack must have been like, āHey, Mike, I noticed that you use Snowflake. Already moves data into Snowflake in near real time and we handle all the hard stuff like schema evolution and merging and everything. Does this sound interesting? Would you want to chat?ā And that was the gist of it.
Turner Novak:
And that was your first big customer, right, with Substack?
Jacqueline Cheong:
Yeah. That was not even our first big customer. It was our first customer.
Turner Novak:
And it was a big customer though too.
Jacqueline Cheong:
Yeah. Yeah.
Turner Novak:
It was a pretty sizeable customer.
Jacqueline Cheong:
It was terrifying. I mean, we had only tested our pipeline with maybe a couple thousand rows, which in database land is very, very, very, very tiny. And the moment they wanted to onboard or in the POC, they were like, āYeah, weāre going to start streaming a couple billion rows.ā.
Turner Novak:
So were you guys like, āShit, what do we do?ā
Jacqueline Cheong:
I mean, yeah, effectively. And then we were like, āWe have to make this work.ā
Turner Novak:
Yeah.
Jacqueline Cheong:
And so we obviously, there were a lot of hiccups with onboarding a table that had tens of billions of rows of data in it, but they were very nice about it. They really wanted the promise of what our product could do at the time, and they really helped us go through the hurdles of... I mean, they had very, very strict requirements of how you could connect to the database, how fast you could pull from it, because they wanted to protect their databases, a very reasonable thing to do. That rigor really helped us make our product better. And then actually, the next 10 customers we onboarded after Substack were significantly easier, nowhere near what we had to do to onboard them. But the funny story is with Substack, at that time, I donāt even think we really had a UI.
Turner Novak:
Oh, yeah. I remember it was a big deal when you just like, āThere was a dashboard.ā
Jacqueline Cheong:
Yeah. That was a big deal because we didnāt have a dashboard when Substack onboarded.
Turner Novak:
So they didnāt know if it was working basically? It was like, āThe productās running, but how do we access it?ā
Jacqueline Cheong:
Yeah. I mean, they could see it after it landed in Snowflake.
Turner Novak:
Okay.
Jacqueline Cheong:
And then they could query the data and stuff like that, but-
Turner Novak:
But thereās nothing telling them that it was working?
Jacqueline Cheong:
No.
Turner Novak:
Okay.
Jacqueline Cheong:
Well, what we did, we had a shared Google Doc.
Turner Novak:
Okay.
Jacqueline Cheong:
Or a Google Sheet and then they listed out the hundreds of tables that they needed synced over.
Turner Novak:
Okay.
Jacqueline Cheong:
And then we had a status column.
Turner Novak:
That you would just update?
Jacqueline Cheong:
We would just be like, āThis is backfilling.ā
Turner Novak:
Okay.
Jacqueline Cheong:
And then when they completed, weād be like, āNow this is streaming. Itās done backfilling.ā
Turner Novak:
Okay.
Jacqueline Cheong:
And weād put timestamps on things, but it was just Robin going in there and updating it over multiple days because it takes that long to...
Turner Novak:
To upload them. Yeah.
Jacqueline Cheong:
Yeah, yeah.
Turner Novak:
And so they still went through with this, even though that doesnāt sound like a good user experience, honestly, but it was worth it to get the.
Jacqueline Cheong:
Yes.
Turner Novak:
I mean, it sounds like, was it real time because Robin was manually doing this stuff?
Jacqueline Cheong:
Oh, oh. I mean, the onboarding was absolutely not real time, but once data was streaming-
Turner Novak:
Oh, so this was actually the hookup, the onboarding process?
Jacqueline Cheong:
Yes. This is the implementation and doing the historical backfill, but then once it caught up, then it was just streaming. And so, the backend infrastructure of all of that actually worked really well, but the UI experience was definitely not there. I mean, it didnāt exist.
Turner Novak:
And then you obviously built almost scaffolding, observability alert systems all around that.
Jacqueline Cheong:
Yep.
Turner Novak:
So then what did you learn about just implementation? Because I know thereās one customer, itās actually, I think it was kind of hilarious. Dedicated listeners of the show will remember thereās an episode with Tommy, the CEO of Alloy, which is one of your customers. Youāre actually in the office implementing Artie while we were recording. And I think maybe thereās a couple of people that actually remember us having that conversation and we brought you guys up. You were in the office and that was the first time I think you did hands-on, you went to the office.
Jacqueline Cheong:
Yes.
Turner Novak:
And did the onboarding. So how did that come about starting to do that?
Jacqueline Cheong:
Yeah. That was different because it was our either second or third implementation of our BYOC offering. So basically, we have our cloud solution where Artie Cloud processes your data and then moves it into, from your databases into your warehouse or data lake. There are a ton of fintech, govtech or just heavy compliance companies where theyāre like, āOur data cannot leave our AWS Azure or GCP environment.ā
And so with those, we implement the entire data plane in our customerās environment so that data processing doesnāt have to leave. And this can actually, we actually do a lot of this remotely now because weāve done it enough times, but I think it was more like, it was early, we just wanted to make sure that we had live communication because if we do... I think we went there, we sat next to their security and networking team. It wasnāt even actually the team that would use us day to day. It was like, āLetās get security and networking figured out.ā
Turner Novak:
Because it was a big compliance issue.
Jacqueline Cheong:
It was a compliance thing. And so it was like, āHow do we figure out the minimum amount of networking permissions that our data plane thatās sitting in their environment needed to have to connect?ā So it was really honing in on these things and then also making sure the team was comfortable and had someone live to talk to whenever they ran into issues so we could move a lot faster.
Turner Novak:
Because you ran into an issue where their security team just said, āNo, we canāt do this and we have to kill the deal,ā or something like that.
Jacqueline Cheong:
Yeah, yeah. I mean, it never went to the point where they were like, āWeāre going to kill the deal,ā but it was like, āHey, we canāt do this. We canāt do this. We canāt do this.ā So we were like, āYou know what?ā And this was over Zoom and over Slack. Weāre like, āYou know what? Weāre just going to come there and then weāll just go into a conference room together over a couple of days and then weāll tell you exactly what we need. And then you can run the command. We can be next to you and figure all of that out in a couple days instead of like...ā You could imagine that dragging out for a month, but we really wanted to make sure that the team that needed this, could get up and running as soon as possible.
Turner Novak:
So you do a lot of this just manual hard product fixes and then you kind of automate them essentially, seems like a big-
Jacqueline Cheong:
Yep.
Turner Novak:
Itās like a big part of it.
Jacqueline Cheong:
Yeah, yeah. And we do this for actually everything. Itās not just engineering, but even across marketing or sales, if itās the first couple times weāre doing something, oftentimes, I will personally just manually do something. Once we realize that it works, we know what the SOP or standard performance is, then weāll create a runbook, automate it, and then itās much easier for the next, either customer or the next people that need to do this.
Turner Novak:
So I think even now, one of the more interesting things, did you hire a head of growth yet?
Jacqueline Cheong:
No.
Turner Novak:
But youāre looking to or youāre trying to make a growth hire, is that right?
Jacqueline Cheong:
We do not have a growth hire unless... We do have a marketing hire.
Turner Novak:
Okay. And they report to Robin, the CTO, right?
Jacqueline Cheong:
Oh, youāre talking about our BizOps?
Turner Novak:
Okay.
Jacqueline Cheong:
Yes.
Turner Novak:
Yes. So I would think a marketing person attached to revenue, they would report to the CEO.
Jacqueline Cheong:
Yeah.
Turner Novak:
But they report to Robin.
Jacqueline Cheong:
Yes.
Turner Novak:
The CTO.
Jacqueline Cheong:
Yes.
Turner Novak:
Why did you do that?
Jacqueline Cheong:
Itās a hack. So business operations, itās a pretty loose... Depending on the organization, it can mean different things, but itās about scaling and optimizing processes. And I think especially today with all of what you can do with AI and even on the go-to-market side, automating with play tables and agentic flows, that is a very technical problem. Theyāre just like, not to confuse with data pipelines, but a different type of pipeline that you would run internally. And so, I actually think, and one of the advice that we got is, itās a superpower if you treat it as an engineering problem. And so, especially with technical leaders, I think their brains are wired to always optimize.
Turner Novak:
Always try to automate?
Jacqueline Cheong:
Yeah.
Turner Novak:
Yeah.
Jacqueline Cheong:
I think itās an engineering thing. All the best engineers that I know, theyāre just constantly thinking... Theyāll walk into a restaurant. My co-founder will walk into a restaurant. Theyāre like, āEh, they really shouldnāt put the seating like this.ā And Iām like, āWeāre just eating.ā
Turner Novak:
This is what, Robin will say this?
Jacqueline Cheong:
Yeah.
Turner Novak:
Thatās awesome.
Jacqueline Cheong:
Heās like, āIf they can rearrange this this way, they could put 20% more seats.ā
Turner Novak:
Yeah. Honestly, so sometimes Iāll have that conversation with my wife and sheās like, āWho cares? Why are you thinking about this at dinner?ā
Jacqueline Cheong:
Iām just like, āWeāre just having dinner.ā
Turner Novak:
Well, actually, I remember one time it was, I think you had, oh, it was the new office party. I was talking to Robin once and he was telling me, it was some sort of like, āYeah, Iāve optimized all these things and I donāt have to do any work anymore. If I chose to, I wouldnāt have to do anything because Iāve automated everything.ā
Jacqueline Cheong:
Yeah.
Turner Novak:
And I was like, āOh, thatās pretty cool.ā Obviously, I think he told me, āItās done. I donāt have to work anymore because Iāve automated every single thing,ā and obviously, thatās not true anymore, but it was a fun story. I just remember him telling me. Heās automated everything about the job.
Jacqueline Cheong:
Yep.
Turner Novak:
I think he automated fixing bugs when thereās a issue with one of the pipelines, he probably had automated a process of doing that.
Jacqueline Cheong:
Yeah. Yeah. We automated a DevOps process every quarter or every half a year, we have to upgrade Kubernetes and a bunch of other stuff across all our data planes. And at this point, between our data planes and all our customers, like BYOC data planes, we have quite a few. So it would literally take someone a week or two weeks every quarter to upgrade everything.
Turner Novak:
Of just manually going into all these different systems and clicking buttons
Jacqueline Cheong:
Literally just upgrading, yeah, following a runbook. And so, we built recently in agentic flow where you just have to... Itās the same process, right? And AI is great with following runbooks. If you can build something into a runbook, now if someone just presses start and it will create PRs, youāre like, āOkay, great. Do the next step.ā And you just watch it do the next step. Wait 15 minutes, see itās okay, and then do the next step. So weāve automated that, but yeah, having someone whoās naturally obsessed about optimizing something, run business operations, I think has been the biggest hack for us.
Turner Novak:
Interesting. So you have the ops team report to the CTO
Jacqueline Cheong:
Yes.
Turner Novak:
Because theyāre automating a lot of things.
Jacqueline Cheong:
Yeah.
Turner Novak:
And so do you still do a lot of just outbound, a lot of cold outbounding at this point or is it you have a bunch of BDRs maybe? Is it a lot of warm stuff? Whatās the process still look like?
Jacqueline Cheong:
Yeah. Itās a pipeline. So we still do some... Thereās still some personalized outreach that weāre doing. Itās highly researched. There are accounts that we think are, we know theyāre dealing with problems that we solve and itās deep research, learning about the systems that they have, the technographics and whatnot.
But we also have optimized and built a go-to-market engineering pipeline with the tools like Clay and a bunch of other tools that youāve stitched together. And it kind of understands our ICP, the right buyer persona, and this thing is just running autonomously. So weāve basically built a pipeline to replace BDRs and SDRs, and we only have AEs at Artie.
Turner Novak:
Isnāt there some of these AI tools like AI BDR? Do you use some of them or you build your own?
Jacqueline Cheong:
We do not use those. And maybe this is like, we should test it because AI is improving so quickly, maybe weāre missing out by not testing it, but we basically built our own internally that... And you know the great thing about this is, itās so much easier to train because with a lot of SDR, BDR functions, thereās a lot of like, āHey, letās make sure we describe Artie properly. Letās make sure thereās no typos, grammar is correct. Letās make sure to remember to follow up to a prospect.ā
Turner Novak:
Exactly 24 hours after the call or whatever time-
Jacqueline Cheong:
Or itās like-
Turner Novak:
... or a certain drip scenario.
Jacqueline Cheong:
A drip, drip scenario. And that stuff AI is great at. It will never drop-
Turner Novak:
Itāll never get that wrong.
Jacqueline Cheong:
Yeah. And it will never misspell words or get how to describe Artie because they understand our messaging framework and our positioning. And so youāre really dealing with tweaking the language rather than training more basic stuff.
Turner Novak:
And actually, this is maybe one of my friends whoās probably one of the best founders at sales Iāve come across, when youāre talking about no typos, he actually showed me that his biggest hack for emailing people is lowercase subject line in the email and also maybe a typo also because-
Jacqueline Cheong:
To show that youāre human.
Turner Novak:
... it knows that itās AI. It shows itās actually a real person sending the email. I think the other interesting hack that he has is you get people on iMessage as soon as possible-
Jacqueline Cheong:
Oh, interesting.
Turner Novak:
... because heās like, āDeals actually close on iMessage, not on email.ā
Jacqueline Cheong:
Yeah.
Turner Novak:
Which is kind of interesting. Maybe especially when youāre doing hand-to-hand combat sales. He does pretty, I think itād be like high ticket ACV, pretty big deals that heās closing and heās very much like an iMessage, deals close on iMessage. Have you found that much or?
Jacqueline Cheong:
Yeah. I mean, our equivalent is actually just Slack or Teams-
Turner Novak:
Slack or Teams, okay.
Jacqueline Cheong:
... like direct the channel between, and weāll pull our engineers into it. And the whole idea is, when you reach out with a question or when thereās an error, youāre not reaching out to a customer support person that might just be going to the docs and copy and pasting answers and then just giving that answer out to you. Youāre reaching an engineer who can-
Turner Novak:
Fix it.
Jacqueline Cheong:
... who can fix it and also not only tell you that itās fixed, but explain exactly what went wrong, what we discovered and then how we fixed it. And I think thatās one of the, actually the best-selling points when we were working with prospects that are like, āThereās a lot of transparency, you guys are actually helpful.ā And yeah, because software is not perfect. There are errors here and there and itās like how you deal with them and how much transparency you give your customers.
Turner Novak:
Yeah. Itās also too, if youāre emailing, it almost feels like too corporate like, āHi, Jacqueline, it was great meeting yesterday. Iām following up, blah, blah, blah.ā Versus Slack, itās like, āOh, no, this broke. Can you fix it?ā Or whatever. Itās just more to the point, like a little more casual.
Jacqueline Cheong:
I think the thing, actually what it does is, it makes them feel like weāre an extension of their team because then it branches out. Theyāre like, āHey, weāve decided to build this new feature and weāre curious about what is the best architecture so that we can serve it to our customers for this use case.ā And theyāll describe it and I guess itās outside our scope, normally, but then we will actually go in and maybe do a quick Zoom call or a huddle or something and think through or we can share, āHey, we have another customer that did exact this thing and this was the best architectural pattern to achieve this.ā And so, weāre very much an extension of their engineering team.
Turner Novak:
Yeah, and I feel like one of the advantages startups have is, you can over support the customer. Itās impossible to overinvest in the customer experience. I mean, maybe itās possible, but just going above and beyond and making people feel good, making it easy to work with you and the different ways to do that, whether itās you go to their office or you have the Slack connection or you have their iMessage or you have dinner with them more often just or the immediate response, instant response. They Slack you and within two seconds, the bubbles pop up, youāre responding or the name like Jackmanās typing. I feel like all of those things, it just makes people trust you more and want to work with you.
Jacqueline Cheong:
And people have been telling us for a long time like, āHey, this isnāt sustainable. At some point youāre going to become a bigger company and you canāt provide this level of support to your customers anymore.ā And I actually donāt believe that because AWS has great support and theyāre massive.
Turner Novak:
One of the biggest companies in the world.
Jacqueline Cheong:
Exactly. I think itās just if you intentionally choose to prioritize this, if this is a core part... I think this is a core part of our product actually. And if you believe that and you intentionally try to keep it, you can make it happen because AWS has.
Turner Novak:
Iāve been having the worst experience lately with my health insurance. This is a shout-out. Iām selfishly saying this, I feel like there needs to be a better startup friendly version of health insurance or insurance. We use this provider in Michigan and I donāt know, we mess up the password on our account and Iāve been trying to pay the bill for a month and I canāt log in and weāve sat on the helpline multiple times for hours at a time. They canāt reset our password and get us into our account to pay. Iām like, āThis is so...ā Itās like 20% of GDP runs through health insurance, basically.
Jacqueline Cheong:
Yeah. If you care about those things as a company, itās such a hack. Your win rate must triple just from responding to you to help you reset your password. Itās such an easy thing to do.
Turner Novak:
So if you are making better health insurance, not only will I probably try to buy it, I want to invest in it. I feel like maybe itās a terrible category. I have no idea, but Iām like-
Jacqueline Cheong:
Probably. I have no understanding of the health insurance space.
Turner Novak:
But itās big. I think itās like an ingredient. You need a big market. Healthcare is a massive chunk of GDP. Thereās also the other angle of going direct and going around health insurance and just go directly to consumers, building a better product that doesnāt incorporate insurance.
Jacqueline Cheong:
It might have been Dalton Caldwell or Michael Seibel that said this to our batch, but itās like, āIf youāre pivoting an easy way to figure out what to do is find a really, really big incumbent thatās doing something and all their customers hate them and then just build a better that.ā
Turner Novak:
Yeah, find customers who are not happy to have a problem and fix the problem.
Jacqueline Cheong:
Yeah. Because clearly, itās a painful enough problem that theyāre still with them.
Turner Novak:
Well, thatās kind of how you started Artie maybe initially, going back to the beginning.
Jacqueline Cheong:
Yeah, it was so painful.
Turner Novak:
So how did that go? Going back to the beginning, whatās the story there?
Jacqueline Cheong:
Oh, gosh. I guess we have to go six, seven years back, but my co-founder, whoās also my husband, he used to work at this marketing automation five person YC startup and they were a really small company, but because of the nature of marketing automation, they were dealing with immense scale. So he was figuring out how to build databases that could store billions of rows of data on a daily basis. And then that startup ultimately got acquired by Zendesk. At Zendesk, thatās when he really learned and used CDC because Zendesk have so many different products.
Turner Novak:
So CDC is change data capture.
Jacqueline Cheong:
Change data capture.
Turner Novak:
Which means?
Jacqueline Cheong:
Which is the way you can grab database changes and get them out of databases.
Turner Novak:
So instead of taking the entire database and copy and paste it, you just say, what has changed and what just-
Jacqueline Cheong:
Exactly.
Turner Novak:
... copy whatās changed and paste that in.
Jacqueline Cheong:
Yeah. Exactly. And because Zendesk had so many products, they were basically using CDC to grab all the changes from all the different products and building a unified experience for their customers. So that was really important there. And the scale of which Zendesk was operating at is also important in the history of this. Then he went to Opendoor and thatās when they really could benefit from getting real-time data into Snowflake because they were such an ops heavy company.
They were buying and selling homes, but every time they bought a home, you have to schedule an inspection, you have to do maybe repainting the walls and then relisting and stuff like that. All the ops people, obviously, you canāt have access to the database, all the ops people worked off of Snowflake data. And so the faster that you could get data in, you can imagine one of their key metrics was days on market of the homes. So thereās a bunch of different factors that go into that, but one of the things is if we can get data in faster and our ops was more efficient, then days on market could theoretically go down.
Turner Novak:
And make more money.
Jacqueline Cheong:
Exactly. And so this was an important project. And then he ended up, I kind of described it a little bit in the beginning, but what he ended up having to do was bake the cake, get all these tools. It was seven or eight engineers at the time building for 11 months to a year. And it was still not fully ready to go into production because of all the edge cases, the schema drift and all that stuff that they had to handle.
And thatās when he was like, āWow, I am moving data from Postgres to Snowflake. How many other companies need to do this? Do they all have to hire out a team and use these tools to build it out? And do they all spend a year or two doing this? What if this could be a product where I could deploy in like 15 minutes?ā Thatās what he wanted. And so, that was the idea. And when he told me about it, actually I was initially like you. I was like, āAre you sure? Not everybody has real-time data.ā
Turner Novak:
That just kind of seems ridiculous. Yeah. Why is that even a problem? It should have been fixed.
Jacqueline Cheong:
Yeah. And I trusted him, but I also wanted to verify. So I went out and I talked to a bunch of my friends that worked at different tech companies in the Bay Area and I found out that, wow, it wasnāt like a Zendesk specific problem or Opendoor specific problem. This is just how things worked. And then thatās when I got really excited because it seemed to be a structural problem with how streaming worked today. And there seemed to be this gap that we could fill if this product could, we could build this product. So we decided to jump in and see what we could, see what would happen.
Turner Novak:
And what were you doing at the time?
Jacqueline Cheong:
I was working at a hedge fund called Balyasny. I had been there just over three years, I believe. But I was investing, it was a long, short fund, I was investing in enterprise software companies.
Turner Novak:
And so you kind of understood how software worked as much as a public-
Jacqueline Cheong:
Oh, gosh.
Turner Novak:
... market investor could understand or?
Jacqueline Cheong:
Correct. Yeah. I mean, now I look back and Iām like, āWow, I really didnāt know.ā I really didnāt understand the technology, but yes, I learned a lot about the business model and what software companies did, what great looks like in terms of financial metrics and stuff like that.
Turner Novak:
You knew what a good software company looked like from the outside, but you didnāt know how to make it?
Jacqueline Cheong:
Yeah.
Turner Novak:
Yeah. Okay.
Jacqueline Cheong:
Yeah.
Turner Novak:
So what was the decision to actually start a company. When Robin was like, āI want to make this company.ā Were you like, āCool, go have fun.ā Or were you like, āOh man, I kind of want to join. This seems fun.ā How did that evolve?
Jacqueline Cheong:
Yeah, it kind of started off like, āI will help you do...ā It was very simple. I think part of it was that we were a little naive at the time because it was like, āHey, weāre going to get your data in faster and itās going to be the easiest thing youāve ever deployed to achieve this. And from a TCO perspective, itās going to be a lot cheaper.ā It sounded like a no-brainer. So our initial thought was like, āHey, weāre just going to build it and weāll tell people weāre building this and people will just buy it.ā That was the thought. So Iām like, āHey, Iām going to help you set up the company. Iāll do the sales conversations. And you build it.
Turner Novak:
So it was almost like a side project almost as you were thinking about it or was it like, āWeāre going to quit our jobsā and started working on this?
Jacqueline Cheong:
We quit our jobs.
Turner Novak:
Okay.
Jacqueline Cheong:
We were like, āWeāll quit our jobs, weāll do this.ā We just thought it would be easier than it was.
Turner Novak:
The classic just, I mean, nothingās ever as easy as you think.
Jacqueline Cheong:
Yeah. Yeah. Yeah.
Turner Novak:
And you guys did do YC. Was it right in the beginning like, āWeāre starting a company, letās apply to YC.ā Or at what point did YC come into the picture?
Jacqueline Cheong:
No. So funny enough, for whatever reason, I didnāt know about YC at all, had never heard of them. I was really not in the startup space. So we quit our jobs, Robin started building. I was like, āI need to buy a book to learn how do people do sales.ā
Turner Novak:
What book?
Jacqueline Cheong:
So I was reading, I think it was Foundings Sales, that was the book, by Pete Kazanjy? Really good book for someone that had absolutely no understanding of how to do sales. So I was reading that book and I was getting the business, the admin stuff set up. I tried to email a bunch of people to have conversations to see if they would be interested in the product. And thatās what I was doing.
And I didnāt know what the ICP was or the buyer persona. So it was a really broad category of people that I was going after. So what Iām trying to say is, I really didnāt know what I was doing. And then at the time, I happened to see one of my college friends post on LinkedIn that he was doing YC and he was starting his own company. So I actually just reached out to grab lunch and to learn more. And at lunch heās like, āOh, you should apply to YC.ā And heās like, āIf you apply and you get in and if you donāt do it, youāre an idiot.ā
Turner Novak:
Thatās how he described it?
Jacqueline Cheong:
Yeah.
Turner Novak:
Okay.
Jacqueline Cheong:
Thatās just how he was... And heās like, āYou should apply. And then if you get an interview, I will... Text me and Iāll tell you how the interview goes.ā
Turner Novak:
Okay.
Jacqueline Cheong:
So yeah. So then I just did it and I was like, āWell, who knows if weāll get in?ā Because he was like, āThe acceptance rate is really low.ā
Turner Novak:
Yeah, itās 1%. I was talking to Garry. I talked to him yesterday. The episode will come out in a couple weeks after people are hearing this probably, but yeah, heās like, ā1% of application approval rate.ā
Jacqueline Cheong:
Yeah. So I was just like, āOkay, whatever. Weāll just submit an application and weāll decide if we get it.ā And this is a very YC thing, but when they accept you, they ask you on the call-
Turner Novak:
If youāre going to take it, right?
Jacqueline Cheong:
Yeah. And obviously, itās a little bit of a knee-jerk reaction, but I was like, āYes.ā And then after-
Turner Novak:
Because your friend was like, āYouāre an idiot if you donāt do this.ā
Jacqueline Cheong:
Yeah. And I was just like, āOkay, I just said yes.ā And then after the call, I remember I turned around and I was like, āRobin, was that okay? I said yes. So like I guess weāre doing this. Right?ā And heās like, āYeah, itās fine.ā
Turner Novak:
Okay. And then how did that go, the whole process of doing YC for someone whoās never done it before?
Jacqueline Cheong:
It was great. It was great. It was such a great learning experience. I spent the whole batch just learning to sell. And because we had spent six months before roughly just building the product... And data infrastructure tools, it takes a bit of time to build. So thank God we actually had that six month lead time and I spent the whole batch really just selling the product, which I think helped a lot.
Turner Novak:
You didnāt have to go zero to one on the first sale in the 13-week YC batch. You were able to go zero to 0.8 or something, and then you go 0.8 to one through the course of YC. Something like that.
Jacqueline Cheong:
Yeah, yeah. Yeah, exactly.
Turner Novak:
So you said you spent the whole time learning sales. How did you evolve your thinking? What did you learn through YC or even through today? What have you learned about sales?
Jacqueline Cheong:
So I guess itās very specific depending on your ICP and the buyers that youāre selling to, but mostly what I learned was how to sell to a technical audience that doesnāt like to be sold to, tends to be more skeptical and how do you win their trust? I think thatās the big lesson. And then this might sound like really basic stuff, but how do you do a great discovery call? How do you-
Turner Novak:
How do you?
Jacqueline Cheong:
I mean, I think thereās a bunch of frameworks out there, but how do you make sure youāre asking the right questions? And that depends on your product, but the goal of the discovery call is like, āDo they have pain?ā A very simplistic thing is like, āDo they have pain and is already appropriate or overkill for what they want to do with their data?ā
Turner Novak:
And youāre basically finding out, is this someone who might become a customer relatively soon?
Jacqueline Cheong:
Yeah. Yeah. And when they become a customer, are they going to be happy? And I think you can gauge that within the first conversation, but itās learning to actually not sell and ask a ton of questions and understanding where to dig deeper to unveil either pain or what outcomes they really want.
Turner Novak:
So itās a discovery call or discovery portion.
Jacqueline Cheong:
Yeah. Mm-hmm.
Turner Novak:
How would you take it a step further?
Jacqueline Cheong:
And then you do your demo call.
Turner Novak:
And you donāt demo on the first call?
Jacqueline Cheong:
We donāt. We kind of explain what the product does on a high level, but the first call is really just to understand their situation, their challenges and their pain. And I think it might feel a little bit weird, but it is very much for us to truly understand them, such that the rest of the process, they can learn to trust us a little bit more because then everything else is catered from that. And you never stop doing discovery. When you do your demo, you do deeper discovery. When you meet again, every time you meet, the rule of thumb is, you should always find out something new about your customer.
But then itās building out the rest of the process. After the demo, do we do a technical deep dive with their DevOps and platform and infra teams before we do a POC? How do you run a POC? What do they need? What type of checkpoints do they need mid and end of trial? How do you talk about pricing? All of that and how to multi-thread across different stakeholders and when to do that and when not to do that. Those are all things that Iāve had to learn over the last few years.
Turner Novak:
What was the one you, of all those areas, what was the area you sucked at the most or the place you feel like youāve improved the most and learned the most on?
Jacqueline Cheong:
Probably the discovery call.
Turner Novak:
Oh, really?
Jacqueline Cheong:
I think thatās the most important call.
Turner Novak:
So did you just not do that? Did you just jump in too quickly and not get to know what they were thinking about?
Jacqueline Cheong:
Probably in the first... Yeah, definitely in the first couple of months, but itās about the first 20 minutes you meet someone, maybe itās actually even less. The first five to 10 minutes of meeting someone, how do you effectively make them trust you? And theyāll trust you in that moment if they feel like you understood them. And the only way you can understand them is to ask good questions and then dig deeper in areas that they care about.
I think that takes a lot of knowledge around the technicals, knowing the nuances. And when someone... Someone can tell you a hundred different things. How do you narrow down to the five things that actually matter and dig deeper? I think that takes... It took me a while to get to that point. And then you know what? The other thing is, how do you provide an insight that they didnāt even ask for after hearing everything they want to achieve?
Turner Novak:
So why is that important, providing an insight? Does it make it-
Jacqueline Cheong:
Because you can be a thought partner to them and it helps with the trust. Itās like, āHey, not only did you take the time to fully understand the challenges that Iām facing, now youāve understood it well enough and then have thought of something else that I havenāt thought of before and I think you could actually help me as I think about this new use case that Iām going to do.ā So for example, someoneās like, āHey, I am currently using another...ā I have a Postgres database and Iām using another Postgres database as an analytical warehouse use case. Itās not meant for it, but I am doing that for now because it was an easy decision at the time.
And one thing, instead of just focusing on, āYeah, we can move data between Postgres and Postgres, we could do this. This is how you set up Artie. What else do you have?ā Itās like, āOh, well, I know this works fine for now. If your data 100Xs or 1,000Xs over the course of the next year or two, have you thought of what youāre going to do because that warehouse is not going to hold? Where are you going to migrate it? What are your use cases? Is Snowflake the best one? Is Databricks the best one for your use case or should you actually be upgrading to a ClickHouse and helping them walk through... It wasnāt even what they asked, but itās helping them walk through what the future could look like.
Turner Novak:
So itās almost like consulting a bit of like, help them with problems, even if itās unrelated to Artie, it sounds like.
Jacqueline Cheong:
Yep. Yep. I think itās all enterprise sales is consultative and getting to that level where you can have a really good understanding and then consult, takes a lot more experience.
Turner Novak:
We havenāt even talked about this until now. You just announced as of, well, weāre recording this, youāre going to announce it tomorrow, but when this came out, itāll be a couple of days ago. You just announced, you raised Series A. So what did you just announce and what are you doing now?
Jacqueline Cheong:
Yeah. So we raised a Series A led by Standard Capital. So itās Dalton Caldwell, Paul Buchheit and Bryan Bergās Series A fund.
Turner Novak:
And they were partners at YC for a very long time.
Jacqueline Cheong:
Yes, for a very, very long time. And they have a very deep understanding of actually developer tools. So weāre really, really excited to partner with them and for them to be a thought partner as we grow and scale or go to market. But with the funding, weāre really excited because we are going to invest even deeper into the core infrastructure. We already started, but weāre going to expand our integrations to other sources and destinations where real time matters and then obviously, growing the team. So weāre hiring across engineering, sales, BizOps, support, marketing across the company, to help us achieve our mission faster.
Turner Novak:
And so I think it might be interesting for people to hear more about Standard Capital. I feel like at this point thereās been a couple announcements. People are like, āOh, interesting.ā Because I feel like when they first came out with it, there was like a, āI wonder if any founder will work with them or what type of founders will work with them.ā
Jacqueline Cheong:
Really? Okay.
Turner Novak:
Well, it was a new model, right?
Jacqueline Cheong:
Yes, yes.
Turner Novak:
So whatās the model that they do thatās kind of different from maybe the other Series A funds that you talked to?
Jacqueline Cheong:
Yeah. So number one, itās an application. So itās very similar to YC, very different questions, but itās an online-
Turner Novak:
Itās the same vibe, kind of.
Jacqueline Cheong:
Yeah. All the questions are online. You put in your application, you have to tell them how much you want to raise from them, at what valuation. And then, they ask you a series of more post product market fit questions. It was very different from the YC questions. And the whole thing, I think, literally lasted just under two weeks. So you submit the application by whatever deadline they have. You get a first interview thatās 20 minutes, which is double... YC was 10 minutes, so this is double the length. And then if you make it to the second round, theyāll meet you in San Francisco for a 45-minute second round meeting where youāll go much more in depth into your business and how you think about certain things, and you meet the three of them with your co-founder. And then literally, I think it was the next day or the day after, you find out if you got in.
Turner Novak:
And they just give you a yes or no, basically?
Jacqueline Cheong:
Pretty much. Yeah.
Turner Novak:
Yeah.
Jacqueline Cheong:
Yeah.
Turner Novak:
Because itās definitely different from... How would you describe the Series A process with other funds you were talking to? How was that so much different?
Jacqueline Cheong:
Well, I was told the average Series A process can be three to six months. Is that right?
Turner Novak:
Maybe. Yeah, it just depends.
Jacqueline Cheong:
Yeah, it depends. And you have to do, get warm intros.
Turner Novak:
Yeah. They wonāt talk to you unless you get introduced by someone.
Jacqueline Cheong:
It is a little bit weird.
Turner Novak:
Yeah.
Jacqueline Cheong:
Yeah. You have to do a warm intro. I was actually sitting there. I was like, āDo I need a warm intro to someone that I already know?ā
Turner Novak:
That Series A investor that youāre meeting, they donāt have an application, but thereās 50 people that they might meet and they get, weāll just say in one day, Iām exaggerating this a little bit, but thereās 50 potential warm intros, but 20 of them, two people mentioned it. And for four of them, three people, and thereās one Series A company that six people were like, āHey, have you met blah, blah, blah yet?ā
So when youāre just squaring that up and youāre doing meetings all day and calls all day, youāre like, āAh, six people told me I should meet these guys, this one, only one person. And I highly value these six people and Iāve never actually met this one person before or I met them 18 years ago when we worked at Yahoo together.ā So when you just kind of think about how theyāre trying to figure out, Iām going to spend time on things, six people that I trust said I should meet this founder. So maybe thatās the one that I will pick and prioritize.
Jacqueline Cheong:
Yeah.
Turner Novak:
So it can be really hard to fight through the noise.
Jacqueline Cheong:
So yeah, you spend basically a little bit of time, you have a spreadsheet and youāre like, āThis investor, who are all the people that know them?ā And then you go through each one and youāre like, āWhich one is the best one to ask for an warm intro?ā And then anyway, you finally get the warm intro and maybe you do a coffee chat.
Turner Novak:
And they might say like, āJacqueline, great to meet you, adding Anna to find time to chat next week or in two weeks,ā or something.
Jacqueline Cheong:
Mm-hmm. Yep, yep. And maybe itās a coffee chat, maybe itās a Zoom call. And then you just donāt... The process is different across funds. So it can be two, three meetings, it can be six, seven, eight meetings until you get to an answer, and it can be one to two weeks between meetings. Who knows? Itās just a lot more of a black box and it goes on for a lot longer. Standard is just like a... Thereās a formula that they follow and they actually follow it. It actually is after the deadline within two weeks, everybody knows, itās a yes or a no. And you donāt have to...
Turner Novak:
Wonder.
Jacqueline Cheong:
You donāt have to wonder, oh, thatās the other thing. If a VC says no, itās probably just, theyāll just ghost you.
Turner Novak:
Yeah. Thatās true.
Jacqueline Cheong:
So itās not a hard no, so you have to deduce that and figure it out.
Turner Novak:
And then randomly three weeks later theyāll be like, āOh, hey, I was on vacation, but we should catch up.ā And itās just because you got a term sheet from someone else.
Jacqueline Cheong:
Yeah. Yeah. Yeah.
Turner Novak:
Yeah. And so, but you could say, these are just kind of three guys, no oneās heard of them before. Thereās some funds that they have a bunch of people on the team that will help you and theyāll join your board.
Jacqueline Cheong:
Mm-hmm.
Turner Novak:
How did you feel like Standard Capital matches up against maybe what you might get from a board member who will show up to your meetings and add all this value to you, and they have a platform team with all these people that will help you?
Jacqueline Cheong:
Yeah.
Turner Novak:
How did you think through that?
Jacqueline Cheong:
Very different. So thereās no board. They donāt take a board seat. What they do is, every quarter, you do group office hours.
Turner Novak:
This is with other, because they do batches, right? Itās like, we funded six companies in a batch kind of.
Jacqueline Cheong:
Yes. So every quarter theyāll fund around five or six companies and then each quarter theyāll, I think presumably theyāll move people around, but youāll have a group of founders where youāre doing your board meeting too.
Turner Novak:
So itās just a bunch of other founders come to your board meeting?
Jacqueline Cheong:
Mm-hmm. And the idea is, you learn a lot faster when you hear about other companies that are roughly in the same spot. What are the challenges that theyāre facing? How are they thinking about solving it? Because youāre probably like, itās similar-ish problems.
Turner Novak:
So have you gone to some other board meetings now or not yet?
Jacqueline Cheong:
Yeah, yeah. Weāve done one so far. It was with the entire group because it was just the first cohort.
Turner Novak:
Yeah. Did you all do all your board meetings all at the same time?
Jacqueline Cheong:
Yeah.
Turner Novak:
Okay.
Jacqueline Cheong:
You literally go up one by one and you present your, you have 10 minutes to present, what are your challenges, what do you need help on? And then we all talk about it together. And yeah, I mean, I donāt know what a normal board meeting looks like, but I thought that was really helpful because we all are roughly the same, having the same problems.
Turner Novak:
So itās all different products and maybe industries and markets, but weāre all working on the same types of things?
Jacqueline Cheong:
Yeah. Itās like, how do we build a team? How do we do... Weāre all hiring salespeople now. How are you thinking about that? What does marketing look like for all our different companies? The things that you donāt think of pre-product market fit that is necessary now, weāre all working together and brainstorming. And then we all have a Slack channel and you can talk to all three of them whenever you need advice. You can choose to do recurring monthly chats with Dalton or something like that. And because their whole model is like, āWe donāt need to be on your board to be helpful. We can just be helpful.ā
Turner Novak:
Yeah. Iāve always had this weird personal feelings about for my strategy, do I need to be a board member or whatever? And thereās also like, man, that kind of seems like a pretty big burden of like, got to be on the board of a company. Iām on the board of one company and in one aspect, I donāt do anything. I just text the founder a lot, but I donāt actually do anything. Youāre on the board and you have to sign and approve things. You have to read a document. And we do board meetings, but theyāre not... Itās on Zoom at this point. Itās like one of those things, it sounds really daunting and like you do a ton, but also, you donāt really do anything at the same time. And you donāt necessarily have to be on the board to do the same things.
Jacqueline Cheong:
That was what I was going to bring up. If you werenāt on the board, can you still do all those things that you do as a board member?
Turner Novak:
I have your phone number and we just text about random things.
Jacqueline Cheong:
Yes.
Turner Novak:
Maybe if I was like a board member, thereād be more pressure of, you have to actually do the thing.
Jacqueline Cheong:
Oh, like proactively text.
Turner Novak:
Yeah. But I donāt know. I feel like I do try to, when you do ask for something, I do try to help as much as I can, but also, what context does an investor have of the business to truly actually help? Thereās maybe a couple cases where you can actually move the needle like, āHey, trying to do a podcast to announce a Series A, can I come on your podcast?ā Maybe thatās a place I can actually tangibly move the needle, but also youāre like, thereās always recruiting. Everyoneās like, āOh, we help with recruiting.ā
And I think thereās, maybe you sit down with them and think through what a process might look like or maybe itās we have a recruiter on the team that works with you or I know someone who maybe could be a good fit and Iāll just convince them they should join. Iāll help you sell it to them. Or maybe sometimes I see that the investor will try to help close the candidate that the founder and the team worked on. So thereās all these different elements of ways you can kind of help with things, but again, you donāt have to actually be on the board.
Jacqueline Cheong:
Yeah. I mean, our best investors, they all help with customer intros. Like you said, we have a candidate way down the funnel. We gave them an offer. Weāre competing against this other offer that they have from a startup. Yeah, can you talk to some of our investors? And those are all very helpful things and referrals. I think an investorās network is probably one of the most helpful things there. And then pattern matching, right? If you guys have invested in 100,000, maybe not 100,000 companies, but 10,000 companies, you can kind of tell us what path weāre... Weāre thinking about doing this and youāre like, āOh, Iāve seen this fail 99 times out of 100 and this is exactly why this doesnāt work out.ā Itās still good context to have. We may still decide to do it because weāre a little different from that context, but itās good context to have so you go in eyes wide open.
Turner Novak:
Yeah. That seems to be the value prop of YC is like, āHey, weāve actually had three YC companies that did exactly the same thing, and you can talk to the founders of why they failed or someone actually did figure this out, talk to them and see what they can do to help.ā
Jacqueline Cheong:
Yeah.
Turner Novak:
So maybe last question, do you have a favorite founder, CEO, or business that youāve learned a lot from or gotten inspiration from just when it comes to building Artie?
Jacqueline Cheong:
Yes. I think one of the best... Heās no longer alive. Heās an author. One of the best books Iāve read in the last couple of years is, The Score Takes Care of Itself.
Turner Novak:
Iāve heard of that before, but who wrote it?
Jacqueline Cheong:
So I believe someone had to take over and write it for him because he passed away, but he was the coach for the 49ers and it was during a period of time when the 49ers were doing a really terrible job.
Turner Novak:
This is Bill Walsh?
Jacqueline Cheong:
Yes. They were the worst team in the NFL for multiple seasons. And he took over and within, I think it was within two seasons, they won the Super Bowl.
Turner Novak:
Wow.
Jacqueline Cheong:
And then they continued to win the Super Bowl for many years after. And itās his philosophy of how he ran his team and his standard of performance and how he implemented it. And a lot of it is directly translatable to running a business.
Turner Novak:
What were the biggest takeaways from the book?
Jacqueline Cheong:
I mean, the biggest, biggest takeaway is the title. Itās like, the whole thing is about if you just focus on everything that you can control and you make sure that every single person on your team is performing to the highest standard of performance and you have your standard of performance strictly written out for everybody. It was not just the players, not just the coaches, but up to the people, like the janitors that worked in that building had a standard of performance. And if you control all of the inputs that you can control and you do a really good job, you donāt have to care about the outcome. The outcome will just happen.
Turner Novak:
And thereās a lot of takeaways obviously for building a company with that.
Jacqueline Cheong:
Yeah. Yeah. Itās focusing on all the most important things, focus on what you can control. Donāt think too much about whether or not this deal closes or this product is... If you focus so much on doing all the right things and youāve done work that youāre proud of, at the end of the day, you donāt have to think about whether or not it was a good or bad outcome. And then in the long run, itās almost for sure that youāre going to have a good outcome rather than a bad one.
Turner Novak:
Oh, makes sense. Well, Iāll throw a link in the description for the book if anyone wants to read it.
Jacqueline Cheong:
Yeah.
Turner Novak:
Well, this a lot of fun. Thanks for coming on the show.
Jacqueline Cheong:
Yeah, thank you for having me.
Stream the full episode on YouTube, Spotify, or Apple.
Find transcripts of all other episodes here.
Brilliant breakdown of why data infrastructure remains such a stubborn problem at scale. That stat about 95% project failure in streaming hits hard becuase most companies underestimate the edge cases until production. I've seen teams spend months buliding what looks like a clean pipeline only to discover sharding drift kills accuracy downstream. The cold email customer acquisition strategy is lowkey genius too.