Kelly Kohlleffel (00:00)

Hi, folks, welcome to the Fivetran Data Podcast. I'm Kelly Kohlleffel, your host. On the show, we bring you insights from top experts across the data community. We cover AI, machine learning, enterprise data, analytics, and much more. Tim Veil is a Principal at Two Bear Capital focused on identifying new technology-focused investments, advising existing portfolio, go to market teams, and supporting 2Bear's data analytics and enablement efforts.

Prior to joining Two Bear, Tim was the VP of Solutions Engineering and Enablement at StarTree, a real-time analytics platform. And before this, he was head of sales engineering at Cockroach Labs, creator of CockroachDB, a cloud native distributed SQL database. Tim, it is great to see you. Welcome into the show.

Tim Veil (00:47)

Hey, Kelly. Thanks for having me. I'm looking forward to our chat today.

Kelly Kohlleffel (00:52)

I am as well. We've had a chance to catch up a little bit over the last few months, so this should be a lot of fun. And you've had a lot of experience, a lot of different, I'm guessing, lessons that you've learned over your time at these organizations. And would love to hear how some of those lessons could help inspire data leaders today across different industries. Why don't you give us a quick overview of your background? I dug into it a little bit, but a little deeper dive and then your current role.

Tim Veil (01:20)

Sure. Yeah, absolutely. So my background, I started really writing code for a living. So you know, I consider myself kind of a software engineer by trade and kind of rose up through the ranks doing a whole bunch of different kinds of development, primarily using Java, but really all sorts of other languages, you know, did a number of different kinds of engineering leadership roles and ultimately decided at some point — and I'm glad I made this decision to kind of jump to the dark side and move away from kind of writing code for a living and actually trying to sell the things that I was building. So, you know, as you correctly pointed out in my background was a sales engineer as my first job, kind of away from professional engineering and then, you know, rose up through the ranks and ultimately leading sales engineering organizations, which ultimately brought me to where I am today, which is Two Bear Capital.

You know, Two Bear Capital is a venture firm that focuses kind of on the intersection of life sciences and technology. And I'm helping support the technology investments at Two Bear Capital. So an interesting journey, but I think the highlight being, you know, really grew up writing code for a living. And to the extent that I can, I try to do it nearly every day. I really enjoy being very close to technology and writing software to this day.

Kelly Kohlleffel (02:33)

I have, I've had a number of conversations recently about software engineering today and how that's even changing so, so quickly. If you're, let's say you're starting today as a software engineer, you want to get rolling. What would you do differently than, what you may have done years ago when you started?

Tim Veil (02:54)

Gosh, the whole world has changed, hasn't it? Just in the last couple, I mean, I almost wanna say months, it's obviously been a lot longer than that, but you can see behind me, at least for those of you on video can, I've got a ton of books and that's how I learned. I remember this distinctly. I remember trying to teach myself the Java programming language and sitting in front of Java For Dummies or one of these books and kind of pouring through on how to write things.

That was tough. You know, it was a really tough way to get started. And it amazes me the amount of information that people have at their disposal today with AI and other things. I mean, you know, we could probably spend an entire podcast alone just talking about, you know, how AI can accelerate software development. But I do still think there's tremendous value in really learning these things the hard way. And by the hard way, I mean, not only reading and absorbing the knowledge, but actually writing code. I think that's a really important thing. I mean, is AI gets more and more powerful. I think there's an opportunity for AI to do more and more of the development for you, just more and more of the thinking for you. And I think that might be unfortunate. If I were coming up in the software engineering world, I think I'd want to still focus on learning and doing things the hard way.

Kelly Kohlleffel (04:10)

There's such an emphasis now on being really, really good at prompt engineering and kind of getting there quicker if you can. I think, know, data engineer, software engineer, anything that's working with code, I think, yeah, if you're doing that full time, there's a need to kind of rethink what you're doing. I was even having a conversation about consulting companies. If you're starting a data consulting company today, I mean, you almost, you really have to re-is this the same as I would do it, say five years ago even?

Tim Veil (04:42)

I don't know, it's a tough question to ask. Certainly being on the VC side now, we spend a lot of time talking to people who are attempting to build businesses around AI and certainly our portfolio companies are being transformed by AI. I mean, it really is changing so much about what we do that I think, I don't know that we've hit the full potential. I'm still surprised to some extent by the amount of people and the amount of companies that aren't properly leveraging it to do the work that they should be doing.

So I think it's very much a balancing act. think it has to be part of kind of the tools that you use on a daily basis. If it's not, you're probably doing something wrong, but I don't think we're at the point yet where, you know, it's gonna replace kind of human intuition and experience and the things that I think we're really and uniquely good at. Therefore, you know, back to my previous comment, I think it's important to really build kind of a library of personal experiences, you know, whether that's like technology or, you know, just working with people.

Because you I don't know if that's what we're going to talk spend a lot of time talking about today, but certainly people are big a big piece of the picture in building successful organizations

Kelly Kohlleffel (05:50)

Absolutely. And I want to, before we kind of move off of this segment or this topic, you mentioned that your focus primarily has been on technology portfolio based companies on the VC side. Any particular area on the technology side that you personally or your firm has been more focused on recently.

Tim Veil (06:11)

Yeah, think there is. you know, as I said, Two Bear focuses on this intersection between life sciences, or maybe a better way to say it is about 50% of our firm is focused on life sciences. You know, emerging therapeutics from everything to neurodegenerative disease to cancer. Obviously, that's not my specialty or strong suit, but about the other 50% of our portfolio is in technology. And I think within the technology, that's obviously a very, you know, broad space.

We're making a lot of investments in security. There aren't our only investments. We've made some investments in the data space and a whole host of other things, AI certainly. But I think where we've spent a good bit of focus when you look at across our tech portfolio is really in the security space, just because as more and more people are doing more and more things online, and certainly with AI and agentic AI and all those things, you know, really understanding and being able to manage the security vulnerabilities across the network is becoming increasingly important. So we're spending quite a bit of time making investments in that space, at least in our tech portfolio.

Kelly Kohlleffel (07:17)

Yeah, no, it sounds like a sounds like a fun, fun role, a lot of challenging, interesting problems to help solve. And I'm, interested in, in this too, with all these organizations that you get exposed to, there's got to be a lot of different ways that data architecture is thought about, approached, actually implemented, everybody has to use data. I mean, I just don't think there's an organization that is not.

What are you seeing out there? What are…any big shifts? Are there any thematic elements that are kind of coming into play right now that you've seen over the last two or three years that everybody's taking advantage of?

Tim Veil (07:53)

Well, you know, I think there, a lot of what I'm seeing are things, different variations of the things you and I have seen for many, many years. But I think, you know, as reflecting on this conversation, I think one of the big changes for me is that the ease at which, you know, people are able to build these kinds of pipelines of data and be able to gain insight out of data has really changed in the last couple of years. You know, back in the day, you know, to do large scale data analytics, you had to invest in technologies like Hadoop and these were very complicated infrastructures and a lot of the pipelining and a lot of the movement, you know, from point A to point B had to be kind of rolled your own to some extent. And that was difficult. And so, you know, for small to medium-sized businesses, for startups that are just starting the idea of, you know, building this very sophisticated data infrastructure can be overwhelming. 

And I think one of the interesting things that's changed in the last couple of years with the introduction of technologies like Fivetran and others is the ease at which people can do some really sophisticated things without having to have full-time people or very, very sophisticated teams of people. Because again, if I'm a series A company or a seeds company or really any small to medium-sized business, I may not have the capability to invest in large teams to set up infrastructure to do data and analytics.

And I think just the availability of tools that allow you to do that much easier is incredible now. You combine that with now the things that AI can do sitting on top of data if you have access to data. It's a pretty remarkable place, I think, for companies to be in. We're starting to see, and it's certainly one of the things we're looking for, is for companies to be taking advantage of that. If they're not, if they don't tell a good story about how they're managing their business and outlook with data, that's kind of a red flag. And certainly, that's true on the investment side, but in terms of the work that we do with existing portfolio companies, that's some of where the coaching comes in, is how we can help our companies take better advantage of the data they have to make better informed decisions about their business.

Kelly Kohlleffel (09:54)

I totally agree. Totally agree. Where do the disconnects typically occur when you're talking about value, you're talking about data, you're talking about what an organization needs, regardless of size? Could be one of your smaller pre-IPO C series A companies, could be an enterprise today. Where are the disconnects? What have you seen, I guess, over the years between value and what an enterprise or what a company thinks they need?

Tim Veil (10:24)

I think, you know, when you say disconnect, one of the things I think back to, and I've certainly faced this many, many times in an organization, is I think, and it kind of goes back to what I was saying, is that these companies tend to be moving so terribly fast. You know, there's always an impending deadline. There's, you know, this stiff competition for resources and time. And what I have observed many times over is that people are often making decisions with incomplete data. 

They are making a lot of assumptions and many times, unfortunately, not the world's best assumptions about what they think the data is telling them. And so, you're making important decisions about maybe where to go from a marketing campaign or other really important decisions at the business or strategic level, and they're being made without a full picture of data. And I think that's one of the things that I work or trying to work a lot with our companies and work at the past is helping companies kind of gather data from multiple sources, put them in a single place where we can begin to do some reporting and analytics and surface that data in such a way that it becomes part of the decision-making process kind of in a natural way. 

Because again, I think people are moving so fast, there's this pressure to make a decision, make a decision, and people tend to just react based on what they think or assume to be true. And I think what I have found is the more we can begin to infuse data into those conversations, the better decisions people are making, the better outcomes we get. But that's been a challenge. Just again, not because people don't necessarily want to. Nobody's going to sit down in a podcast, for example, and say, no, I don't want to make decisions based on data. That doesn't make sense. Of course, everybody wants to do it. But it's very hard to do it in practice when things are moving so quickly. And I think that's part of the work that we try to do is just say, let's slow down to ultimately go fast. And by slowing down, what we're going to be doing is analyzing the data we do have and make the best decisions with that data.

Kelly Kohlleffel (12:25)

Yeah. Now, well, you've, you've had a couple of things I want to, I want to, you, you and I got connected back up when you reached out on some projects that you had going on internally. And one of the things I wanted to talk to you about today was a Fivetran Connector SDK. We a lot of most of the time on this podcast, we don't talk about five-trans specific capabilities or features or anything, but

Tim Veil (12:47)

Yep. Sure.

Kelly Kohlleffel (12:55)

I was really interested in your story. This is a relatively new offering for us. And I've been personally really, really excited about this. You say, okay, well, Fivetran, got 700 plus connectors out of the box. To me, this just opens up thousands and thousands of potential sources. And, I'd love to hear from you, Tim, what, what were the challenges? What happened with you that said, Hey, I need to explore building a custom connector and why Fivetran Connector SDK.

Tim Veil (13:19)

Yeah, absolutely. So I'll give you just a little bit of a highlight of kind of what problem we were trying to solve. And it kind of part and parcel to the things we previously discussed. And we're doing this actually at a couple of our portfolio companies. We're also doing it internally at Two Bear. And that is that data lives all over the place, as we know. It's tied up in all sorts of different systems. And I'm a firm believer that when you can pull that data out into a single place, you just have the ability to make better, more informed decisions.

And I'm not even talking yet about the possibility of, you know, pointing AI at things like that. I think there is just tremendous value in taking data across, you know, a handful of key systems that we tend to see time and time again and putting them in a place where we can do singular analytics. 

And so the kinds of systems I'm talking about are CRMs, you know, whether it's HubSpot or Salesforce or in our particular case, Affinity. And then, you know, a whole host of other systems, you know, at our portfolio companies, oftentimes,we wanna pull data from things like JIRA or ticketing systems, customer success systems, but at Two Bear, I guess I'll use the Two Bear example. In addition to wanting to pull data out of things like Salesforce for fundraising and Affinity for some of our investment stuff, we also wanted to pull data out of tools like Standard Metrics and Carta, which have a lot of interesting information about the health of our portfolio companies.

And again, the driver for doing this, sure, I could pull up each individual website and look at a dashboard that each individual one of these tools presents. But what we really wanted to do was take this data out of those systems and make them available so that we could do kind of a single pane of glass, a single series of dashboards in order to do kind of executive level or investor level reporting. And that's kind of a difficult problem to solve. 

And I didn't want to get in the habit of writing as, you know, a whole bunch of manual code, which are things I used to do in the past. So, Fivetran has been wonderful for some of the more well-known connectors, but where we saw a drop-off was some of the lite connectors and then places like Carta where no connector existed at all. But we really loved the kind of the pipelining and some of the capability of Fivetran. So you and I got connected, and you introduced me to this idea of, why not just build your own custom connector? And at first I'm thinking, you know, is this really what we should be doing? Turns out it was.

Kelly Kohlleffel (15:45)

Now I remember your one of your first questions. Wait, can I do this in Java?

Tim Veil (15:51)

Yeah, I did. I did. I did. I only knew Java when I started this, by the way. 

Kelly Kohlleffel (15:56)

Right. My answer was no, by the way.

I don't think you like that answer too much.

Tim Veil (16:02)

But anyways, yeah, the answer was no. It's Python only. And I didn't, but, but. So anyway, you know, after playing around with some of the light connectors and really, you know, investigating and spending a lot of time on what it was we were really trying to get out of these source systems and the way we wanted that data transformed and stored, it seemed to me like building a custom connector was the right way to go.

So I, because I was an engineer by trade and still am, forced myself to learn a little bit of Python. But it turns out it was super easy to do. And I'm really glad that we connected on it. And I went down that path because what it's enabled us to do or me to do is have incredible now fine-grained control on how the data, how I'm taking data out of these source systems and moving them into, our particular world internally, our data warehouses, BigQuery.

In fact, and maybe I shouldn't share this, but I will. When we were using kind of the out of the box lite connectors based on kind of how they were implemented originally, it was costing us nearly a thousand dollars a month to pull that data out and store that data. And why? Because it was pulling a lot of data I just simply did not need for the kind of analytics that we did. And so I won't tell you how much we're spending now, but I will say that when I implemented the custom connector, which was very easy to do, I was able to significantly reduce our cost, both on the storage side and the processing side, because I was able to really pull out just what I needed. So very, very simple to do. 

Since you and I talked, I've built three connectors, a Carta connector, an Affinity connector for our purposes, and a standard metrics connector, and they're great. It's been very easy to use, and it's been able to pull exactly what we need, dump it in a way that's easy to interrogate. And now essentially we have a single pane of glass serving our investment professionals. And that data covers everything from our existing investments to our pipeline from both fundraising and investment. So it's been fantastic, and I'm glad we connected. I'm glad you suggested it. And really I'm glad I learned Python. It may be my new favorite language. I don't know what I'm gonna do with it, but yeah.

Kelly Kohlleffel (18:15)

That's great. no, you're converted. You converted. Okay. Well.

Tim Veil (18:17)

I think I kind of am. It's kind of nice. Kind of nice not to have to do all that other stuff.

Kelly Kohlleffel (18:22)

Yeah. Well, so you mentioned lite connectors. We're talking about Connector SDK, just, just to set the stage for the audience. So, so Fivetran has got, you know, call it standard connectors. Think of Salesforce database connectors kind of do everything under the sun. And then we have something called lite connectors, which really are targeted at a particular use case. These are first-party connectors. Fivetran supports them end to end, from interaction with the source all the way through interaction with the destination.

What Tim's describing, connector SDK, Tim and team have coded up the source interaction side, how Tim wants to interact, like he said, very specifically with Affinity, Standard Metrics, or Carta. And then Fivetran handles the middle part, any normalization, data typing. I don't know how you built your connector, but Fivetran can infer data types and columns and all that kind of stuff, certainly with destination interactions.

There's a ton of automation to your point earlier when you and I were in Hadoop space, everything end to end had to be coded up, the source interaction, the middle normalization all the way through to destination interaction with Hive or H base or whatever it was. Now you get to really create your own experience almost with the source interaction that you wanted, because you, as you said, the lite connector, for instance, with Affinity just didn't, it did not do what you wanted it to do. The rest of everything else is automated. Talk to me a little bit about that only having to do kind of the one third of the work in having Fivetran handle the other two thirds.

Tim Veil (19:58)

Yeah, no, it's really important. And I have some, you know, I have some personal experience with this. I mean, you know, in my, at both my roles at Cockroach and StarTree, you know, prior to using Fivetran, I was essentially rolling my own platform for this. So I'm very, very familiar with kind of the work that it takes to kind of write the full stack. I did it in Java, of course, which, you know, at the time that was, that was my favorite, but yeah. So, you know, there is quite a bit. I mean, you brushed, brushed over it, but there is quite a bit that you need to do in order to kind of orchestrate this data movement if you're building it truly yourself from end to end. So the nice thing about the API that you have via the connector is that it does, it shields you from a lot of that nasty complexity and allows you to focus really on what specific things do I want to pull from an API and how do I want to store that? And all that other complexity is shielded from you. And that's actually pretty important. So it really allowed me to focus on the things that matter, which is what APIs from these systems do I want to consume? How do I want to, if at all, transform that data into something useful? And then just kind of pushing that down, not even really being terribly specific about how it's pushed down, by the way. The connector has some really nice capabilities there. It's just like essentially update here. Here's a JSON record. 

So it's made it very, very easy to do and, yes, has enabled me to, you know, essentially implement one Python file as opposed to a very, complex Spring Boot project, which, you know, did the whole kit and caboodle. That was fun to write by the way, but yeah, from a maintenance perspective, it's not the easiest in the world.

Kelly Kohlleffel (21:39)

Well, this also is, as you continue to grow the team, I would think from a maintenance perspective, maybe refactor whatever needs to happen at that source interaction side with affinity or Carta or any other connectors that you develop should be a little bit easier to pick up too, especially since I just have to deal with this one piece, not the end-to-end.

Tim Veil (22:01)

Yeah, and there's something that maybe others can appreciate, and I certainly have. Part of it, too, has to do with, I think, the sophistication of the team around you, or the skill set of the people who might be consuming the data model that ultimately ends up in your data warehouse. And what I mean by that is I have observed that a number of the APIs that I'm working with, some are more complex than others.

And I think you kind of have two choices as an SDK developer or a Fivetran SDK developer is do I take this complexity from the API source system and just kind of pass that right on through into the data warehouse or wherever my data store and let somebody else worry about it? Or do I begin to kind of reshape the data as I'm pulling it from these APIs into something that's a little simpler? And I chose to do the latter partially because I'm not the world's greatest SQL expert, but it seemed easier to me to do a bunch of transformations kind of at the connector stage. 

Now this may fly in the face of some best practices and paradigms that you guys talk about, but not all APIs are created the same. And so what I tried to do is simplify a little bit as I was taking some of this data and making ultimately a much easier to consume schema on the data side so that you know, my consumers didn't have to write uber sophisticated SQL in order to consume the data, but could, you know, operate with a data model that, that was a little bit more sensible. 

So that's one of the things I really appreciated about kind of implementing the custom connector is it gave me a lot of flexibility, I think, to kind of dictate how sophisticated my downstream consumers needed to be.

Kelly Kohlleffel (23:38)

I love that. I love that. Great, great, great story. Let me transition just a little bit. It could be on the topics we've already talked about, but just interested in some personal perspectives. Any lessons learned, you know, kind of the running into brick wall type thing maybe that you've had, or maybe better experiences in that over the years that could apply to you're talking one-to-one, one-on-one with a data leader today. What are those one, two or three things that you go, this is, these are the things that I, I learned that I would either not want to repeat or always repeat.

Tim Veil (24:14)

I think I touched on it earlier, and this is something I see a lot of organizations struggle with. I don't think there's any grand revelation here. It's that it's never too early to start collecting and analyzing data. I get this pushback sometimes, and I've been part of organizations that pushback on this. You know, we're not big enough. We only have a handful of opportunities or a handful of customers or whatever. The excuses are kind of the same.

And I just remember, I remember the early days at Cockroach, or maybe not, you know, kind of the middle days after we'd been there a few years, I remember we had not done a particularly great job of collecting some of that data early on. And I remember thinking, if we could just go back and begin to collect some of the things that we forgot to collect or didn't collect, how much more valuable would our data set be now? How many lessons could we avoid relearning, you know, by having access to some of those? You know, those loss reasons or whatever the details were. 

So that's, you know, that's one thing I really work with teams on is I know it feels like maybe it's the cart before the horse. know maybe there's this, this idea that we're asking you to, you know, implement process or tooling that is better suited for a company that's later stage. But I just, this is time you can't get back. This is data you can't get back. And there can be tremendous value in capturing and analyzing this data now. And I think that's something that we learned the hard way at a couple organizations, quite honestly. So that's one of the big ones that I really push on is let's absolutely take the time to do this right now. You will thank me later.

Kelly Kohlleffel (25:52)

That's fantastic. Great, great advice. What about obviously we've we've hit on AI a lot today, any any other emerging trends, whether it's in data, whether it's in the PE space, or whether it's in this, the this intersection that you're working in with tech and life sciences that you see, trend wise emerging, especially as it might relate to data leaders to help them prepare as we continue to roll through 2025.

Tim Veil (26:21)

Well, security is going to continue to be incredibly important. Data security, as we move away from human identity from a security perspective and introduce more non-human identities into the picture, agents and things like that, companies are really going to need to have an understanding of who's doing what in their systems and that who may not be a person, it may be an agent. 

And so I think one of the things that we're looking at very seriously is how do we secure this future in which the people that are interacting with the systems may be very, very ephemeral and may not be human at all. And what does that really look like and mean? So I think security continues to be important, but also just monitoring. You know, AI costs are going to continue to rise. And with any new and emerging technology, takes enterprises a while to get a handle on things like spend. How much am I spending? Where am I spending? Am I getting real value?

And so we're looking at ways that we can help companies keep an eye on those things. Because I think that's going to be interesting. AI is going to transform so many businesses. How exactly and when is yet to be seen. But we think it's going to be absolutely transformational. And so the things that we can invest in that help secure data now and those environments in the future are going to be pretty important, we think.

Kelly Kohlleffel (27:50)

Tim fantastic getting to catch up a little bit today. Thank you so much for joining the show. I really, really appreciate it.

Tim Veil (27:57)

It was great to be here. Enjoyed it.

Kelly Kohlleffel (28:00)

Absolutely. And definitely let's stay in touch on everything that you're doing at Two Bear. I want to hear about those next five or six connectors that you build out as well. 

Tim Veil (28:08)

You'll be the first to know.

Kelly Kohlleffel (28:09)

Thank you so much. And thank you to everyone who listened in today. We really appreciate each one of you. We'd encourage you to subscribe to the podcast at any of the major platforms, Spotify, Apple, Google. You can also find us on YouTube. Visit us at Fivetran.com/podcast. And please send us any feedback or comments at podcast@fivetran.com. We'd love to hear from you.

See you soon. Take care.