DataMastersCDO VIDEO INTERVIEWS

Subscribe to get the latest CDO Interview delivered to your inbox.

Voya Financial, SVP, Head of Data Management, Architecture and Salesforce Development: Developing Enterprise Architecture Road Map in Alignment with Business Priorities

 

Julia Bardmesser

SVP, Head of Data Management, Architecture and Salesforce Development at Voya Financial

(US & Canada) Julia Bardmesser, SVP, Head of Data Management, Architecture and Salesforce Development at Voya Financial together with Andy Palmer, CEO and co-founder of Tamr Inc talks about dealing and managing data lineage and provenance in their projects. They also discuss cultural shift as the biggest challenge in data and the excellent time to progress in public cloud.

Transcript

Andy Palmer:
It’s really a pleasure to see you and it’s really spectacular to connect with someone that is such a pro when it comes to data. Your career is really remarkable, from Bloomberg on, you’ve always been working at the forefront of data. Could you just take a few minutes and tell us about your career and how you ended up getting from Bloomberg to where you are now and kind of your story?

Julia Bardmesser:
So I got hired into Bloomberg out of NYU. I was doing master in economics. And back then, Bloomberg was not a well-known company at all. It was 90 people and they were hired in their first training clause of people right out of college that wanted to play in the way Bloomberg was doing things. First they had a very, very unique environment in terms of using foreground, using proprietary database package, using a very unusual or not widely known aggregation system. So they hired specifically people not with computer science background and then taking them through the training program. And that first training program I was in, they had a few after that, so that was a very successful hiring mechanism for them.

Julia Bardmesser:
And at the end of the training program, I was assigned onto the mortgage team, and there were two of us out of the training program that got assigned and basically completely randomly. I got assigned most data-related tasks and he got assigned more security tasks. So I kind of fell into data. It was not a grand design. It was not anything I was thinking about that, you know, this is what I want to do with my life. I have no idea why they decided to do it that way and not that way at all, but I got assigned a lot of, as I understand now, back then I didn’t have those words and it was not really in a practice around data integrity of most of the databases that Bloomberg had.

Julia Bardmesser:
And from then on, and I’ve done some UI work as well, and then as I went on and I transitioned to Bear Stearns and then Thomson Financial and Freddie Mac. I did a lot of data work, but a lot of it was outside of traditional data paths for people, because data warehouse and data modeling, data analysis, DBA type of work, it was more on how do you effectively use the data to bring applications’ capabilities to the business. I’ve worked at Bear Stearns with support of the mortgage trading floor, the same in security sales, and trading group and Freddie Mac, I was on the trading floor supporting the traders for the first year and a half of my career in Freddie Mac, and it was a lot about what functionality do you build to support the business needs of the traders?

Julia Bardmesser:
And then I went on on there, and again, it was also clearly accidental. I wanted to get into management, and the managing position that opened in technology was managing the data team. There were two positions. One of them was managing, and it’s interesting enough, was managing more of UI team, and the other one was managing the data team, and I didn’t get the job on the UI team and I actually wanted that one more for whatever reasons. I can’t remember the reasons anymore. But I got the job managing the data team and I’ve learned a lot, and from then on, I stayed a lot within data. But again, a lot of it on how do you effectively use data to bring value to your business?

Andy Palmer:
Wow, that’s an amazing story and it highlights so many interesting things that are going on right now. That perspective of the consumption of the data as being the primary thing is sort of antithetical, right? There’s a lot of people in data management that come from the sources and think about how to just bring the data together, and maybe one of my opinions is it’s one of the failures for data warehousing and data marts and data lakes has been, it’s just a lot of data collected but not in context of consumption. How do you think about what’s going on right now in terms of technologies and new capabilities that are coming out and all the new analytic capabilities that are enabled by data science? Do you feel like we’re now thinking about consumption and you have the tools and the infrastructures to deal with consumption?

Julia Bardmesser:
We do have tools in development. So one of the things that I have noticed that a lot of data disciplines, from data pooling to data consumption, data sourcing, data profiling, was very, very immature state until about financial crisis hit. And then with financial crisis and all the regulations that came out [inaudible 00:05:30], it’s famously data-driven to [inaudible 00:05:35] RDA principles, to GDPR, all of that. There was enormous focus on data, and what you had is a lot of focus on the tooling side of it. And you have a significantly larger set of choices around data profiling, and of course all the data analytics tool and data catalogs, which I think is a wonderful set of tools not yet reached the potential in terms of how do you effectively use it. This is actually one of the challenges that I see around data catalog, a lot of promise, but not yet a practice on how to do that.

Julia Bardmesser:
But going back to your question around going from the data consumption, this is something that I have had to explain many times in my current jobs and before was people are really have the idea of enterprise data platforms being built from sources independently of how the data’s consumed. And I’ve been here now at Voya for two and a half years, and especially around the technology, but business and technology people who haven’t had all the scars on my back that I’ve had from trying to build the data from the sources and not succeeding or not really building for any specific usage. All it was, if you grew up in that area have those scars, so people who haven’t had those go, “But you enterprise data. You can just go and bring all the sources in and just integrate it and it’s going to be beautiful.” And I go, “Well, no, not really. It’s maybe beautiful, but it’s going to not going to be very useful.”

Andy Palmer:
It’s-

Julia Bardmesser:
And now it’s unbelievable beauty that has no use.

Andy Palmer:
Yeah.

Julia Bardmesser:
So that still is something that I think we need to communicate and propagate more, that effective enterprise data is building from consumption side. Now what makes it enterprise is where you have a lot of thinking that needs to happen, so it’s reusable. Because you don’t want to build silo. I don’t want to build enterprise silo data. The only way I’m going to build for this specific use case needs to be fully usable by a similar use case coming from a completely different business area, and that’s where a lot of our attention and thinking and pain is coming from, is how do you do that? Because that’s what’s making it valuable.

Andy Palmer:
So you’re really focused on the reuse of the data in context of the consumable data.

Julia Bardmesser:
Exactly.

Andy Palmer:
Not just the raw form. That’s [crosstalk 00:08:06]. So could you tell us a little bit more about the team that you’re leading now and what your most important initiatives are and the things you’re most excited about?

Julia Bardmesser:
I have three teams under me now. So there is enterprise data team. I also lead enterprise architecture team [inaudible 00:08:22], and I also lead sales force development team [inaudible 00:08:26]. That’s a very interesting combination that works amazingly well and I’m really excited about having that combination of the teams under me. Of course, I was hired for my data background, and what attracted me to the job and that makes it incredibly interesting is building basically the video film development for business. So I have done video film development, but what my jobs at FINRA, Citi, Deutsche Bank, it was developing from scratch. And I like my job in Citi and Deutsche Bank, I don’t want to suggest that those were not good jobs. They were very excellent jobs.

Julia Bardmesser:
But a lot of the focus on the data was on the regulatory compliance, which was incredibly important, because if you weren’t able to satisfy regulated requirements, it was a matter of survival for Citi Bank and Deutsche Bank. However, what attracted me significant to Voya and the business that they are building these reusable enterprise data capabilities to advance the business. And to increase the business to understand our customers better, to bring more customers to Voya, but it’s not only for sales, it’s also the mission of where is the time and to improve the time and outcomes for American companies, for people. And so a lot of data being used, how do you get people to invest more into data requirement? How to be more secure? So that’s an amazing mission for the company and it makes it much more interesting to build the data capabilities to do this effectively.

Andy Palmer:
So your mission is not just sort of a technical one, but it’s kind of integrated into the core of the product offering for the entire company.

Julia Bardmesser:
Exactly.

Andy Palmer:
[crosstalk 00:10:11]

Julia Bardmesser:
Just technical missions aren’t particularly interesting. Not that technology isn’t interesting, it is, but it needs to be, again, I’m going back to my roots. It’s how do you effectively use the data to promote business goals, business strategy of the company? And that’s what made the job at Voya very interesting. And so we’ve been building this out. It’s been two and a half years and it’s been a very interesting and challenging journey of establishing these reusable data capabilities. And we are well on our way, and I think we’re doing very well. One of the projects that’s coming up that’s really exciting for me, it’s actually on intersection of all three teams, and that’s data driven distribution. We recently hired an executive on our investment management business to drive data driven distribution for investment management, and that’s where all three of my teams are working together, from enterprise architecture for establishing architectural principles, and they then integration tooling and principles.

Julia Bardmesser:
My sales force team on serum data and my data team to bring a master intermediary and organization data to support our serum strategy and to support distribution on the [inaudible 00:11:26], and that is incredibly exciting. This project has taken off in the next couple of weeks. All three teams are just really raring to go and start working that.

Andy Palmer:
Wow. It sounds really challenging. So what are the things that you’re wrestling with as you do such a big project that integrates so many different people and teams? What are the biggest challenges that you’re going to face and what are you most worried about?

Julia Bardmesser:
Well, as always on data project, and it’s not only at Voya, it’s pretty much any time I’ve had a data driven job, the biggest challenges are cultural. [crosstalk 00:12:03]

Andy Palmer:
The human stuff, yeah, yeah.

Julia Bardmesser:
It’s the human stuff, yes. It’s humans. It’s not technical challenges. I mean, there are technical challenges, but you also have much better [inaudible 00:12:14] to address technical challenges now. It’s a cultural shift. It’s a cultural shift to think about the underlying data, about how we do operations in this business unit that enables reusable data assets. And it goes back, for example, example in Citi, we had, we called it our [inaudible 00:12:37] and apple pie slide, and it showed that every business pretty much has the data they need at the data quality level that they need it. It’s when you want to bring the data together and use it across multiple business lines that when you run into problems where you need data management, data governance, and reusable data asset.

Julia Bardmesser:
But to get there, it needs to be priority change on the business line side, and that’s a human problem. That’s a business process problem, it’s a cultural problem, it’s who owns the data, who’s accountable for data, and accountability from front to back. So that’s always been in all of my [inaudible 00:13:19] the biggest challenge that you have in data, and it doesn’t matter what tools we have. We can have tools and we do have tools to make data quality more [inaudible 00:13:29] and data definition easier to manage, to consume the data once you have it easier to integrate data to use machine learning and AI, to recognize data patterns. All that exists and all that is easier. The culture, that’s hard.

Andy Palmer:
I get it. When I was managing data engineering over at Novartis’ research group, we used to describe every lab with its own island of data, and getting them to share, collaborate, or even align their data was the biggest problem, this human problem. So how do you do that? How do you get people to believe that it’s worth it to change in the interest of serving this broader, bigger data asset?

Julia Bardmesser:
[inaudible 00:14:21] slowly. Slowly. A lot of it is communication. And a lot of it, again, it goes from consumption. So I can go in the beginning, two and a half years ago, when the start of the data foundation initiative and when we made the business case for it to be funded, a lot of it was it’s a mess right now and it’s going to be so much better after we’re done. But a lot of it was what I call [inaudible 00:14:46] there was no proof, it was just words, words, and, of course, the history that I have with the growth of the job, but again. So in two and a half years, you go from consumers and you go here is how… So we built a little bit and now it’s used and it’s making somebody’s life a little bit easier.

Julia Bardmesser:
And now if only on the business side we change decoration process and make it a little bit better, there is a better outcome here. And interestingly enough, because of the pandemic, what’s been [crosstalk 00:15:17] right now, made it actually easier.

Andy Palmer:
Really? How so?

Julia Bardmesser:
[crosstalk 00:15:21]

Andy Palmer:
How so?

Julia Bardmesser:
Everything is becoming much more digital. So analog experience is what probably, well, I don’t know, but if we weren’t in this situation, weren’t all sitting at home, we probably would have been doing this in person, so it wouldn’t be lying on this platform, which I would much rather, mind you, do it in person [inaudible 00:15:46]. But that’s a different conversation. But a lot of communications are much more digital, because people don’t want to… There’s just no physical way. There’s really letters, well, used to be, now that we know the recent transmission of services, it’s not as much, but in the beginning, letters were suspect as well.

Julia Bardmesser:
So a lot of communication is digital. Also the situation was changing, and it’s changing now so quickly that you need to do analytics much quicker. So the whole infrastructure, the reports that used to come out once a month, because it took three days or five days to load the data and then another 15 days to claim the data, and then another 15 days to run the reports. Now a business asks for data much more often, and now you want to go back into history. Majority of the data warehouse… It’s very difficult to build history. This is one of the most difficult things enterprise data warehouse does, figuring out how do you build history? And a lot of existing data assets don’t, and now you want to have that, because you want to see how people’s behavior changed on a much more [inaudible 00:16:54] level.

Julia Bardmesser:
So actually that’s helping also to get people to understand the underlying data and enterprise data capabilities are very important to get the analytics, to get the business reports, as they were.

Andy Palmer:
This is a really interesting topic. Can you talk a little bit about this idea of data lineage and provenance and I know you’ve done a lot of that in your career, and financial services is like the cutting edge of this. I think that benefit is hearing for other people in other industries to hear how you’ve been dealing with and managing data lineage and provenance in your projects.

Julia Bardmesser:
Data lineage is a very challenging subject that I find. I have worked in and around data lineage all the time back from Citi, and if you have a legacy companies, and majority of the companies are legacy, or they have a large footprint going back many years, they usually have “spaghetti”, quote, or spaghetti diagram. When you go to data flows, that’s the best example that you can give. It’s a spaghetti diagram. And if you will not do the lineage by hand, it’s pretty much impossible. Of course, by the time you finish documenting all of your flows, the changes, they’re not true anymore. The changes have been made. Now you’re going to need to start all over again, and it’s not an amount of investment and money that need to be spent to keep keeping it up to date and [inaudible 00:18:31].

Julia Bardmesser:
There is some tooling out there that tells you that it can discover lineage through code [inaudible 00:18:39] or collection scrolling and we have tried to do that at Deutsche Bank, and it works to certain degree. It does some limitations on how much it can do and how complicated it is. There is also some process changes you can make then when you keep track of your changes using your usual [SDLC 00:18:57]. What I have found the most effective way of dealing with the lineage is to build from metadata driven architecture. So the environment they are building here in Voya is actually self-documented, so of course they have the limitation because you can’t really build material jobs if you don’t have the lineage documentation, this [inaudible 00:19:21] goes here and this is [crosstalk 00:19:23].

Julia Bardmesser:
But as we develop and as there’ve been more and more, just in [inaudible 00:19:28], it doesn’t mean how much discipline they have. The documentation’s going to uphold [inaudible 00:19:33]. [crosstalk 00:19:34]

Andy Palmer:
So make sure I got this right, so what you’re describing to me feels something like swagger for data. Is that a fair-

Julia Bardmesser:
What do you mean by swagger?

Andy Palmer:
Swagger is automatically generated documentation for APIs, as you build APIs, it sort of automatically publishes [crosstalk 00:19:50].

Julia Bardmesser:
Actually, that’s an interesting thought, and I will take it back to my team and we’ll talk about that and see if we can do that. But actually what I talk about is that everything you build generates metadata that gets saved. So to build a product that brings raw data into data lake. It generates metadata, but it brings that data in and it saves it in the metadata depository. So any transmission, every single transmission, you see how many [inaudible 00:20:25] went, what were the [inaudible 00:20:27] on the source system, what were the [inaudible 00:20:29] when it landed, almost all the ETL tools have that.

Andy Palmer:
Gotcha.

Julia Bardmesser:
[crosstalk 00:20:33] tool tell you that.

Andy Palmer:
Yeah, right.

Julia Bardmesser:
So in our tool we’re distributing the data, we’re exactly the same. It generates metadata as it works.

Andy Palmer:
Gotcha.

Julia Bardmesser:
Not as a documentation at the beginning, but you need to. Not part documentation. But generating as you move data, generating metadata with your move, to me, that’s the only way to effectively [inaudible 00:20:58].

Andy Palmer:
So if the continuous generation of the metadata and all the changes that are reflected in that repository, then you can go back and prosecute that as a tool to understand lineage. Yeah.

Julia Bardmesser:
Exactly, exactly.

Andy Palmer:
Wow. Have you done that effectively? It sounds like you’ve been practicing this. Have you been able to make it work?

Julia Bardmesser:
Well, that’s one of the design principles for the enterprise data platform that they’re building at Voya. So to build a platform, we have made it work. Of course, not only is it a huge environment before us and then after us, so then you go back to the combination of things, so everything that you build is new you build based on that principle. And then you do combination of documentation and SDLC changes and that’s where [inaudible 00:21:49] enterprise architecture and architectural review and the process for controlling the changes and what information projects have to put in when they read the change, that way it’s help. So you have a combination of things.

Andy Palmer:
So it sounds like your mandate is much bigger than data. Again, it crosses over into product, but also into the overall technical architecture supporting the entire company.

Julia Bardmesser:
Yeah, enterprise architecture is outside. It’s not only data. It’s overall enterprise architecture for Voya, so including architecture governance, including our technology and application landscape, and one of the more interesting efforts that I’ve been part of that’s been allowed by enterprise architecture team is our public log journey. So I’m learning a lot from all parts of the company, from our security team and our infrastructure team and my enterprise architecture team is how do you take the company, Voya’s footprint to the cloud? They’re not there yet, so it’s still work in progress.

Andy Palmer:
So tell me more about that. There’s a lot of people that are curious about how they’re going to get to the cloud and is now the right time? What stage are you at? You say you’re taking your first steps. Do you have stuff running natively in the cloud yet? And are you running experiments or do you have stuff in production?

Julia Bardmesser:
Well, yes to all of this. We have some pieces that run on the cloud. We have POCs, we have from data perspective done quite a bit on the cloud, or we’re planning to. It’s still not in production. So we have a warehouse multi-cloud [inaudible 00:23:34] applications as well majority of the company. I mentioned sales force. That’s a source. [inaudible 00:23:41], that’s a source. There’s just tons of apps out there people don’t even realize they are on the cloud. So on the public cloud, I think now is an excellent time to go to public cloud. It’s not bleeding edge anymore. So it’s not you have security on the cloud, you have incredibly with people, all of the public clouds, well, three major public clouds, invest incredible amount of money in the security. Of course, obviously any security breach would be front page news and it’s huge impact on their business.

Julia Bardmesser:
So from the security perspective, why do you of course got to pay attention to the security and they work very closely with the [inaudible 00:24:21] and our security team at Voya. It’s been very well-developed and it’s quite mature from the public cloud perspective. It offers a lot of new capabilities to take business forward. And as companies get to the cloud, that’s an edge that you have. Eventually right now, it’s an edge. In a couple of years, it will be table stakes.

Andy Palmer:
Wow. So a couple of years [crosstalk 00:24:54] so it’s moving fast.

Julia Bardmesser:
Yeah, so think about what I’ve heard recently from one of the consultants in that space was that to think about transition from on prem to the cloud the same way we thought about mainframe to client server.

Andy Palmer:
Gotcha, gotcha. So it’s-

Julia Bardmesser:
So it’s not for cost, per se. There is a core structure there as well that potentially could be very favorable in the cloud, but its new capabilities, its ability to bring more to the business from technology perspective to support business growth. And then eventually it becomes you can hire people who know how to do stuff on prem. Everybody wants to work on the cloud. How many global programmers can you hire currently?

Andy Palmer:
Right. [crosstalk 00:25:36]

Julia Bardmesser:
I know my father, who’s a global programmer, still is getting calls. Yeah.

Andy Palmer:
Yeah, I think about all those people that are used to racking and stacking servers inside of these data centers and 10 years from now it’ll be less important or they’ll all work for Amazon.

Julia Bardmesser:
Yeah.

Andy Palmer:
Of the existing cloud vendors from the data practitioner’s perspective, of the existing cloud vendors, which of the vendors do you like the most or are there strengths and weaknesses between GCP, AWS, and Azure? Things that you prefer [crosstalk 00:26:14]

Julia Bardmesser:
There is such a tight competition with the three of them, that if any one of them is ahead on any one capability at any moment, the other one or the other two will catch up pretty quickly and then step ahead of one. So I feel that it’s less of an issue of which cloud is good for each specific purpose. It’s more overall what makes sense for you as a company. And I also generally suggest not to spend a lot of time figuring out which cloud, because whatever this creation is right now, whatever the capabilities is right now, the other cloud is going to catch up soon. So it’s more about where you have relationship, where do you have service, what’s coming out in terms of the course perspective. What’s your current stack? A lot of other criteria that will get you which cloud.

Julia Bardmesser:
But I think spending a lot of time figuring out which one’s ahead, by the time you finish figuring it out, the other clouds will catch up.

Andy Palmer:
So it’s kind of great for us as customers to be able to consume that. They’re just going to compete and drive the cost down and you’re saying there’s really going to be parody in terms of the features over time.

Julia Bardmesser:
Yeah, I’m sure, and I don’t want to endorse any one vendor, so I’m not going to speak about what I’m seeing specifically for each of the vendors, but again, whatever it is at the moment, I’m speculate now, let’s say, and I’m completely making it up, AWS has made great strides in analytical [inaudible 00:27:45]. And right now, that’s really important and that’s my criteria. By the time I finish making my criteria and I’ve signed the contract and to build the lengthen zone, Azure is going to be right there, if not slightly ahead, right?

Andy Palmer:
Right.

Julia Bardmesser:
And then Amazon is going to catch up and build something over there and then Azure will develop… So it’s really, I don’t think it’s a capability question. I think it’s more what makes sense from the relationship and tangible, what kind of stack you have what makes it easier, where do you have more other vendors that you use, and then if your SaaS apps that you have [crosstalk 00:28:22] are they running them on any one of those public clouds?

Andy Palmer:
Gotcha. Gotcha. So and tell me how you think about multi-cloud. Is that something that’s worth factoring in at the beginning of the journey to the public cloud? Or do you have to build it in from day one, or can you think about having multiple cloud providers over time?

Julia Bardmesser:
Well, the reality of it, that we are on multi-cloud. Not just Voya. Pretty much everybody who has any cloud footprint is on multi-cloud because again, we have SaaS apps. Even the companies that are not a tool, thinking about public cloud, just don’t want to go there, I’m pretty sure they probably have either worked their sales force or something like that.

Andy Palmer:
Right.

Julia Bardmesser:
So they’re on the cloud and they’re on prem, so that’s already… Or you have private cloud, which is… Right? So the reality that majority of the companies are already on multi-cloud. Now, if you want to be evenly on multi-cloud, meaning that you can switch from one cloud to another cloud seamlessly as a [DR 00:29:26] in case something happens, that’s an investment. And companies need to think about it. It’s a significant investment to build out those infrastructures, because they’re not the same. If you want to take advantage of native capabilities on whatever cloud you’re on, then it’s not an easy transition to a different cloud. If you want to take advantage of some platform as a service on top of the cloud, that’s an investment. That’s more expensive.

Julia Bardmesser:
You’re paying for the cloud and you’re paying for pass. So that will allow you presumably to switch. We haven’t tried that, so I’ve actually never tried that, so I don’t know, but that’s a selling point of pass, that you can take something from one place to another place.

Andy Palmer:
Yeah.

Julia Bardmesser:
But again, that’s more expensive. Or you have to develop in a way that’s neutral of underlying cloud capabilities, well, then you lose on other [inaudible 00:30:14] capabilities. So as always, the reason one way or the reason the magic [inaudible 00:30:21] that you, and just go, “Oh, I can run on any cloud,” it’s an investment the companies need to make and they need to understand the outcome or the value from that type of investment is worth the amount of money that that’s going to take on the ongoing basis.

Andy Palmer:
Well, and what about some of the traditional, high end, on prem workloads? Like I think of the kind of workloads that I used to run on Teradata and such. Do you see those kinds of workloads moving over onto public cloud or-

Julia Bardmesser:
[crosstalk 00:30:53]

Andy Palmer:
… like the Snowflake guys are really popular right now and Databricks on the [inaudible 00:30:57]. Do you see these Teradata-like workloads or Exadata-like workloads moving over any time soon?

Julia Bardmesser:
I think so. I don’t know in terms of how quickly that’s going to happen, because again, we haven’t run a lot of performance tests on the cloud. Like Snowflake is absolutely one of the things that the feature of Snowflake is spinning up as many computes against the storage. So that gets you the performance. Well, again, how cost effective, what is this exactly that you’re running? Again, and also what I would want to think about is I think [inaudible 00:31:39] is not the right way [inaudible 00:31:41] if you just [inaudible 00:31:44] or infrastructure or whatever data footprint or application on it, if you just move it from your data center to the cloud, you kind of meet some opportunity.

Andy Palmer:
Mm-hmm (affirmative).

Julia Bardmesser:
So there is a lot of re-engineering and refactoring of the application’s state or architecture to take full advantage of the cloud.

Andy Palmer:
Gotcha. So in addition to the cloud, what are the other big changes that you see happening in data management in the next three to five years? Are there patterns that you’re seeing and/or tech that you like or things that you think are going to be important and useful?

Julia Bardmesser:
The tech that I would like to see more of, and we’re at the beginning of this. But as I said, I think with the catalogs, we still don’t know. It’s very cool technology. I still don’t know if they have an effective way of using it to really bring the value. But the tech I would really like to see… So the hardest part in building data environment is not bringing data in. It’s integrating. It’s integrating data and finding common keys.

Andy Palmer:
Common keys, yeah, yeah.

Julia Bardmesser:
[crosstalk 00:32:59] Yes. If you have one dataset, you can run all kinds of data analytics on that and it’s fine. But if you have two datasets that you want to relate to each other, you need to find your common key. And they’re not going to be standardized the same way. They’re going to have different code values. They’re going to have different identifiers. And that’s where a lot of work from my team or from anybody really, all the analytics team, I think are spending a lot of time figuring out common keys. So a lot of data prep business and Alteryx, EXata, others, is all about finding data common keys without having to do ETL and doing it programmatically.

Andy Palmer:
Yeah, yeah, yeah.

Julia Bardmesser:
And doing it on the fly, and of course, doing it on the fly, you have problems because you do it one way and then somebody else comes against the same dataset and does it differently and you get [inaudible 00:33:47].

Andy Palmer:
Yeah.

Julia Bardmesser:
… different numbers of the same set of data. So where I would really like to see more growth and more development and the way I think the machine learning really comes in or artificial intelligence, different things, but [inaudible 00:34:00] to find common keys easier. So the data integration, it means the most difficult problem out of the culture problem that I spoke about, technical problem that you have. Now you can actually, and that’s what I tried in Citi, tried to solve for data integration problem culturally, meaning institute common keys and enforce, have them standardize common keys, standardize valid values, and do it all the way from the data ingestions all the way through. That’s a huge culture change. I think it’s going to actually be easier to do it with the tools and [crosstalk 00:34:37].

Andy Palmer:
Yeah, yeah, yeah.

Julia Bardmesser:
Especially in the large companies. Now maybe in the smaller companies, especially with the new ones, they can start that way from the beginning. That might be better. However, I’ve been told by people who are in those types of companies that they’re in such a hurry that that discipline doesn’t exist to begin with because you have to spend more time doing it. And they don’t have any time because they’re racing to get the next level of [inaudible 00:35:01].

Andy Palmer:
Yeah. And God forbid you go out and you buy another company, and then you’ve got [crosstalk 00:35:06]-

Julia Bardmesser:
Yeah, then you have the same problem all over again. So I think that might help it a lot more, and that’s where I would like to see development is the data integration common keys problem.

Andy Palmer:
The common keys problem. It’s near and dear to our hearts at Tamr. We’re trying to get people to use the machine aggressively with some human expertise. Have you found yourself in a battle… One of the battles that I had was people competing to be the common key, right? That you had one group that had a key that they really liked their key and they wanted to see that key adopted as the primary across the… Back to these human problems.

Julia Bardmesser:
Yeah.

Andy Palmer:
Have you experienced that?

Julia Bardmesser:
Oh, quite a bit, yes. And see, so that was, for the first two years, that was my job in Citi.

Andy Palmer:
Really? Really?

Julia Bardmesser:
So finding, identifying… So that was trying to address the data integration common keys problem top down. So to find what the common key is and then get everybody to adopt it all the way back to the sources. So the first problem was exactly what you just spoke about. Obviously everybody has their own common keys and they don’t want to change and they want everybody else to use theirs.

Andy Palmer:
Right. My key’s the best. Why doesn’t everybody just use it? Yeah.

Julia Bardmesser:
Yes, exactly. So golden source for client data, golden source for [crosstalk 00:36:28] data, golden source for country data, golden source… So the way we try to get around it, so of course, there is always somebody who yells the loudest or has the most support from the management. That’s the deal, that’s always a consideration. We try to put a little bit more process and structure around it, so something more objective criteria of what would be authoritative source of data. So anywhere from what’s your data coverage? So if you have an excellent key for 10% of securities that the company trades and to get another 90% is going to take you five years, you’re probably not the most effective common key.

Andy Palmer:
Yeah, yeah, yeah.

Julia Bardmesser:
Right? Or if you have a common key that’s not easily extensible or you’re not using in these three identifiers or whatever, so there is that set. And then there is a set of how does it change management? If somebody wants an addition to common keys, usually those are depositories that have other information about whatever the entity that you have a common key for. So if you have organizational entity, you have things like industry codes, for example. So you have a coverage of that type of data, and if there is a consumer of this data that needs an additional piece of information and you don’t have it, how long it’s going to take you to get it? What’s the lie out of your door? Is it going to take six to 12 months? Because if it’s six to 12 months, you’re probably not a good candidate for being enterprise common key.

Julia Bardmesser:
So that’s what they have tried to do, is to come to an agreement what would constitute an authoritative source and it goes from technical to change management criteria and then have that agreement in and then use that as a criteria to say, “Out of that many people who have something like common keys, but then we narrow it down.” So did it always work? No. But it helped. It helped.

Andy Palmer:
Well, it seems like you’ve done this in practice and based on our experience at Tamr, we just see a lot of companies that are just beginning the journey of even considering doing this kind of work. It’s amazing to hear the stories about doing this in practice at large scale. It seems very, very challenging. What about integrating third party data? How do you think about that at Voya and how have you seen that done in the past? Obviously completely different set of issues involved when you have lots of third party data coming in versus data you’re generating internally. What are the practices around third party data that you think are most effective?

Julia Bardmesser:
Well, have a lot of leverage with your data supplier helps a great deal. The best way to integrate third party data is when you can provide… There is either the common format or the layout in this accepted definition data, digitally and so on. Or you can say, “Here is how we want the data from you.” The first job that I talked about when I thought of managing data, that was actually reference data, pool level data coming from multiple external suppliers. So that’s why [inaudible 00:39:46] managing data, enterprise data in one is loading external data in one mortgaged security, single clause and multi-clause. I don’t know how many people in this audience will remember Intex dataset. I think it’s probably still there. It was an interesting dataset [crosstalk 00:40:05].

Julia Bardmesser:
So I think one of the easiest or one of the ways that really helped to deal with external data is to have industry standards. So at least some would get past the problem of common keys. If your common keys is bad to that, because you have data diction that is usually external data. You have [inaudible 00:40:29] data dictionaries that go with that. But again, the work of if you want to work this dataset with the other one, you go back to the same problem, to well define different dictionaries [inaudible 00:40:39]. That’s not in here, it’s I think on your side.

Andy Palmer:
I think it is on my side. Sorry.

Julia Bardmesser:
Are you in Boston?

Andy Palmer:
Yes, Cambridge, yes. Sorry.

Julia Bardmesser:
Oh, oh, good. I’m in New York. No, that’s okay. I’m just like, “No, I don’t think it’s here.” So well-defined data dictionaries, well-defined data [inaudible 00:41:03]. Accountability on the part of vendor to define and provide data quality information and as accountability to fix data quality on the data that has not come through. How successful you are with that, it depends. But from my experience, that pretty much the only thing that helps. It’s following the industry standards and having a well-defined data dictionary and making sure the vendor accepts accountability for data quality and fixing data quality in a reasonable timeframe.

Andy Palmer:
So the data standards in financial services, you have lots of experience with this, do you feel like they’re standards that have worked well and are there reasons you think that they’ve worked well, or ones that are particularly bad and why they may have failed?

Julia Bardmesser:
I think definition of the standards, and it’s actually the same external industry standard or internal, for example, for Citi. Definition of the standard is difficult work where you have a lot of workers come together and agree on the standard. However, the hard part is adoption of the standard. The LEI standard, for example, and it’s been a while since I’ve been in the middle of LEI work and I think you have women on your list that actually much more in-depth on that, so I’m very much an outsider on that. But what I have seen, standards itself is pretty good. It’s adoption of the standard by wide variety of companies. And it is a requirement, I think there is a requirement that to trade, you need to have an LEI identifier, but a lot of small companies don’t need to.

Julia Bardmesser:
And then if you’re trying to source this data and have LEI identifier as your common key, you have a pretty large footprint that does not have that. So that’s the problem. I think it’s not so much creation of the standard. It’s adoption of that standard by broad enough community, so it’s actually valuable, and it doesn’t become yet another key that you have to build a [crosstalk 00:43:16] against.

Andy Palmer:
Yeah, at some level, like these standards can make the problem worse, not better, right?

Julia Bardmesser:
Yeah. I think on the security side, like [inaudible 00:43:27], those are pretty good standards.

Andy Palmer:
Yeah, yeah.

Julia Bardmesser:
They’re pretty prevalent and they have a reasonably good coverage. Of course, [inaudible 00:43:36] would do better if they didn’t reuse the [inaudible 00:43:38], but that’s my personal pet peeve.

Andy Palmer:
Right. Yeah. Every one of these things has, there’s like physics to it, right? There’s strengths and weaknesses associated with them and how they’re designed.

Julia Bardmesser:
Yeah.

Andy Palmer:
So a little bit more on the personal side as you think about your career, this is kind of the age of data is happening and you’re at the center of it. How do you think about the next three to five to 10 years of your career? And the kinds of things you’re most interested in?

Julia Bardmesser:
I’m really enjoying right now that I’m not only doing data. I’m also responsible for enterprise architecture and sales force. And I think my background in data especially in terms of digital transformation, so if you think around digital transformation, the way I define digital transformation is removing of friction in operations of the company, all the way from communication with the customers all the way back to financial [inaudible 00:44:43]. So it’s end to end. It’s not only UI, it’s your experience with the customer, but it’s operational experience. It’s financial [inaudible 00:44:51] experience. It’s analytics experience. So it’s removing of the friction.

Julia Bardmesser:
And removing friction that we usually have is manual work to enter data, to remove data, to move things around, and that’s where I think my background of understanding how the data moves and what do we need to do to make it frictionless comes into play. So that’s something that I’m looking forward to to be able to do it on a broader scale. Similar to what the plans we have around the sales force projects that we have again removing a lot of manual intervention, to normalize how we deal with data, to provide a much broader report and using sales force data, external data, [inaudible 00:45:35] data, other internal data from Voya to improve our business outcomes, taking it on a broader scale.

Julia Bardmesser:
So I think it’s a lot of what I would like to do in the next three to five years, and 10 years I’m hoping I’ll be retired as [inaudible 00:45:48].

Andy Palmer:
Okay, yeah.

Julia Bardmesser:
That’s what I would like to get to, is just to take a broader, but not only building the data platform, but looking at data end to end.

Andy Palmer:
Well, it seems like the time has come that people with the data backgrounds are providing, especially like yourself, with a focus on consumption, providing the context for how and why we should organize our enterprise architecture to generate managed data as an asset. It’s really very unique position you’re in, and this has really been an amazing conversation. You’re an inspirational leader and we really appreciate you taking the time to talk today. Thanks, Julia.

Julia Bardmesser:
Thank you, Andy. It’s been wonderful.