Champions
of Data + AI

Data leaders powering data-driven innovation

EPISODE 19

Data for the People

In this episode, we explore how Jack’s technical and product background is helping him lead and execute on the ADP data and AI strategy to bring more awareness to pay equity and build new services and products beyond payment processing. He will also share his views on creating a data strategy centered around defensive use cases first — in other words, how to protect your data and your brand before tackling the offense-based ones — the ones that make you more money and reduce cost.

headshot
Jack Berkowitz
Chief Data Officer, ADP
Jack Berkowitz joined ADP in August 2018 as the Senior Vice President of Product Development for DataCloud, ADP’s People Analytics and Compensation benchmarking solution. As the leader for the DataCloud team, Jack is responsible for ADP’s vision and approach to artificial intelligence and the development of cloud-native machine learning solutions that span ADP’s HCM product suites. With data across ADP’s population of nearly 30 million employee records in the U.S., ADP has a unique position in the market to deliver unmatched insights to clients.

Jack came to ADP from Oracle, where he was Vice President, Products and Data Science for Oracle’s Adaptive Intelligence program. In this role, he oversaw the market, technical and sales strategy for Oracle’s suite of next-generation intelligent applications, combining web data, data science and platform cloud computing. Previously, he oversaw product strategy for analytics in Oracle’s cloud applications, and product management and strategy for Oracle’s analytics portfolio.

Prior to Oracle, Jack spent 20 years in both product development and implementation of intelligent information systems, most recently involved in web-scale search and recommendation systems, data-driven applications and the Semantic Web. He has been on the executive team of four startups involved in search, reasoning or meta-data-driven applications (Attivio, Siderean, Cerebra and Reef Software) and he co-founded edapta, which enabled dynamic user interfaces and personalization for mobile and web clients. Jack has delivered solutions for a wide range of Global 50 clients, spanning financial services, consumer web, and healthcare. Early in his career, he was involved with DARPA- and FAA-sponsored programs for user experience and intelligent systems, FAA Aviation Security, and the certification of the B777 flight deck.

Jack has a master’s degree in industrial engineering and operations research from Virginia Tech and a bachelor’s degree in psychology from the College of William and Mary.

Read Interview

Chris D’Agostino:
Welcome to the Champions of Data and AI. I’m your host, Chris D’Agostino. In this episode, I’m joined by Jack Berkowitz, chief data officer at ADP. Jack and I explore how his technical and product background is helping him lead and execute on the ADP data and AI strategy to bring more awareness on pay equity, and build new services and products beyond payment processing. He will also share his views on creating a data strategy centered around defensive use cases first. In other words, how to protect your data and your brand before tackling the offense based ones, the ones that make you more money and reduce cost. Let’s get started.

Chris D’Agostino:
All right, Jack. Thanks for being on Champions of Data and AI. Hope you’re doing well today.

Jack Berkowitz:
Hey, I’m doing great. Thanks for having me, Chris.

Chris D’Agostino:
Awesome. So, let’s dive right in. So, couple things want to have the audience get to know you a bit. And I know that you have a background in software programming, and want to talk a little bit about when you built your first production application.

Jack Berkowitz:
So, the very first production application was a long time ago. 1983. We were trying to correct some accounting system in a print shop. So, we had to write a little bit of print shop code. And so, a long time ago, Apple IIe, little floppys going back and forth.

Chris D’Agostino:
This is post-college, pre-college?

Jack Berkowitz:
I was in high school. Junior in high school.

Chris D’Agostino:
Awesome. So, all right, let’s talk about data. You collect a lot of data. You represent thousands of companies and millions of employees. I was struck by the pay equity data analysis that your company had done. And you presented a bit of it at the Reinvent Conference. For those viewers that didn’t see Jack’s presentation, I would just do a YouTube search on it, and it’s great. Can you walk us through, though, quickly, the highlight of that? And let’s chat a little bit more about some of the insights that you found.

Jack Berkowitz:
Yeah, so, we collect all this information and we help people understand it. One of the big things that we do is we build a people analytics product. So, we have clients that will use the people analytics to track their head count, but also do things like understand the compensation footprint for their business.

Jack Berkowitz:
One of the things that we’ve built over the past year is a view for them to look at pay equity gaps. So, pay equity gaps are what most people identify. A woman makes less than a man for the same type of job, or the same work. But it also extends to people of color. It can extend to people with disabilities. It could extend to veterans as well. So, there are pay equity gaps throughout that. And what we did is we took a look at pay equity gaps. We rolled out this new capability and we saw what companies were doing with it.

Jack Berkowitz:
To step back, unfortunately, over the past couple of years, what we see in the data overall across the country is that pay equity gaps, whereas they looked like they were in proving over the past couple of years, actually because of so many people that have lost jobs, the pay equity gaps are actually getting worse. Worse for women overall. But there’s an upside to that. By using the types of tools, by putting the data at the fingertips of clients, both their data, as well as combined with all this external information, people can take action. And, in just a few months, we saw over 1000 companies take concrete action, return 720, and it’s probably even more than that now, $720 million into their communities. We equate it about $3500 per person to make adjustments, to close that gap. And so, the hope is that people keep using data for good in this way.

Chris D’Agostino:
Yeah, so, one of the interesting insights, if I’m interpreting correctly, is that prior to the pandemic and the change in the workforce, if you looked at the average salary for people in these underrepresented groups that you mentioned, women, vets, disabled people, people that lost their jobs during the pandemic tended to be lower wage earners, and were disproportionately among those underrepresented groups. So, during the pandemic, the average pay looked like it might have gone up, but when you factor those lost jobs back in, and the lower wages associated with them, that pay gap actually was worse.

Jack Berkowitz:
That’s right. It got worse for everybody. And it’s because of certain industries being affected. So, you and I work in the software industry and it’s boom time. Nothing’s ever been faster than in the cloud industry the past couple of years. But if you look at accommodation, or if you look at transportation, if you look at retail, even if you look at manufacturing here in the states. Just a massive impact. Massive amounts of people are losing work.

Jack Berkowitz:
Or people dropping out of the workforce because of childcare or because of needing to take care of adult parents, people choosing to drop out of the workforce. And it turns out that’s created really tough times for a lot of people.

Chris D’Agostino:
Yeah. Yeah. Well, let’s talk a little bit about that talk, because you presented some interesting statistics about ADP and the amount of money that you process in your platform to deliver a paycheck to an individual, so that they can feed their family is and pay the rent or pay the mortgage. And you plotted. You said if ADP and the money it was processing were the GDP of a country, it would sit somewhere between France and Italy, which when I saw immediately, reminded me of Italy winning the 2006 World Cup, and Italy just, of course, won the 2020 Euro tournament, played just in 2021.

Chris D’Agostino:
And so, we’re now coming up on the World Cup, which Italy will, of course win. So, I found it fascinating that you’re processing that much money on behalf of individual consumers, employees.

Jack Berkowitz:
Yeah. It was an analogy that I’ve needed to make, to understand the impact of what we’re doing, and the criticality. If we were to have either issues with processing, or if we don’t take our software incredibly seriously, it’s not like missing some advertising. It’s not like, “Oh, gee, I gave somebody the wrong ad.” If I mess with somebody’s paycheck, well, it’s a substantial impact on people. And a substantial impact, not just on one individual, but on the economy overall.

Jack Berkowitz:
And so, I’m always struggling for analogies, how to make data approachable, and I’m glad that was memorable. I know there’s debates. My friends in France, and ADP has a big business in France, as well as Italy. This has been impactful to them right now. There’s a lot of arguments about who has better food as well. So, I’m willing to go referee.

Chris D’Agostino:
Sounds good. Well, I was just in France just a few days back. Theirs is quite good, but I’m biased towards the Italians, of course. Speaking of bias, let’s talk about bias and data. Let’s talk about the importance of the data that you work with. Not only in terms of just keeping people’s livelihoods going, but also, you mentioned ADP as an HR platform and the ability to do recruiting, ostensibly, a client of yours could post a job through ADP, advertise for a particular position, maybe federate that out through different job sites, and then collect resumes. And you can run machine learning models against the applicants to figure out who might be best qualified.

Chris D’Agostino:
So, with all this data that you’ve got on people and the importance that ADP plays in the role of employment and livelihood, talk to me a little bit about ethics and ensuring your data sets are of good quality. How do you approach solving for that? How do you measure if you are using the data that has no bias baked into it? Or maybe an example, if you’ve determined that there’s some bias, what do you do to remediate it? Give us some insights there.

Jack Berkowitz:
Now, it’s a great question. It’s something we think about every single day. So, as we started to really ramp up in our machine learning and data journey, this is going back more than a couple years ago, we stood back and we said, “We really need to get some principles outlined.” And so, we stepped back, we formed a group, both internal ADP people, as well as some external experts on the team. And we stood up a data and AI ethics board.

Jack Berkowitz:
And we set forth, and it’s published on our website, the principles that we’re going to use to guide ourselves. And one of those is that bias exists. We don’t want to make the statement that things are not biased. We know bias exists. But therefore, we have a responsibility to monitor, understand and advise when these situations are there.

Jack Berkowitz:
And so, fast forward now two years, we’ve implemented a pretty robust ML operations playbook, as well as system, that allows us to understand what the data is, the data shape, data shape changes over time. Things about data drift, looking at scores and understanding those, and actually running very specific tests that are applicable to the HR industry in the ML ops system. So, we know if there’s a situation going on, and we can immediately correct it.

Jack Berkowitz:
Now, we have people who’s got eyeballs on this constantly, as well as being monitored. Is this just most software is monitored. Why do you monitor for uptime and not monitor for bias? Of course, we’re going to monitor for both. And then, so we take actions. We have policies and procedures in place for that, as well.

Chris D’Agostino:
So, let’s talk a little bit about the technology stack. If you look at some of the public presentations you’ve given, one could maybe say you tend to be pretty open with the best of breed, considered best of breed, the right tool for the right job kind of approach. So, one is that characterization fair? Number two, how do you think about open formats, open standards, transparency in the platforms that you use? Tell me a little bit about what goes into your thinking when you’re selecting a particular component of your architecture.

Jack Berkowitz:
Yeah, it’s a great question. So, I believe really strongly in openness I spent the mid part of my career working in the semantic web technologies, which were all around the sharing of self-describing data and self-describing knowledge formats. I believe strongly in things like SQL. Why? Because there’s 5 million people that know how to write SQL, and it’s important for you to be able to interact in a common language.

Jack Berkowitz:
And I believe that, by keeping things in that way, with the sharing of things, we can build architectures and components that allow us to build things that we could never imagine, over time. Now, the thing about selecting best breed technology, there too. At a certain point, there will be platforms that can encapsulate all the pieces of tech that I need, or that we need. But at the same time, if we stick to openness, then we can plug and play those pieces that we need over time, as well.

Jack Berkowitz:
And so, yeah, today there are eight components that we might need to stitch together to perform a function. Maybe tomorrow I can get them from just one component, one vendor or one open source thing. That would be fantastic. As long as I’m building it in an open way so that the data can be moved, so that the queries can be readdressed, then I have a chance to do that. If I go lock in to just a core platform and I can never get out of it, then I’m constrained by the development pace of the platform that I’ve chosen.

Jack Berkowitz:
And I don’t know what’s going to come the next two years. I don’t know what the demands are from our company and more importantly from our clients. And so, I need to have a system that’s flexible enough to allow us to meet those demands. And that was part of what we talked about with AWS. That’s part of this entire cloud move. That’s why we work with folks like Databricks, is to be able to build platforms that allow us to be flexible, to allow us to build new capabilities that we hadn’t anticipated even four or five months ago. And that’s what we’re doing.

Chris D’Agostino:
Yeah. So, maybe for me to summarize or play this back to you, it sounds like, if you standardize on data as a product using standards and formats that are portable, you can mix and match the tooling to meet the need of the use case, or to meet the demands of the business. And if you can find something from an architectural standpoint that maybe simplifies your architecture a bit, reduces the number of moving parts that you’ve got to worry about keeping up and running, without sacrificing that flexibility and portability of the data, and resisting some kind of lock into a platform or to a proprietary data format or something, that’s how you guide through [crosstalk 00:14:41].

Jack Berkowitz:
Yeah, that’s right. That’s right. That’s exactly right. Over time, things come together, things break apart. We just need that flexibility. At least for my career, it’s been important for this interoperability to happen over the past 30 years. And I see no difference today, not to continue down that path. And it has served ADP really well. It’s served our clients really well over the past couple of years.

Chris D’Agostino:
So, talk to us a little bit about out the ADP data cloud. What is it? What’s the objective? I know data sharing is an element of it. And help us understand what your goals are there.

Jack Berkowitz:
Yeah. So, at ADP, we want to make use of the data for our clients’ benefit, for the country’s benefit. One of the things that we publish is a thing called the National Employment Report, which, every month our research institute puts out. It’s a view at the information, all anonymized, all aggregated, so it’s not personally identifiable, but a view of the information to inform what’s happening in the US, as well as in Canada, as well as some other places, from an economic view.

Jack Berkowitz:
So, to do that, we have to bring together all that information. We have to align all that information. How do we take 20 million job titles and align it down to 9,000, so people can look up compensation benchmarks? So, the data cloud’s really about putting all that together. And then on top of that, we build products or capabilities that allow people to take advantage of it. So, people in analytics, which is a set of products that we sell, or data that we license, or that we provide for people doing computation analysis for their companies. Or maybe they’re doing demand planning, so supply chain demand planning, based on the economic shifts inside of the country.

Jack Berkowitz:
And so, we build those types of things, and then we also build some things that are just to service our clients better. So, how can we use that data to create shorter wait times for clients, or to have the system proactively repair our HR system? So, some of what we’re doing with that information, it’s actually providing better HR or payroll services to our clients.

Chris D’Agostino:
Oh, that’s great. Well, I would imagine, with that much data across that many industries, and as many countries as you operate in, you’re seeing labor trends probably earlier than almost anyone else.

Jack Berkowitz:
Yeah. Yeah, we do. Right? So, we update that information for the public monthly, and that allows you to see, for example, we just published a few weeks ago. We know how much job switchers they’re making in improved salary. It was about 6.6%. And we can see that literally within a few days, at the end of the month. We can watch the economic recoveries. We can watch those swings. And that information, companies, like I was talking about demand planning, companies can make adjustments to what products they’re stocking, where their warehouses are, even their driver schedules or their delivery schedules, based on understanding that information.

Jack Berkowitz:
Are people commuting? And if they are commuting, have there been changes in the acceptable commute over the past two years? And it turns out it has. It’s been cut about half. So, people are willing to commute still, but not as far as they used to.

Chris D’Agostino:
Perhaps they’re not required to as much, with hybrid working models and things like that.

Jack Berkowitz:
Yeah. There’s hybrid working models, but again, only about 30% of the population’s allowed to do that. So, you and I are working from home, but there’s still a bunch of people that aren’t.

Chris D’Agostino:
Yeah.

Jack Berkowitz:
Right?

Chris D’Agostino:
No, I mean, that’s interesting. Yeah, sometimes, in our industry, we take for granted because our lives are oftentimes so very different, especially from people in the services industry, where they’re tethered to the place the service is offered.

Jack Berkowitz:
That’s right. It’s great me and you can be digital nomads, and you and I can order GrubHub from a mountaintop in I don’t know where, right, Colorado? But somebody’s got to actually drive that up to us. And if you read the tech papers, you think everybody was the ones being digital nomads, but there are other people and they’re important. And so, we need to make sure that both sets of groups, or both groups of people are treated fairly.

Chris D’Agostino:
Yeah. I’m wondering if there’s any insight. This is off the cuff question, about where people are moving, given that people that have the ability to now be more flexible with where they work from home, they’re opting to leave expensive priced areas to maybe areas less expensive.

Jack Berkowitz:
Well, it’s funny. So, there’s two things that I know are true right now. It’s true that everybody’s moving either to Austin or Miami. Okay. That’s true. But it’s also true that the people in Austin and Miami that were living there are actually getting displaced. So, lower income people, actually moving to the suburbs or to other towns that they can afford. So, you actually see that ripple effect, where people are leaving California, leaving New York, but then you see that ripple effect on the economy. Which opens up new opportunities. Chattanooga or Asheville or Nashville. These are all areas where there’s going to be new vibrance in terms of people moving, people wanting to work.

Chris D’Agostino:
Yeah. That’s great. Well, you’re talking to somebody, San Francisco and I’m here now in Miami.

Jack Berkowitz:
There you go.

Chris D’Agostino:
You hit the nail on the head. All right. So the role of the CDO. Let’s switch gears and focus on that for a moment. As I said earlier, I speak to a lot of executives in the data space. Very few of them span both skill sets on the policy side, as well as the technology side. And oftentimes, the ones that are on the policy side, what I hear from them either publicly or privately is the challenge around not having enough teeth in the role on the implementation side.

Chris D’Agostino:
So, it’s this very real struggle between, “Okay, they’re setting the policy agenda for the organization, but now they’re having to partner or defer to the IT organization to implement things. And the way IT organizations are oftentimes structured, it’s less about a data platform, data ecosystem environment, and how you bring together data from across the enterprise. It’s more about one off applications and their backing data store and the security model and those kinds of things. So tell us a little bit about the CDO role that you have, and its ability to influence both the policy and the implementation details.

Jack Berkowitz:
That’s a great question. So, because it came up as a product development person, and my first few years here were really about product development, when we defined the CDO role, we defined it a little bit different. I’m the first CDO at ADP. And so, we defined it maybe a little bit different than other companies, but it’s proven to work for us. So, we really look at it simply. We have defense and we have offense.

Jack Berkowitz:
Now, you notice I put defense first. Normally people say offense first and then defense. But in today’s environment, with everything going on in terms of threats and situations, we find that defense for the data, and defense is not just protection of the data, but also our ability to recover. But also things like data governance. Where is the data? Is it being handled appropriately? What’s master reference data? What third party sources? What are the models? Are the models actually compromised? All of that is an important part of the CDO. So, we can set policy, but we can also set in practices, work with our CIO, work with our chief security officer, bring that together.

Jack Berkowitz:
Now, at the same time, I retain my product development role. And so, we build products. We build products for internal. We build products that we monetize. And monetization’s an important part of the modern CDO. I think a few years ago, the CDO for companies were like, “Well, I’m just an internal thing. I’ll set policy.” But today, companies want to see a return in one way or the other. So, the ability for us to do that, either directly or through enterprise value calculations, is an important part. So, both of those go hand in hand, and that’s allowing me to partner with our CIO really closely and our chief security officer really, really closely, to bring together a view across data that’s just refreshing. Right. I joke. I built these tools for years, now I have to use them. And so, it’s pretty funny, right?

Chris D’Agostino:
No, it’s interesting that you talk about offensive and defensive. I have an ebook that I authored on behalf of the company that goes through a 10 step strategy. And one of the things that it talks about are use case selection, and the use case is being bent in a couple of different ways. In the first, most granular way, or course grained way, sorry, is in offensive versus defensive, and starting with the defensive ones.

Chris D’Agostino:
The ones that are going to protect your brand, the ones that are going to protect your data from a data breach. Those are the ones that you’ve got to get to first.

Jack Berkowitz:
That’s right.

Chris D’Agostino:
Ensure that you’ve secured the data. I was at Capital One during the data breach. I have some inside knowledge on how the breach took place, what caused it, what the remediation was, and then what we needed to do differently as a company and how we had to align resources around ensuring the broader security of the data within.

Jack Berkowitz:
It’s no good to talk about billions of dollars of benefit if it’s built on a creaky house of cards, or whatever you want to call it. Right? So, we really are spending a lot of energy. The data platform, people immediately say, “Oh, well, that’s so that you can get business advantage.” Yeah. But the data platform is really about security and defense. Lineage and everything else.

Chris D’Agostino:
Yeah. So, then, you take that, you take the offense and the defensive ones. You tackle the defensive ones first. The offensive ones, in terms of increasing revenue, reducing costs, those kinds of things are great. And then I tend to bend them into three performance categories, subsecond SLA response time. So, very fast, high speed event-based type decisioning. The second is multi second. So, where the customer can afford to wait a few seconds for something to execute. And then multi minute, some of these longer run, big analytic processing model training exercises. Things like that.

Chris D’Agostino:
And so, really thinking about it as which ones do you tackle first. The defensive ones. How do you figure out what the architecture needs to be to support a given use case? And then the final piece, with use case selection, is really around data adjacency. So, easier to tackle use case A and B, that have an 80% overlap in the data that they need to execute, versus use case C, which has zero over lap, and it’s a completely orthogonal data set that you need to go after and figure out.

Jack Berkowitz:
One of the things that we’re really focused on right now is, for that data adjacency, is which data domains do we onboard into this full protection lineage governance model. And as we start to bring on which data domains are fully managed, then do we get these adjacencies? It’s hard because you start to think, “Well, I got to bring on my clients’ data first,” or I got to bring on this or that. And it’s got, as anything, becomes a spider web or you get this octopus effect. But eventually, all that information needs to come together.

Jack Berkowitz:
An area that we’re working on and we don’t want it to sound like object models from the 2000s, but we are working on semantics. And because you’re starting to see now, in the industry, “Oh, I need a metric store.” Well, for those of us who’s been around a while, you’re going to need more than a metric store. You’re going to need dimension representation. You’re going to need object representation. And all of this type of sharing needs to come together.

Chris D’Agostino:
Jack, it’s been great talking to you. If you can just tell us, what advice would you give to an aspiring CDO, someone, like you said, there are people that have helped shape the direction you’ve gone in your career. What things would you recommend to people that really have a passion for data, want to elevate their executive presence and maybe get to a role like what you have, dealing with the volumes of data you’re you’re dealing with?

Jack Berkowitz:
Yeah. I think there’s really two things. First of all, do what somebody did for me, give somebody a chance. Give somebody a chance. You don’t need to be the smartest person in the room. We talked about being the dumbest person in the room, but really it’s about building those teams. Those teams of talent are really the thing that happens. I always like to say, “Ah, I just took credit for what these guys did.” And it’s exactly the intent.

Jack Berkowitz:
Because giving people a chance to rise and shine actually bolsters the entire company as you go. So, that’s the first thing. The second thing is really start to listen. A lot of people like to talk. We’re talking right now, and a lot of people like to hear, but they actually don’t like to listen. And so, actively listening to what your company needs or what your client needs, or what the individual engineer needs at 2:00 AM. Well, she’s fixing the latest security issue that you need fixed the next day, for the CEO. These are important traits.

Jack Berkowitz:
Notice I didn’t talk about technology. I didn’t talk about going to some conference someplace. Didn’t talk about posting something on LinkedIn. It’s really about building a great team and being part of a great team, and really listening to what people need. And I think these are the two most important things. This is what my mentors told me. Sometimes in a very brutal way like, “Listen, damn it.” But it’s important, and it’s really important to share that with others.