JD Williams @ Zoetis | 6-steps for POC to production & long-term value | Data Science Hangout

So, welcome to the Data Science Hangout. A little different today in that I'm hosting, not Rachel. Rachel is on a train right now, right now down to the Philadelphia area. But this is the Data Science Hangout. We host it every Thursday at noon, same time, same place.

There are a few ways to participate in the Hangout. It's meant to be super casual, open conversation, questions about data science. You can participate just by typing into the chat. We obviously encourage that. Raise your hand and hop in live and ask a question. And then there's also a way to submit questions anonymously via a Slido link that my colleague Hannah will provide. So, do whatever you feel comfortable with.

But I'm super pumped to introduce our featured leader today, J.D. Williams from Zoetis. And I'll hand it over to you, J.D., if you want to introduce yourself.

Great. Thank you. Welcome, everyone. Glad to see you all here. I was scheduled to be here, I think it was three weeks ago, four weeks ago, I can't remember what it was. And I got COVID. And you would not want me to have been here on the call at that time. It's actually my third infection of COVID in the past three years. I do not recommend any more than one or any infection of COVID. But glad to be recovered. A little bit of a groggy voice, kind of still hanging out after this time.

JD's background and journey

So, I have a slide that I've used to introduce myself and that other team members on my team have used it to introduce themselves as well. I'll pull it up as a way to get started here. So, just a little bit about me and my background. I grew up in the Rocky Mountain West. So, this is the area of Colorado, Idaho, Oregon, Utah. So, I love the mountains. And so, I live now in flat old Indianapolis, Indiana.

So, coming from that background, I went to Brigham Young University in Provo, Utah for my undergraduate degree, which was in mathematics education. My original plan was to be a high school math teacher. And then, through the course of some events, I ended up getting into statistics. The master's program there at BYU in the statistics program and absolutely loved it. And then, took that opportunity to then go on to do a PhD in statistics and that was at Virginia Tech in Blacksburg, Virginia. The stats program at Virginia Tech is the third oldest statistics program in the country. And the area of my dissertation research was in the area of multivariable statistical process control, which comes actually full circle.

After graduating with my PhD, started to work at GE General Electric in their research headquarters. This was in upstate New York and really loved it there. GE, at the time, owned a little bit of everything. So, we would, you know, lean loans and leases of all kinds of equipment. So, that's where I got into the area of financial services. And that led to a role with JPMorgan Chase and their analytics headquarters. And that was actually in Columbus, Ohio. And that's where I really learned the concept of big data.

And by the way, this is like 2010, it's kind of timeframe, big data and just really, you know, using vast amounts of data to make decision making. And the banking, the financial industry, I feel like has been kind of well ahead in the analytics curve. You know, for example, at JPMorgan Chase, I never heard of anything in terms of an analytics project. It was just the way things were. It was part of the DNA of the company.

Whereas when I took my next role at Dow AgroSciences, I was asked to come and start up a brand new business analytics team. So, I went and hired a small team. And really, we were just starting from scratch and trying to get analytics use cases going. And everything was in terms of project and use cases and just trying to get momentum and get buy-in from various stakeholders. Whereas where I'd come from a place where there was no major decision made unless the analytics team signed off on it. And it was such an integral part of how decisions were made.

And then two years ago, this is during the pandemic, so February of 2021, I had the opportunity to come here at Zoetis with just an amazing opportunity to basically create and lead a brand new business insights and analytics team. And so, basically, I was given like a blank sheet of paper to draw the organizational design, write the job descriptions, and go and hire the team to accomplish the vision of leading in digital and data analytics, which is one of the company priorities that's been publicly stated.

Intro to Zoetis team

So, I've had to do this twice now. This is the second time. In this case, Zoetis is, just a little bit of background on the company, Zoetis is the largest animal health company in the world. So it's an $8 billion company with 13,000 or so employees globally. So if any of you have dogs or cats, then very likely there's a chance that your dogs or cats are being prescribed Zoetis medications from your veterinarian, or any of you are associated with livestock farming. So cattle, beef or dairy cattle, pigs, chickens, sheep, and even salmon. So fish farming as well.

So in 2020, we had a new CEO, Kristen Peck, that took the helm. And one of the first changes she does was to bring in a new chief information officer who is Wafa Mamili. And I had known and worked with Wafa at Eli Lilly previously, and she is just an absolutely spectacular leader. So Wafa came in and completely took a new direction into the company to be technology and digital focused. And that's where the priority of leading in digital and data analytics came about. And that was really the foundation of why my team was created and funding it. So not just words. We don't say, hey, by words we're supporting you, but actually by dollars and resources.

Building a data science team from scratch

And so I was brought in along with others to build this data, this business insights and analytics team. And so one of my first responsibilities was to understand what the business needs are and to understand how we can play a role in that. And when I came in, I realized that there were some data scientists and pockets of data science throughout the company, mainly in our U.S. commercial, which is our major commercial unit, as well as in our R&D group, there were data science statisticians, et cetera, in those areas.

And so what we did was design the organization to be one where we had dedicated focus of data analysts. So those that are not necessarily have master's and PhDs in data science, but have bachelor's or so forth degrees in data analytics type of degree areas that we can focus in particular areas. And then I designed a pool of data scientists that could then be deployed across any of the use cases and could be very flexible kind of a pool there.

And so we focused a lot of our initial work in the area of manufacturing and supply chain as well as our international commercial business, where there was very little or none in terms of data science support. And so they were immediately, there's so much pent up demand in these spaces that it didn't take long for us to generate momentum in those areas. Then we also did things like with finance and HR kind of other enabling functions, but we focused mainly on our manufacturing supply chain and our international commercial.

And as we gain momentum, we've expanded now to more use cases within the U.S. commercial as well as in our R&D space. So in terms of designing the organization, I basically just designed to have an analytical lead. So an analytic lead that I see as a person who's a translator between what are the needs of the business and what are the analytic solutions that can help to solve those needs. So I have analytics leads for each of those areas, and then data analysts that report to them, and this pool of data scientists that work across all of those verticals to serve the needs across each of those areas.

Being flexible on location when hiring

One of the key factors for us was being able to be flexible in terms of location. I don't know how many of you are considered to be remote employees, but I was hired in during the pandemic and I was told, look, you're in Indianapolis, but we're a New Jersey based headquarters. We want you to be in New Jersey if you want to, but you don't have to. So when I hired the team, it was always a remote option if they wanted to stay remote. But that was really the key to be able to recruit people from all over the globe, actually. So not just in the US, but leave that flexibility for remote work.

And just as a side note, I was able to discover that there are about 10 or so Zoetis colleagues that live here in the Indianapolis area, and we get together once a month for lunch. And so that kind of gets that face-to-face interaction you need to take care of. So I'd recommend that to anyone too. If you're working remote and there's colleagues that live in your area, have a little chat that you have, put that together and see if you can get together on a periodic basis just to interact and meet new people.

Just kind of rough sizing, we're about 25, so not like super huge, but a very powerful and mighty team.

Promoting data science in the company

So one of the things that I wanted to talk through is how we have helped to promote data science in the company. And so a lot of you are probably leaders or you're working in your companies trying to get momentum in terms of making data science and the power of data science to be more prominent in the company. And so when I joined Zoetis, it was pretty early on, it's like within the first few months, really noticing that, okay, there's some data science groups that exist and we're bringing in data science team. We don't want there to be confusion around, okay, why are there so many teams? Who's doing what? We wanted to be more united, be really a unified, one Zoetis team of data scientists.

So what we decided to do was to create this thing called the Zoetis Data Science Forum, where we would meet once a month and share success stories with each other, share use cases, share code and best practices. And so it started off with roughly about 50 or so people that would attend across all of these different data science areas. And we met for several months, we'd be sharing things and it was great. And then the attendance started to build up a little bit more and more, and then more and more non-data scientists started to come in and want to be interested in what was going on.

And so it was at that point we realized, well, wait a minute, why don't we open up this to be much more accessible to anyone in the company, not just have really kind of technical kinds of topics, but talk more about the business value and what's the business impact and have the data science in there as well, but have it be in a very easy to understand manner and then have that be a way of promoting what the great data science work is across the company.

So as we started to do that, more and more people started to comment and enjoy it. And then our CEO, Kristen Peck, got wind of what we were doing and wanted to come and join one of the meetings. And this was in September of last year, so nearly a year ago. And with her joining, the meeting attendance bumped up by 50%.

And afterwards, she was just so thrilled with how the meeting went. She learned so much and wanted to continue to attend those meetings that she told her administrative assistant that I want you to put this meeting on my calendar every single month and I will try to come to as many as possible. And I can tell you that in the past, I think nine or so months, she's been there about six of the past nine times. Our CEO personally attending the Zoetis Data Science Forum, and she has become one of our greatest proponents. In fact, in our town hall, just two weeks ago, she talked about the Zoetis Data Science Forum in the global town hall, and how this is something that you should all go to and participate in. And by the way, over time, we've built up now where there's 300 participants.

Our CEO personally attending the Zoetis Data Science Forum, and she has become one of our greatest proponents.

Having the CEO and your CIO, all those senior leaders, be not just supportive, but just be strong advocates for what you do. There's really no replacement to that. It's incredible what you can do when you have that kind of senior level support. And we are so fortunate to have that at Zoetis.

Tips for building relationships with execs

So Zoetis is a big company, not sort of like a five-person startup. 13,000 employees worldwide. So getting the CEO on board is no small feat. How do you, what are some tips and recommendations you can give to people to form that kind of relationship?

I take no credit whatsoever. It's completely her. Kristen Peck is just an A-plus CEO. Honestly, so one of the things that we have is what's called digital fluency. We were trying to raise the digital, you know, understanding, advocacy, you know, usage of digital technologies in all its forms. And so we have this initiative called digital fluency. And our CEO is taking that completely serious to the fact where she said, okay, I want to learn how to code an app, code an app. And so our CEO, who's as busy as she is, you know, got some of our leaders who are, you know, coders and can help her to mentor her how to code an app. And she did it. She taught her kids how to do it. And so she learned how to code an app. And then when it comes to these data science topics, you know, she's personally told me that she considers herself a student trying to learn.

Example app showing community growth

So this is the app now. Some of you use Zoom and we're using Zoom now. I didn't realize this until just a few months ago that you can download the participant data of all those that attended your Zoom meetings. And so we downloaded, I think you can only get like a year's worth of history. I wish I had known this at the beginning. But anyway, so we downloaded all this data and Vincent in his whiz bang, our coding skills, was able to really quickly put this together and show what's the meeting progress over the past more than a year here. We've had over 800 unique attendees over the course of these many months. And you can see here, in the past couple months, we've had over 300 attendees.

So there's 309 on that one and then 331, the one in April, and our CEO attended both of those as well. But it also shows the attendees so we can track who has attended the most meetings, who has attended the most number of minutes. So you don't just log in for one or two minutes, but you stay for the whole meeting, et cetera.

Making an impact on manufacturing

I would say that where we use Posit the most is in the area of manufacturing. So when it comes to how you can make a real impact, whether it be reducing waste, whether it be finding defects, whether it be you're noticing trends in the manufacturing process that over time that are drifting in the wrong way so you can make corrective action, all that requires quick and easy visual access to data as well as the smart analytics to go on top of that to help you get alerts and to know when things are going the wrong way.

And so one of the things that Catalina and others, other really brilliant data scientists on the team have done is they've used Posit as the base platform to build a continuous process verification tool. Okay, and it actually has multiple facets to it. So there's different modules to help you go through the entire process of, okay, I have all this data that's been ingested. Now, how do I set control limits that make sense for this particular control metric of this product. So I'm measuring this, whether it be percent dissolution or some kind of quality metric that happens at the end of the manufacturing process and that product and that site.

And there's a module to then select the control chart limits. And then there's the more general module will be to have the continuous monitoring of those and another module that allows you to go in and actually make annotations. So you can type in, okay, this particular data point is, you know, out of spec. Here's the reason why. And so that gets tracked. Anyone can go back and see that of why it went out of control and as well as things like email alerting. So your process is, is drifting out of control and hit the hit control limit. Now you'll get an email alert right to your inbox. And so then you can quickly go and investigate.

This is one of the, the, the benefits of the way that Kathleen and team have designed the database and asked to be scalable. And so it's not just a bespoke solution for just one of our major sites or two, but now the way the data comes in and the way they formatted it, we're able to take the solution and basically say, you know, if your data is being loaded into SAP and so forth, or into our electronic batch record system, then you have it and be able to be exposed to you.

So for, you know, a couple dozen of our sites or so we're able to do this type of work. We're in the process now of validating it and having some alpha users use it. So we haven't fully deployed it yet. But we've made quite a bit of progress in that space of doing a continuous process verification.

So one of our major products that we manufacture is a monoclonal antibody. And, and these are four different products for dogs or cats. One is Cytopoint, which is a product that is used for basically, you know, bitching, you know, dogs. The other one is for Labrella and Salencia. Those are for osteoarthritis in dogs and cats. And these are monoclonal antibodies that need to be manufactured as part of those drug substances.

So, one of the things that, that Catalina and team have done is they have, they have built a Posit app that takes a lot of these, you know, real time, I'll call it real time, but very close to real time process data from the batches that are being, you know, actually, you know, scaled up and grown, you know, in our, in our bioreactors to be, you know, on our cloud platform, deposit and then present it back to the users as, you know, I'll call it near real time visuals of what's going on with the current batches as well as some smart analytics.

So, we have another data scientist involved in this that's from our team in Austria, who built some algorithms to help detect when these very complex multivariable, you know, relationships show an out of control signal and then send that signal to, you know, could be an email alert or just signal it on the actual chart itself.

And, and this has been extremely valuable for one, it's, it's really hard for these shop floor engineers to be able to get access to data because they might have to, you know, log into SAP, download a spreadsheet, or they may have to log into their MES system, download data to Excel, manually, manually, you know, manipulate it, which is so cumbersome. Whereas now we have it all flowing, you know, automatically up into the cloud platform, which then can be, you know, easily and readily, you know, deployed through tools like Posit to be able to build these visuals and the smart analytics on top of that, to be able to make better decisions.

So that's, that's in the area of manufacturing. That's, that's the area that we've actually just scratched the surface. So, the monoclonal antibody manufacturing process has an upstream and downstream process. We've only just tackled the upstream part of one of those three products. We want to do upstream and downstream across all three products and scale across multiple sites.

Building up a data team at a startup

Thanks for having me. I just wanted to go straight to the point that we were saying about this time. I'm based here in the UK, in London, and I'm a student presently working on a startup with my co-founder. And so, I was like asking myself, USP, and how you actually come in, setting up, let me say, a small practice. Yeah. And so, that's what I was like coming from because you work with big companies. So, I was like wondering in my institution, how do I go about that?

Yep. Got it. And I can see that the, thank you for that question. I did have a little bit of that experience when I did a kind of startup business analytics group earlier in my career at Dow AgriSciences. Basically, I came in and I was given the headcount to hire three people. So, there were four of us. So, a small and mighty team. I'm really dedicated to help make a difference using data and analytics in our commercial space. And I learned a lot from that. I mean, I learned a lot from mistakes and actual successes. So, let me tell you some of my mistakes, and then maybe that'll help guide you.

So, one of our first mistakes was to try to tackle too many things at once. But I think we just spent too much time trying to work on some inherited projects that we got instead of really trying to focus on where's the real biggest impact going to be. It took us about two years to get to the point where we're like, okay, we found an area that can really make a difference.

And then by the fourth year, that's when things started to really happen for us. And that's when we narrowed our focus, and we honed in, like put all of our efforts on just a couple things. We're a small team, we can only take on so much. Just focus on one or two things and just really do well at that.

And, you know, you're this thing, this cliche about quick wins. Well, it's a real thing. Having some, showing some progress along the way, you know, your senior leaders, they want to know that they're not just wasting their money by bringing on an expensive data science team that's going to just go off and do, you know, academic, you know, exercises. But are going to actually make a real difference.

And so, again, my advice would be narrow in, focus on the highest value areas with maybe two, one kind of shorter term, quick win, and one kind of longer term, according to the company priorities.

Six steps from POC to production and long-term value

Good, good. So we, we created up front a six step process where, where we, we kind of went through each this sort of stage gates to get to, you know, all the way to, you know, value, you know, long term value.

And the first step is identification slash prioritization. And so we had a, so for our manufacturing and supply chain leaders, we had a small steering committee of leaders and analytics leaders that came together and identified and prioritize the highest value use cases. And that was the first step. Once they identified those, then we did the second step, which is feasibility, where we took a little bit of time to gather the data to, to look at the, the data available. Do we have enough data and the right kind of data to do a proof of concept? And if so, then we went to the next stage, which was proof of concept.

You know, it's, it's, we're talking, the ideal would be between two and four weeks quick proof of concept using the data that we have sort of in a sandbox or offline. We're not building production ready code. We're not building something that's ready to go, to go live. We're just trying to prove the value. So that's really the idea of a proof of concept. Two things. Can we do it? And does it have the value? Those are the two questions we're trying to answer with a proof of concept.

And if the answer to both of those is yes, there's real business value and we were successfully able to do it, then we go to the fourth stage, which is the pilot. And that's where we would make it more production ready. Here's where we were going to, you know, partner with our data engineering team to take the analytic solution and put it into production on a small scale. And think of it as, you know, maybe one manufacturing site, one line, you know, kind of thing where you're not scaling it across, you know, many different places, but, but being more deliberate in how you do it.

And those pilots that are successful, for example, the, what I talked about the monoclonal antibody is very successful. We want to scale that to other, other sites, other products, other stages of the process. And, you know, leverage what we learned there and scale it across other areas. And so that's that scale up, you know, stage, which would be after the stage five is scale up after piloting. And the last stage is once we scaled it up as sustaining. So how do we sustain the value long-term?

The ideal would be between two and four weeks quick proof of concept using the data that we have sort of in a sandbox or offline. We're not building production ready code. We're not building something that's ready to go, to go live. We're just trying to prove the value.

And within the first year or so of us getting started, the total number of use cases in our, in our, in our use case tracker was about a hundred in various of those six stages. You know, some, some were just ideas, some were, you know, being, you know, prioritized somewhere in the feasibility step, some of the proof of concept, et cetera. So, but from there now we've taken the most high value of those, and we're actually focusing more on those and how we can pilot them and scale them.

Tech stack and tooling

Yeah, yeah. So we, we use, we use a lot of tools, I would say. We're trying to consolidate into one kind of workspace, but we're not, we're not dictating to every data scientist that you have to use, you know, this tool or have to use that, that language. We've got Python programmers, we've got R programmers that are very strong in both of those. Posit's nice because it allows you to do either one of those. We also use Databricks as a way to use, you know, Spark and Python code or R code as well, but mainly Python. And a lot of the Azure tools are what we use as well. So we're, we're an Azure shop.

So technically we were within our IT organization, but we rebranded. So IT kind of has this connotation of, you know, network and laptops and all that stuff. So we're called Zoetis Tech and Digital, Technology and Digital. And, and because WAFA came in and restructured all that, it's been just a very vastly different, you know, culture and landscape than it was prior to her coming on and making those changes. So the work style is, is very collaborative and very supportive in terms of our analytics initiatives.

So within Azure is our data Lake, you know, we've created data Lake within Azure and honestly, you know, how the data engineers ingest it. I think it can come in many different formats. We land it into a raw zone. Then from there it gets transformed, put into Parquet, into, into our, you know, governed zone. And from there it can be consumed among, you know, various downstream applications and consumers and data scientists, et cetera.

Data science vs. data analyst roles

So we have created just last year and launched what we have is the career framework, career progression framework. So when I first designed the org, the idea was that we wanted to hire people that can be really good at manipulating data that can create visuals, create dashboards that can communicate well with the business, but don't necessarily have an advanced degree in data science.

So, you know, analytics is a spectrum. There's this whole spectrum continuum between someone who, you know, on one end is, you know, competent and familiar with data analytics tools and capabilities, but not necessarily, you know, you don't have, you know, years of math background and all the math that's underneath it. Whereas someone who has an advanced degree in analytics, multiple years of grad school, you know, has, you know, all the mathematical foundations behind and really knows it very well. And so we felt like there was a lot of value in having, you know, both ends of that spectrum.

And so the design was to have some data analysts who have, you know, bachelor's degrees. But basically these are people that don't want to necessarily build models. They want to, they're interested in data engineering, they're interested in data visuals, they're interested in, you know, solving those problems kind of more of a descriptive way and aren't as interested in, in kind of machine learning or AI.

And those are our, our key, you know, data analysts and they're a smaller group. We have a much bigger group of data scientists and data scientists are those that have those advanced degrees in data science, statistics, et cetera. And, and then they, they really want to focus on building the models and refine those models, you know, MLOps, all those kinds of processes.

Now in, you know, late last year, we, all of ZTD or Zoologic Tech and Digital rolled out a career progression framework where we connected those together. And so data analysts that wanted to progress to be a data scientist would have that opportunity. It may mean that they have, you know, many more years of experience or maybe they get a master's degree along the way. And, and, you know, we're a company that supports our, our team members in, in, you know, furthering their education.

Staying ahead of trends and generative AI

So I get, I get some, I get, I get a couple of daily emails that come in from, I think one is a CEO, CIO journal, which has some technology headlines. So I try to keep up with some of the technology trends and companies what they're doing. So I'd say LinkedIn. Just try to stay plugged in whatever source you want to stay plugged into those kind of feeds it doesn't take long to read some highlights, and, you know, the title of the article was just brief summaries you can get through those pretty quickly.

I would also say that some of the most forward thinkers are really in the areas of like joint ventures or startup areas. These people that are thinking about new business opportunities. They're seeing things they're seeing trends and they're putting lots and lots of money to it, whenever you put your money where your mouth is that to me says okay there's something there that we should pay attention to. You know, and so some of these people saw chat GPT coming you know well in advance they knew that this this type of tsunami would come and they're already starting to create startups that would kind of ride the wave early on.

So, yeah, the whole generative AI space has been an absolute tsunami of interest that has come from Kristen Peck from waffle all the senior leaders to us as an analytics team to try to tackle and and we were able to onboard some of the generative AI tools using our Azure and Microsoft partnership. And we're starting to test those out. We're very early stages in that.

I attended a call last month with probably 50 or so CIOs from various companies and and we're not too far behind there's some companies that are farther ahead and their generative AI. You know, use cases and so forth. But what we're what we're doing is we're kicking the tires on the tool testing them out, and we have a couple of use case areas that we're going to start to dive into here shortly.

The public version, you know, just word of caution for all you out there never put company proprietary information or customer information into the public chat GPT, we've already seen some headlines of some bad actors that have done that and the consequences of that. We've onboarded these tools within our as well as firewall. So it's all you know privatized and any, you know, internal data that we use stays within the internal network.

Hiring and what JD looks for in candidates

Well, hey, I would say that connect with me on LinkedIn. I like to connect with as many people as possible. Stay connected that way constantly search the job board. And I would say that there are some jobs, not necessarily my team, but there are other data science teams within Zoetis that I know have been hiring. So definitely check the job listings within careers.zoetis.com

Um, so I think of it in terms of a couple different dimensions. One is, is the analytics skills dimension. And that's just kind of a, that's table stakes, you have to be able to talk the talk, walk the walk, you have to be able to code, you have to be able to do the stuff. And so, you know, that your technical expertise is absolute must have the next dimension, I would say, is more interpersonal side. So it's being able to communicate, being able to understand how to best communicate with others, how to work with others that don't have a technical background.

Um, and, and, uh, so there's a lot around the communication dimension. There's really a third dimension in my mind, which is more of business understanding business acumen. Now I hire people. I don't expect them to have any animal health background. I didn't have any animal health background. But we do anticipate that people will take the time to learn it and try to, uh, you know, expand and what are the company products? How do we make money? Who are our customers? You know, what are their needs and to really understand the business side of it. Cause that's, at the end of the day, that's what we're here for is to help our customers, help our company be successful. So three things, analytics dimension, I have to have, you know, kind of baseline level of competency, communication, soft skills, being able to understand and communicate well, and then having the business acumen to go with that.

Cause that's, at the end of the day, that's what we're here for is to help our customers, help our company be successful. So three things, analytics dimension, I have to have, you know, kind of baseline level of competency, communication, soft skills, being able to understand and communicate well, and then having the business acumen to go with that.

So my boss and I got interviewed by Forbes magazine, uh, just recently. And so if you look on May 2nd, you can see an article there on Zoetis. So if you just, um, if you just actually Google my name, JD Williams Zoetis, it'll show up on one of the things there, but it's, uh, it's on Forbes. Encourage you to go and read the article to talk about what we're doing in terms of analytics and data science and Zoetis. So thank you all. Appreciate the opportunity. Sorry. I wasn't here about a month ago. COVID's nasty. Don't recommend it for anybody. Stay away from COVID.

Awesome. Well, thank, thank you, JD. Thank you everyone who joined and participated. We'll be here next week. Uh, same time, same place. Have a great day, everyone.