Data Science For Non Data Scientists (Guest: Brandon Cosley)

Andreas Welsch: 0:21

Today we'll talk about data science for non-data scientists. And who better to talk to about it than someone who's passionate about doing just that? Brandon Cosley. Hey Brandon. How's it going?

Brandon Cosley: 0:35

Hey, it's going well. Thanks for having me on the show. I really appreciate it, Andreas.

Andreas Welsch: 0:40

Great. Hey, why don't you tell us a little bit about yourself, who you are, what do you do?

Brandon Cosley: 0:45

Sure, absolutely. I direct a data science organization. We are responsible for all of the customization that happens for the business artificial intelligence, machine learning, data science our buzzwords. Everybody's selling them. But what I think the real challenge is trying to figure out how to use those capabilities in the fabric of your own business. And so I'm responsible for an organization that tries to do just that, bring those capabilities into the things that are very unique and very special to the industry, and to the sort of the way in which we work in that industry. Because not every company is going to sell you something that's going to fit perfectly with what you do, your infrastructure, all the old things that you have going on your legacy systems. So I help bring those capabilities to those systems.

Andreas Welsch: 1:40

Awesome. And I know following your content for a while, you're also quite passionate about teaching data science to non-data scientists outside of work. So I'm really excited that you're able to join me. And so for those of you just joining the stream, please drop a comment in the chat. What business stakeholders should learn about more around AI? What do you think? But hey to kick things off, Brandon, should we play a little game?

Brandon Cosley: 2:08

Okay, let's do it.

Andreas Welsch: 2:10

Okay, perfect. So this game is called In Your Own Words. When I hit the buzzer, the wheels will start spinning and when they stop, you'll see a sentence. And I'd like you to answer with the first thing that comes to mind and why, in your own words. So Brendan, are you ready for, What's the BUZZ?

Brandon Cosley: 2:36

Let's do this.

Andreas Welsch: 2:37

Good. Alright, so then let's get started. If AI were a rock band, who would it be?

Brandon Cosley: 2:45

That is a really fun question for me to think about. And I think if AI were rock band, the most obvious choice probably would be Daft Punk. I'm a little bit of a techno guy, but if AI were a rock band, it's definitely Daft Punk.

Andreas Welsch: 3:03

Alright. And why do you think it's them?

Brandon Cosley: 3:05

One, they dress like robots. It's their shtick. Two, they write great music. It's all electronic. So, when I think of AI, that's what conjures up for me They're dance they're oriented towards digital and and they even wear suits that make them look like artificially intelligent robots. But they're not, they're people.

Andreas Welsch: 3:29

Awesome. So one more time, huh?

Brandon Cosley: 3:33

Absolutely. Cool. That's right. Yeah. Makes sense.

Andreas Welsch: 3:36

Perfect. Thanks for answering that on the spot in your own words. Hey I remember you had mentioned a while ago that you have a dual role and you alluded to that earlier in the intro as well. On one hand, leading a team of data scientists and on the other, enabling your business stakeholders on AI, what it is and what you can use it for. Now I'm really curious, how do you do that kind of upskilling and what kind of prerequisites do you see that these individuals on business teams should have to make this a successful endeavor?

Brandon Cosley: 4:11

Yeah, so I think that there's a couple of different angles. I think that from the one aspect, you have data scientists who are coming into new environments. So data scientists coming out of boot camps and degree organizations coming into businesses trying to figure out. What does it mean to take what I've learned about training models and turning that into actual data science products? I think that there is a lot to be learned, not just from those who understand how to train models, but I take that same framework and try and teach that to business stakeholders. Let business stakeholders understand that there is a framework just as there is for building. That there's a framework for building data science products. And if we can think about that framework together, we can identify those use cases that help us advance the business forward by building upon that framework, and then allow those subject matter experts who understand the modeling part to come in and add those pieces. But to me, the most important thing is building understanding around the proper framework for building data science products. Fabric of the business understands those use cases. Make sense when they make sense, and then how to actually go about implement implementing them because they know who to talk to. You don't need to know how to do each part of the framework, right? What you do need to know how to do is who to talk to, the right people when each different piece of that framework needs to come into play. As you're building those. Hey, enablement. It's all about teaching people what framework is for data science as a solution.

Andreas Welsch: 5:57

That's awesome. I would like to pick up on one thing that I noticed because I feel in a lot of cases we think about AI as projects, but I hear you talk about it as a product. How do you see that being different? And why do you feel it's important to call it a product as opposed to a project?

Brandon Cosley: 6:16

So for me, user experiences are fundamental. And if you can't have the proper user experience, then at the end of the day a project will fail. And so for me it makes a lot more sense to call artificial intelligence agents a product, right? Because what it's doing is it's enabling a user. And so users work with products, users work in the context of projects, but at the end of the day, what they want is something that they can take away from that project so that they get more value from it day in and day out. And to me, that's why a product is a much more fitting term than the word project. Sure, it takes a project to build a product, but at the end of the day, what you want is a product that lasts for a very long time that continues to drive value for that team, for that business, for that process. So I really prefer the term product. I think that what we're building as data scientists are truly products. And to me, that orients us towards that user experience, which is fundamental for the way in which this, those products can be successful.

Andreas Welsch: 7:23

I think that's a perfect way to summarize and to frame it and to call it that distinction. I really like how you think about it and how you seem to approach it. Really putting the user at the center. Yeah. So thanks for sharing that. Now, like you said there should be a framework. There are data scientists, there are business stakeholders, and they will eventually work together on building an AI product. How should business stakeholders work with data scientists? What should they be prepared to answer? And on the other hand, what should business analyst, data scientists be prepared to answer, to help understand the problem much better and much more quickly?

Brandon Cosley: 8:06

Great question. I think probably the most fundamental challenge as a data scientist is really trying to understand how to turn a business problem, or what I often call real problems into data science problems. Fundamentally, there are different degrees of sophistication that we can turn business problems into data science problems. Those data science problems are the ways in which we reframe business problems in the context of the capabilities that we have available to us, whether they be tools, technologies, models, you name it, right? All the things that fit under that umbrella of data science. But the better we a, the better able we. To take all of those real problems and reframe them as. Data science problems, the better able we are to identify whether or not there are data science solutions for those problems. Now, will the problem fundamentally change or the solution fundamentally change? Absolutely. And that's where communication becomes fundamental to that entire process. So there needs to be a communication between the business stakeholder and the data science team that work. To say, Hey, here's how we translate the pain point that you have, using the capabilities that we know to how to potentially solve that pain point. And so trying to tie those two things together is where the magic happens, right? That's the magic in the middle. And that I think is the most challenging part to what I do on a day-to-day basis, is trying to bring my business stakeholders. Where they're dealing with real everyday problems into my world to say, Hey, there's a data science solution for that, but we need to reframe it in different ways. And that means we need to reframe the way that you think about your real problem. And so we look at the capabilities of data science. To try and do, it doesn't always fit. And when it doesn't fit, then we have to understand that and be ready to move on. And more importantly, we have to be ready to experiment because oftentimes we don't know if it fits. So when we look for that magic in the middle, we have to have that clear communication. We take those real problems, we turn them into data science problems, and then we find all the different possible. That may fit in the middle, and that's where the value is. Whenever we fail fast, we identify the solution more quickly and hopefully, Deploy that solution project so that it becomes, here it is again, product

Andreas Welsch: 10:45

Perfect. I've seen in my own experience that a lot of times you start with something like, Hey, let's let's either see where or how we can apply AI to this problem. Or here's an idea that will help somebody. Reduce the number of clicks or get better insights, and then you need to start drilling down, right? Why are you doing this? What does it mean? What's the next step? What does the business impact look like? What are you seeing there? Do you see that be part of that conversation as, as well in your work.

Brandon Cosley: 11:22

Yeah, absolutely. So I think to revisit what I just said in, in the previous answer, real problems have KPIs tied to them, right? They have key performance indicators, businesses about metrics. Obviously it drive our bottom line, but where data scientists really have chows trying to understand how. Work drives and affects those KPIs. And so what's really important is, again, to try and understand how we turn all these that we've been trained to do and bring those in to impact those KPIs. That's where the business can understand you. That's where you can have an impact, but that's also fundamentally the hardest challenge lies because it's not always the case that when we translate that real problem into a business pro or into. Data science problem, excuse me, that we have an impact on that kpi. So we need to make sure that fundamentally the things that we're doing with our data science are having an impact on that KPI, so that way the business understands the value of what we do build.

Andreas Welsch: 12:34

Great summary. And again, tying it back to business KPIs and something that your business stakeholders care about and are it's eventually measured on. So keeping an eye on the chat. I see Lisa saying, Hey, keep discussion simple and using analogies. It's helpful when you work with business stakeholders. And quick shout out to Oliver in Germany who said, Hey I've built a course in, in English and German called Data Science for Business Leaders for small medium enterprises. So check that out. And Sujata is saying, Hey we have to be willing to experiment like you. She emphasizes with that so true that it's not always a fit. And that also needs to be the expectation that it is experimentation in some cases more research than, a straightforward process in that sense. Now to close it out with our last question. I know you've seen your fair share of AI projects, obviously, and I think one perspective that is usually not covered that much is what does it actually look like from a data science manager's perspective? How can your business stakeholders and your business analysts make your life as a data science manager easier? What do you see in your work?

Brandon Cosley: 14:00

Fundamentally, for me, it comes down to communication. I'm maybe sometimes described as an over communicator. I always check in. I really like to have the conversations. And I think to Sujata's point, it's really important to understand where the language crosses different boundaries and different domains. You have to take a lot of time, a lot of attention, and you have to spend a lot of real care to try and understand where the language is using different terms, but it means the same thing and that's fundamentally where I think we struggle as data science managers to help communicate the value of the work that our teams do with our business stakeholders. And so the more that we can get in front of them, the more that we can have those conversations, the better able they are to articul. The relationship between their real problems and the ways in which my teams and my developers can solve for them. So I think at the end of the day, for me, if I were to wrap it up in a single sentence, it would be communicate, spend lots of time talking. And I know it seems so fundamentally simple. But it is so important that we spend time communicating with those individuals. It is always people in the loop. If we don't keep them in the loop. Then whatever we're doing is destined to fail. So it's always keeping them in the loop. It's always making sure that we communicate, and then being willing and also being empathetic. Understand that we are talking about the same things, just using different words. And so trying to find those relationships is where we really drive value.

Andreas Welsch: 15:44

I think that's golden that, that advice really keeping people in the loop and working as one team to have a shared understanding. There's not much more I can add. Maybe one last question from the chat before we wrap it up altogether. Arek is asking, Hey Brandon when do you conduct projects, are you mostly looking for specific domain expertise or more AI strategy?

Brandon Cosley: 16:08

Fundamental question, both actually. It's important I always train both data scientists and non-data scientists to understand that the more knowledge that you have of how data science algorithms work to train their models, right? The. The more options will be available to you for potential solutions, right? For turning real problems into potential data science solutions. So it's important that you have AI strategists that understand the framework through which you build products. But I am most interested in domain expertise because at the end of the day, they're the ones who are struggling with the real problems that we're trying to. And so what we need to be able to do is we need to be able to connect the two together. So my answer to art's question is that you need both. Fundamentally, the sophistication that you have on the AI side is important because it opens you up to new possibilities. But at the end of the day, it's really fundamental that your domain expert helps to articulate what their real problem. And how those potential solutions may actually impact the metrics that they're trying to change.

Andreas Welsch: 17:36

That's a great summary. So having both domain expertise and technical expertise to, to make the biggest impact possible.

Brandon Cosley: 17:44

You need apples and dough to have apple pie.

Andreas Welsch: 17:47

Perfect. Now you're making me hungry. Hey let's wrap it up. And maybe if I can toss it over to you to quickly summarize the key points for each of the topics that we talked about.

Brandon Cosley: 18:00

So for me, for data scientists and non data scientists understanding the framework through which we build solutions which I have called products understanding the full product lifecycle of a data scientist is fundamental and communication. Constant communication with those domain experts who are dealing with the real problems that data scientists need to be able to turn into data science problems so that way we can look for experimentations to find proper solutions that actually fit. And have impacts on the KPIs that those business domain stakeholders are really struggling with. So communication, understanding the framework, and then enabling that communication so that you find that magic in the middle and understand that it will take experimentation. And then oftentimes that experimentation does take. But the more we communicate, the faster we're able to fail. The quicker we able, the quicker we are able to get to a solution that develops a product that has a lasting impact for a given organization.

Andreas Welsch: 19:07

Awesome. Thank you so much for wrapping it up. And folks in the audience, we're coming up to the end of our show today. Thank you so much for joining us Brandon, and for sharing your expertise with us. It's been great having you on. I really appreciate it.

Brandon Cosley: 19:20

Thanks, Andreas.

Andreas Welsch: 19:21

Alright, thanks Brandon.