Episode 1: Building Data Science Teams with Chris Matys

Media Thumbnail
  • 0.5
  • 1
  • 1.25
  • 1.5
  • 1.75
  • 2
This is a podcast episode titled, Episode 1: Building Data Science Teams with Chris Matys. The summary for this episode is: <p>We live in a world that’s increasingly fueled by data and analytics. But for companies to unlock the value all of that information represents, and improve business processes, they need some very specialized help. In this episode, Jon Prial talks to <a title="Georgian Partners" href= "https://www.georgianpartners.com/" target="_blank" rel= "noopener">Georgian Partners</a>’ own Chief Analytics Officer Chris Matys about data scientists. Find out what they do, where you can find them, and what it takes to build a successful data science team.</p> <p>You’ll hear about:</p> <ul> <li>Why companies are <a href= "https://georgianpartners.com/resources-data-science-team/">building their own data science teams</a> (1:56) -What data scientists actually do (3:18)</li> <li>The four skill sets necessary for data science (4:07)</li> <li>Why it’s important to start your data science team with a product manager (5:08)</li> <li>How to approach data to glean insights (6:46)</li> <li>How data science needs change as companies scale (9:15)</li> <li>Why companies struggle to find data scientists and build data science teams (11:10)</li> <li>How to find and vet data scientists (13:10)</li> <li>When it does and doesn’t make sense to outsource data science (16:01)</li> <li>What factors determine whether data science teams succeed or fail (22:07)</li> </ul>
Warning: This transcript was created using AI and will contain several inaccuracies.

Welcome to the impact podcast. I'm John croyle the matter with your reading mainstream checking the Galore business articles, but analysts free stuff free news access to social media free trials. You name it the information that's being collected to help them sell more to their consumers. They want to know what you looked at what you've checked on what websites you visit in and they want to figure out what products you're most likely to buy and this doesn't just happen to be to see businesses and B2B. It's no different yestoday. The software companies have huge amounts of data available to them to help them manage these understand this.

Proof specific business process seat I just can provide on today's episode. I'm talking with Chris mattes Georgian Partners Chief analytics officer. What do they do to find them? And what does it take to build a successful data science team?

Thanks for joining me today Chris. Thank you. John. Great to be here president analytics for years what's changed? The Legends teams in business intelligence teams have a variety of tools at their disposal to query and mine information from data warehouses gate of arts and so on the key is so the analysts are the ones that discover the rules that lead just might predict certain, or groups similar things together what's really changed is

We've gone from the information age until this almost Big Data age where the data simply become too big too complex and too fast for people to write the rules anymore. So in response machine learning is being rapidly adopted and machine learning is essentially letting algorithm actually analyze the data and come up with the rules and predict outcomes or distinguish screws for us and machine learning requires new skills that a business intelligence team doesn't have

At the core a data scientist is really selecting combining training and tuning algorithms to best learn insights from Big Data. But in reality that's really only represents 20% of their effort because to get there they actually spend the majority of their time wrangling data that is identifying cleaning an integrating the right data to power of this machine to have a shot with a team. Do you need one person to come to do this data wrangling Hothouse? How broad is this become?

You know, it's really required for primary overlapping skill-set. So first and foremost a scientist has to understand the mathematical underpinnings of Tabasco and to get to that point a deep understanding a business context is required. You've got to figure out what are the most impactful insights to go after and how they they can be used to effect change. Another still said that comes in as once I do have the answer. It's come out as output from statistical models and I have to be able to turn that into insights and communicate the back to the business in a meaningful way.

of course, the last part is

I also need if I'm talking about dealing with data. I need both Technical and Engineering skills to be able to experiment with the data and eventually productionize these insights the product manager job. How does the product manager and he's dated sign Works work together who's on point?

Yeah, so if you do well when we look at the data science scientist, and those four skills, that's an exceedingly rare skill set to get it almost any business function weather expert at the technical aspects there. They're really deep business understanding they have great communication and storytelling skills and they can actually do the work stands on very hard to find an any display of science is no different. So when you're looking at that the best approach is to build a team.

And when you look at the team, the very first question is getting what is the Insight? What is the highest impact insight and how will that benefit the business computer scientists mathematicians and so on may not be the best person to figure it out. That's why I always recommend that a product manager is actually the first person on the team figure out. What do we want to know? What's the question specifically as possible. Then you bring in the data scientists to start working on it. But do you need that that link to the business of somebody very skilled with talk to companies that say, okay. I've got the data. What can I do with it? Can you help us give an example where the Zeta signs team is really helping the evolution of a product strategy and kind of change that question.

Approach of here's the data bringing you do decide to something. What can I do with it? So rather If the product manager at the business pixelated bukkake insights and carefully defines that Target a data science just like a scientist and engineer, you know something you'll be better data scientist that I can go after an insight to what data is needed. What models will best uncover the free patterns in the data and so on just looking for Value Inn in data will lead to tons of research lots of experimentation. Very little deliverable the applied analytics Workshop how you see applied analytics as a differentiated for our portfolio in any River Gorge in talk a bit about that and

How that affects again this product strategy. This is this is really interesting that we could help our portfolio companies identifying those high-impact and insights and that usually the executive team and their Collective knowledge if you could facilitate that you can draw that information out. So we designed these workshops two-step people through looking at the audiences that they're currently serving with their business process software what might be if we can only answer this it would Wood Drive Great engagement to my drive even more value in into the company's.

So we really developed this this approach to go through and let's get the first question right when we have a few of those high-value inside swell Define, then we can move on from the the workshop and test them on customers and use that feedback now to say, okay. This is the one two or three key areas were going to focus data science on

I'm at Leeds usually to the very best out of science needs for a company change as his company's scale. You mentioned so much about the the growth of data the diverse types of data a little bit about the different types of companies that and then they might have an dead where they are kind of business is a valve companies early growth and

What you were when you look at those in the early stages of a startup identifying what information should be captured and how it should be captured and stored is of critical importance. So can we use later to to Garner insights Adidas Sciences can help that with their experience and knowing the types velocity and the the level of information that should be captured and you start getting to grow stage. You've amassed enough information. Now where the real Focus becomes. Can I get those key insights that will differentiate me from my competitors and provide the highest value to my customers. So you start focusing more in on the modeling aspects and using machine lat learning to answer very specific questions.

By the time you learned through all of that and you hit a mature stage, what you really want to do is reduce the friction for asking and answering the next question. That means architecting analytics framework and data some people call it a day the lake that makes it easier and easier to answer to question and answer more issues as terrific as he is so important now and why are companies in struggling in the building of data science teams, or do they struggle and one of the reasons is there looking for that perfect data scientist and the one that can that has deep mathematical skills knows as an engineer can communicate clearly with business people.

And understands the business process and the insights need to be garnered that is exceedingly rare to find him one person like most other disciplines the answer really lies in building steam with overlapping skills that not only reduces risk in having a team of folks on it but is is scalable in the future for data scientists. What should I be looking for as we balance across two sets of skills. Is it just balance are there are there drill down skills you looking for people to have a key qualities. They might be looking for what? What do you say but you don't know when you get a chance to add most companies today have a good product managers and good product managers can be that what is the Insight? We should be chasing group. They also tend to have good Engineers with their developing software.

Services in particular and they have people that are used to communicating the sales folks a customer service folks mad that are used to communicating out the missing skill. The one that is hardest to buy is really around the mathematics and statistics in particular. So what I always recommend is when you're looking across those the breadth of skills required the one you need the most of you probably have not built up is around the mathematics. I say start with that for data scientist.

As soon as you narrow it to that, they're easier to find lots of poaching of employees across different companies always be going out to find these skills. Then when I'm looking for a data scientist a priority around of the mathematics, you know, 90% Maurice recent recent recent study acting shows that 90% of data scientist have Masters or PhD level prudential's critical thinking the research approach and so on is is absolutely key when I would start looking. I mean that the programs for people can come out from my postage to get post-secondary education with

That would be closest a line would be Applied Mathematics or statistics and operations research. However, those are again also I've been rare and then people are buying for them. What I have found is some of the other research base theoretical areas also produce fantastic data scientist and by that I'm using physics astrophysics Biometrics. There's a whole range of more theoretical research that steals the speed the core skills that the data scientist will need. Excellent candidates when you got this. We met these people that one is

I asked questions that they do they know their craft but as with hiring software engineers in that pretty well figure out if they know their craft quickly, then I really focus in and say well especially if you're building a team rather than adding to a large existing team is you start looking for qualities of one of the first qualities I look for is arlia simplifier. Are they able to explain and simplify what they're doing rather than using lots of technical jargon and so on.

The second thing is are they researchers by Nature? They're curious. They're going after the problem. Also, you want to figure out especially in the early stages is if there actually a do her any roll up her sleeves and do the job because usually when you're starting off, you need to run a couple of Pilots to build success and then build the momentum behind the effort my company all that deliver inside as a service that they're sucking in different sources of data or your data and giving a few different slices of Industry verticals, perhaps. So how do you see that affecting this decision to do your work in-house are potentially outsourced these vendors?

Yeah, I I think if you have to approach that decision just like you would with any other cord function in your business if data science business intelligence being data-driven is cordier business and likely you've already got a business intelligence team. For example, then you're likely going to need a data science team internally. This will be ongoing you want to use it as a core competency to create competitive advantage and not just externally but also bring those skills and apply it against optimizing your own business and curly if you're in the area where it is more about is not as crucial then you can Outsource the skill sets and what you're really looking for there is I just need the answers.

I don't need to develop the capability within my organization to add to answer questions quickly or competitive Advantage. So that's the traditional Outsourcing model. So for example, a company may decide to Outsource sales Outsource marketing outsourced engineering if they're not poor like key Point here is a data science in a data-driven business is core and it should be thought of as court that me push more people to think about building a team internally than just Outsource skill sets and expertise.

Yeah, I do know is that a fair amount of experience?

Creating data science teams and creating them across a wide set of Industries and Company sizes and stages of building a team one is you may have to Pilot in other words management is thinking do I need to build a team. Do I need to invest in this? I need some proof in that case. What I I would either do is use third-party services to answer that burning question to show what can be done what can be accomplished or bring in a data science contractor and some Junior perhaps people pivoting from physics into data science and answer that question that really determine if the organization would like to build a capability giving you want to build a capability you stage it in what's most important we talked about earlier.

It's having that product manager. So the data science team is well-directed. Just like you want to make sure an engineering team is well-grounded second part is the date of scientists. And as I said focus on the map, that's the critical skill set. That is new. I need a scientist can actually have Dana analysts and more Junior people just like senior software Engineers down to Junior software engineer. So you can expand that as need one of those critical areas. It's a bit different than people often don't think about is if I'm Building Solutions round data science. How do I know if the numbers correct is 97 the rights for the one of the things I encourage the people think about is if there's a quality assurance space that is different and does require some more of these math skills than for example in a traditional engineering and bark.

Brown's We've Got Talent Management with build up some capability round data science of the statistics and then engineering really looking at if you need to cheat to new technologies analytics streams big data streams band, you may need an architect and especially some what you might call data ETL Engineers massage the data for you. The one reason thing is you can do most data science using existing database Technologies, and so on, you know from SQL from almost any stores, but what we're seeing is an ecosystem of tools develop around these newer technology and I think that's a strong strong influencer in towards moving to those Technologies, you know when you talked about Outsourcing

You talked about it was sort of thing it completely or keeping it in the house completely. There's actually one in between which is leveraging tools that are out there and in the cloud to do the analytics and at the one extreme you've got something like IBM Watson and at The Other Extreme you have modeling tool kits that can be barely finite and run. So I think it's important for companies also to distinguish between Outsourcing data science or leveraging tools in the marketplace that are not on pregnancy. Can I really like the what I'm hearing is disintegration of not only the product skills and this this data science in this analytics skills or technology from different sources as well as it's interesting you mention some of the more traditional DBA database administrator functions.

What factors would make a these data science teams and these companies succeed or fail first and foremost is data science just need a clear Target and a well-defined target for them to do their job and understanding what it is that's going to really Drive business benefit what burning question needs to be answered or burning questions is really the starting point. I have yet to see a data science project fail when that was clearly defined.

Second thing would be going for mental data science is much more messy than an engineering type of approach because you are talking about coming up with a hypothesis is it early do an experience experiments to prove them and so on so don't over architect and don't over engineered things of the beginning learn to the process when you've answered some questions return back and look at reducing the friction tag at Aspen exit. What would be the best way to attract is this top talent? It is this just compensation. Do we have to think about different factors that might matter more to these type of skills, you know, of course compensation plays a role in specially in in in a field where demand Fire cases Supply so yes compensation.

There's much more to it. But I I found it a few things are critical one is when you're looking to hire and putting out the position describe the problem you're trying to solve data scientist look for a problem to solve and if it's just saying I have tons of data and we're looking for somebody to to to find Value in there will be less interest if they're saying we're trying to improve health care by identifying at-risk patients and being a leader in that that will attract more because they're already saying that some thought about inches chart data scientist don't want to work alone like an engineering or software sales team. They want to work with other people who understand their craft and then they can learn from

So if I were building a team, I would show commitment to building a team. This is not a I want to know if I have a team I would certainly put front-and-center that you'll join our growing team of data scientist focused on driving insights that will change the industry. Whatever that industry is those kind of things are really dry in data scientist. So we'll get them motivated will get them feeling comfortable that they're delivering some real value in the unique space. I very much appreciate that the respect you have and in in calling in a craft you think I rang your programmers feel that way and I'll take what is they go higher and build to these teams and in that respect is very important. So any

My final advice would be to start now. It takes time to build up a craft to take time to build up their bench strength and to amass the data everyday that slips by is the last opportunity. Excellent. I got it Chris. Thank you so much for your time. Thank you John. It's been my pleasure.

Thanks for listening to our compact podcast. If you like what we're doing stay tuned for our next episode when will be talking about why security has to come first for every software company in Big Data.


We live in a world that’s increasingly fueled by data and analytics. But for companies to unlock the value all of that information represents, and improve business processes, they need some very specialized help. In this episode, Jon Prial talks to Georgian Partners’ own Chief Analytics Officer Chris Matys about data scientists. Find out what they do, where you can find them, and what it takes to build a successful data science team.

You’ll hear about:

  • Why companies are building their own data science teams (1:56) -What data scientists actually do (3:18)
  • The four skill sets necessary for data science (4:07)
  • Why it’s important to start your data science team with a product manager (5:08)
  • How to approach data to glean insights (6:46)
  • How data science needs change as companies scale (9:15)
  • Why companies struggle to find data scientists and build data science teams (11:10)
  • How to find and vet data scientists (13:10)
  • When it does and doesn’t make sense to outsource data science (16:01)
  • What factors determine whether data science teams succeed or fail (22:07)