From Ecology PhD to Global Marketing at HP | John Stanton-Geddes | Data Science Hangout
To join future data science hangouts, add it to your calendar here: https://pos.it/dsh - All are welcome! We'd love to see you! We were recently joined by John Stanton Geddes, Senior Manager for the Advanced Analytics Marketing Team at HP, Inc., to chat about his career path from academia to marketing data science, the challenges and opportunities in marketing valuation using data, the evolution of marketing data science teams, and the impact of data privacy on user tracking. In this Hangout, we explore the work of John's team at HP, which involves global marketing valuation. This includes developing media mix models to understand the return on investment across diverse marketing activities, from sponsorships like F1 to digital advertising like Google paid search and in-store promotions. John discussed the complexity of integrating data from these varied sources and the ongoing efforts to refine their models and use experimentation to drive effective marketing decisions. Resources mentioned in the video and zoom chat: FRED (Federal Reserve Economic Data) → https://fred.stlouisfed.org/ R package fredr → https://cran.r-project.org/web/packages/fredr/vignettes/fredr.html Semrush → https://www.semrush.com/ If you didn’t join live, one great discussion you missed from the zoom chat was about whether Excel is the "gateway" to data science, with many attendees sharing their agreement and experiences. Was Excel your gateway?! ► Subscribe to Our Channel Here: https://bit.ly/2TzgcOu Follow Us Here: Website: https://www.posit.co Hangout: https://pos.it/dsh LinkedIn: https://www.linkedin.com/company/posit-software Bluesky: https://bsky.app/profile/posit.co Thanks for hanging out with us!
image: thumbnail.jpg
Transcript#
This transcript was generated automatically and may contain errors.
Welcome back to the Data Science Hangout everybody! If we haven't met, I'm Libby. I'm a Data Science Community Manager here at Posit. I'm also a Data Science Educator. I have experience teaching R in Python and I am based in San Antonio, Texas. If you're not familiar with Posit, Posit builds enterprise solutions, open source tools for people to do data science with R and Python. We're also the company formerly known as RStudio. So if you've used RStudio, you have used Posit. We are known for things like the RStudio and Positron IDEs and package ecosystems like Shiny and the Tidyverse.
I am joined today by the creator of the Hangout and my co-host Rachel. Rachel, would you introduce yourself? Hey everybody, I'm Rachel Dempsey. I lead Customer Marketing at Posit and I'm based in the Boston area. I'm so excited to have Libby here taking over as full host of the Hangout, but I'll be here behind the scenes and chatting with you all in the chat here too.
The Hangout is our open space to hear what's going on in the world of data across different industries, chat about data science leadership and really just connect with other people who are facing similar things to us. So we get together every Thursday, same time, same place here on Zoom and make sure that you have it on your calendar for 12 p.m. Eastern time.
Alrighty, well I'm so, so excited to welcome our featured leader today, John Stanton Geddes, Senior Manager for the Advanced Analytics Marketing Team at HP Inc. John, I would love it if you could introduce yourself, tell us a little bit about your background, what you do, what you like to do for fun.
Thanks Libby and thanks to everyone who joined. I have a PhD in Ecology and Evolutionary Biology from the University of Minnesota. And I picked up a minor in statistics, kind of the equivalent of a master's at that time. For anyone who, like a small fun factoid is that because I had a minor in stats, I had a statistics committee member and that person's Charlie Geyer, who wrote the original MCMC package on CRAN. At the time, I barely did anything daisy and barely understood what it is. I look back years later and I'm like horrified how incredibly naive I was when I worked with Charlie and some of the questions I would ask him and things I did.
So, you know, my message to some of you out there who are new and you're stuck, it's like it's okay to be clueless. Just get there, but also like pay attention to the people you're around and what they're doing, what they think is important, because there's things I, you know, could still probably go back and ask him that would have been helpful or will be helpful in my career.
John's career path from academia to industry
After I did a PhD, I then moved into the genomics realm. I was still in the ecology area, so I was actually collecting ants and crushing them up and then analyzing RNA, which is super cool. But eventually, I found myself moving away from academia and I landed in a data science position at a small company called DR.com that did digital marketing for automotive dealerships in Belmont, Vermont.
The company, I joked I got a crash course MBA, so I came into what was kind of a late stage startup. They had been acquired once. I was the second data scientist at the company, so they didn't really know what data science was, which is great because I didn't either. So, we kind of got to learn together. Then the company was acquired once. They were acquired a second time by Cox Automotive, which is a large kind of umbrella that has a bunch of different brands underneath it. Just fantastic for me because, again, I got wide exposure to different areas that data science is being used. The key one was they own Kelly Blue Book. So, Kelly Blue Book has been doing data science since before data science is a thing because their core business is vehicle valuation models.
I was there for a number of years and worked on a wide range of projects. Again, it's kind of fun about a company that size. They got to do things like building recommendation engines for vehicle recommendation engines for auto trader. Also got to be involved in some of the vehicle valuation work as well as in the digital marketing space. That was a core part of the dealer.com business and auto trader.
So, I have strong opinions about paid search and Google and those types of things. That experience eventually led me to when I looked for other alternatives, moving to HP, which is where I am now.
Job changes are good and challenging for many ways. One is that you have all this experience that you think is great. You're going to land somewhere and then you get somewhere and you find out, oh, all these things you think you knew are only a small part of the picture. HP is huge and the way they do marketing is very different. There's a lot of traditional, believe it or not, that still happens. This led me into the media mix modeling space.
The timing also there was fortunate or coincidental. A few weeks ago, Google released something called Meridian. Facebook has had their Robin package out for a few years. So, there's a bunch of different frameworks out there for media mix modeling. Basically, we had a head start by year two, where we've been developing our own internal framework to do this. That helps make decisions across the different 4P, your product, your placement, your pricing, and your promotion levers at HP for our products.
I have three kids, so I'll be at a hockey game tonight. Myself, I'm in Vermont, so I love skiing. I personally prefer cross-country skiing. In the warmer months of the year, I enjoy getting out on Lake Champlain, paddleboard or kayak as well.
What John's team works on
My team, as you said, we are a global team. We support marketing across HP globally. We have people, the team's mostly in Europe and the Americas. There's people in Romania, Switzerland, France, Spain, UK, the US, Mexico, and Colombia. Within our group, the digital marketing returns and analytics team, we focus on a number of different areas. A core part of what we do is what we refer to as marketing valuation. This is developing the media mix models that we have.
We take in information on our investments across all our different activities. This could be sponsorships. HP made a large investment in F1 with Ferrari. That type of information, as well as Google page search. You're very low funnel capturing people on Google, but there's also TV, there's radio, there's connected TV. There's also in-store activities. Walking to a Best Buy, you're going to see promotions there.
The data is extremely diverse. The data you get from an in-store placement and the data you get from Google are very different. How do you build a model that effectively evaluates the performance of those dollars and then helps you make recommendations is a challenge. It's something that I was like, yeah, check, we can do this. Years later, we're still tweaking and figuring out how to do this properly.
Because this is complicated, there's no one model that can actually solve this. There's a whole part of the team that's focused on experimentation. For in-store placements, you can take the advantage that either you just stagger when those are launched, or you don't do them in some stores. There's some natural experiments you can set up. You can turn your spend off on Google. There's lots of experimentation you can do. We have a group that's focused on experimentation. Then we also have a group that we call our algorithmic decisioning group. They're more in the weeds in some of the specific platforms like Amazon.
Amazon is becoming a huge player in the advertising space. At the end of the day, the point is, if you look, I think in their most recent numbers, I think Amazon ads business made more money than, maybe not more than AWS, but it's up there. It makes Amazon a lot of money. Obviously, that means that if you're a company the size of HP and you're investing a lot of dollars in it, there's a lot of opportunity for optimization. One of the things that's both exciting and challenging about this work is one or two points of efficiency can make a big difference for our bottom line.
One of the things that's both exciting and challenging about this work is one or two points of efficiency can make a big difference for our bottom line.
It's not just a classical analytical or a classical data science problem. It sounds like it's also an optimization problem. Yes, absolutely. We're doing Bayesian modeling using R and Stan, but then we have an optimization framework that's built by them. We have multiple models. These are sequential and hierarchical models that gives outputs depending on the granularity of the type of input and our confidence in that data. We take these and we use that information that we put into optimizers to help the business make decisions.
Moving from academia to industry
For me, it's a very personal decision about where I lived and my family that was an opportunity came to move into this time, very new space of data science, and it seemed to make sense. I'm not going to say academia is a safe route. I think one of the beauty of academia is that there's always a new research project. There's always something different to do there. So it's always interesting, but at consequential points in my life, I feel like I've always sought out something that I really knew nothing about.
I don't consider myself a risk taker, but I look back and I'm like, okay, when I had these options to continue what I'm doing or jump over to HP and try to build media mix modeling frameworks before things like Euribion existed, and now it seems much easier than where we are. I take those decisions. And so I think part of it is looking for a challenge, constantly challenging myself. And part of it is just that I was living in a place that I liked and I had young kids and academia is not terribly supportive of those priorities.
A lot of academics end up having to move a lot in order to find jobs or stay employed. Well, how did you gain experience in marketing then? If you went from where you were to marketing, granted, there's tons of behavioral aspects of marketing, right? Yeah, I got super lucky. And it's interesting how things have changed.
There was one in Burlington. They're like RUG, right? The R user group on Meetup or maybe somewhere else. There still is a somewhat active one in Burlington, Vermont, but there was a very small one. In Burlington, and I would attend that. And when I say small, it was like three to six people often. And there's someone there who basically, he knew enough to understand that the methodology you're using if you're in genomics or marketing is very small, right? You have an outcome that you want to model, you have budget data, you have low signal to noise ratio. And he was basically able to give me a recommendation and a reference that got my foot in the door. And then, so I knew nothing about marketing when I started. I had to learn on the job, which when you're in an office, in person, you can turn that around and ask people dumb questions very easily.
I think that ability to learn quickly is a key asset. And finding people, right? As much as learning, it's like finding the people who will help you.
What makes a good marketing data science team
The people who are really effective are the people who like in my space, you like really understand, not, I would say marketing and digital marketing is like a marketer, like how do you structure a campaign? How do you go out and figure out what someone needs and wants are and build creative that appeals to them? Like that's that right brain, the creative side of brain, right? Like that's still very important.
If people who, and the thing about this is it's so achievable, right. Google and Amazon wants you to spend money, their platforms, they have certifications and training. And if you have some like small side project or you're on a nonprofit for your kids hockey, like go run a meta ad campaign, you know, a couple hundred bucks, just so you see like, what is it like? What kind of data do you get? Like, this is something I needed years ago. So you have some hands-on experience, even if it's not on the job.
And then, because when I do interview marketing or for people for these data science roles in marketing, the people can talk intelligently, like, oh yeah, on Amazon, if you set up an Amazon shopping on our ads, like here's the weird things you run into, or here's the challenges. It's so much more impressive than someone who tells me about the neural network they designed. Now there's, the methodologies change so fast, right? And the frameworks are a dime a dozen these days, right? Like there's a new best in class AI every month right now. And, but knowing in a business sense, you're trying to drive business outcomes. And so knowing how, what the data is, how to work with it, that's, what's really a differentiator.
Getting buy-in from the business
Yeah, that's a great question. And that's going to vary tremendously by, you know, the company you're at and the maturity of your, your models and your data science and, and not surprising. The most important thing is the relationships, right? You have to have integrity and you have to have some accuracy. And I was lucky when I came to HP, like there was an existing framework in place, right? This is a strategic decision made to move it in house. And I would say it's definitely a case where you want to have small wins. You don't want to just like disappear and for six months and go do your work in asylum and come back and be like, here's the answer to everything. Trust us. It's not going to work, right?
It's building your case, finding situations where you can step in and really provide help. And so there's this weird cadence to our work where we have these, in data science, we tend to have fairly long time drivers, right? Projects are like one to two quarters out. But there's, there are these moments where like, if you can put in a three day spike where you like, they need to make a decision tomorrow about where that next million dollars goes. And if you can come back and give that recommendation, back it up with data. And I think importantly is provide one of the proof points that your recommendation worked. So they can look through weeks later and say, oh, look, we followed the recommendations and it was a positive outcome, right? This is what we're looking for. Then the next time you go back and it's a $10 million one, they're not going to be like, no way are we trusting these idiots with this decision. Like we're going with our gut, right? They're going to actually follow what you recommend for your models.
As I mentioned, we do experimentation. So it's both, right? You have models. You can also, we try to always set something up. Like if we're making a big change or a big bet, we frame it like, hey, here's the best knowledge we have and our best recommendation, but we should also do an experiment to validate what we're saying here, right? The trust but verify mentality. And I think it helps to work with data-driven people. It helps to work with people, you know, find those people who want to be data-driven and want to work with you.
Data sources and quality
Yes. So all of the above, we have obviously a lot of internal data. We have data that is external that we either, we get a number of sources. The best data, not surprising, they didn't pay for, right? Because then there's someone who's providing a value-added data service. So they have a vested interest in making it really good quality data. So in the order of what I find the problems with data, the best data is you're paid for from someone else. After that, it's probably the data you go out and pull from public places, right? Like the Fred here in the US and World Bank, that type of data, weather data. And then when you get down to internal data from your, I won't name any teams, but there's teams that manage everything out of Excel and Windows 10. That's your worst data.
AI and the future of marketing
Too soon to say. It changes so fast. To some extent, the thing is, like, there's always been search engine optimization, right? So you could argue this has already been happening for years, right? People are putting in keywords and doing things on their websites to improve their SEO, right? You're trying to reverse engineer Google's search algorithm to get up there, right? You could argue that that was a very simple AI.
And then today, right, when you look at things like, you know, that people, I did hear statistics on, you know, people under 20, what percent of them, if they go to get a new product, if they're just going to Gemini or chat to your team and saying, what laptop should I buy? It's already astoundingly high. So it absolutely will impact how marketing occurs and what's done. I think the big players there don't fully know what to do yet. I believe it's perplexity even has like this whole shopping thing out there. And they're trying to figure out how to monetize it. And it will change things like slowly and then rapidly.
Using Posit Connect
So as Rachel mentioned, it started my nature for a while. The prior company is out. We started using Posit Connect. So I think we're pretty early adopters of it. And it is an invaluable tool to be specifically the tech platform as a data science team. The biggest challenge you have, in my opinion, actually isn't the R squared on your model. It's not. We spend lots of time talking about data, but those are solvable problems with time, smart people and money. The biggest problem you have is that at least for the manager of data science team is this is that Alan's question I guess is getting the buy in from the business, right? Making sure that what you're doing is getting used.
And Excel has long been the solution there, right? Send someone Excel file. In some ways, Excel is the ultimate data product. Your version control your data is in there. You can build pretty impressive, you know, the UIs I've seen built in Excel are pretty impressive. It's all falls apart, right? When you have like really big data or complex algorithms, but often big data and complex algorithms aren't when you get to solving a business problem, that's not the largest challenge.
But what Posit allows you to do is the tech platform is to take your model to take your data and publish it. And something looks very professional, right? You're not sending a stakeholder an Excel file, especially with Shiny and Quarto dashboards or something. We've started using very quickly that are really impressive. As well, again, anyone who knows R or Python can spin up a Quarto dashboard like they can. You can go from zero to a dashboard in like three hours, and if you can't, you probably should be in data science. And it provides a level of polish on top of it that is will impress stakeholders.
And the big challenge there, though, of course, is like once you have this, it looks great is, you know, you can you can host it yourself, there's things you can do. But if you have authentication, there's all these other issues that you have. How do you get it out there, right? And they probably Excel files, you send an Excel file, and you find out there's a bug in your, you know, formally can't fix it, right? If you have a dashboard you've published, you can always change it and fix it. So you don't have to like urgently resend and don't use the last Excel, use version 2.3 Excel and people don't remember which one. And so that's why having this publication platform is so valuable, right? It centralizes what we do.
Non-technical hiring attributes and imposter syndrome
As I said, hiring is stressful and really hard and I actually just don't feel qualified to answer. I've gone through a number of different, the technical skills are still the most important part, right? So there's various ways you can do a technical assessment, whether there's tools like Hacker Rank or providing people with case studies. My least favorite is like the dropping a technical question on someone in Zoom and like solve this, you know, sharing your screen. I've seen that happen. I've done it and I apologize. I think that's a horrible way to assess someone.
From a hiring perspective, what I rely on a panel of people and recruiters to help assess some of those other qualities, right? So it's not just me. And an important part is frankly, someone's resume as well, right? Like what else are they doing in their life? What experiences do they have? What interests do they have? And let that come through, you know, in a screening call or at other times.
In this space, I think there's two things. I think meetups like this, best is in person, but even like a virtual one like this, I think is a really good way to figure out what's going on, get some exposure, things you might not know about and, and hopefully make some contacts. In-person is better. You can find some of these in-person contacts and connections. And, and then I think side projects, I think like those are under evaluated. Any artist you talk to is going to have their, their photographer's going to have their portfolio online or an artist is going to have some information about their art online. So if you're really looking for that change and you've got some time, like invest in your side projects and, and put those on GitHub or get that out there in such a way. Referencing Quarto again, Quarto makes it super easy to make a website, you know, like make a website featuring your side stuff. And then that passion will show through no matter how introverted you might be.
Like passion is obviously something I look for. If someone's really passionate about, you know, bowling statistics and has a website about their bowling league, they've tracked all the stats and they can talk to me about it for a while in the interview. Like, I think that's great. I think that person will be able to translate that passion to their work, even if it's not bowling.
Like passion is obviously something I look for. If someone's really passionate about, you know, bowling statistics and has a website about their bowling league, they've tracked all the stats and they can talk to me about it for a while in the interview. Like, I think that's great. I think that person will be able to translate that passion to their work, even if it's not bowling.
Data privacy and the future of user tracking
Yeah. So that's a tough question to answer in a few minutes. You're right. These privacy regulations are coming in, and because HP is global, a lot of business in the EU, we're very forward focused on this. So they're trying to avoid, even if those things are still legal in parts of the United States, trying to avoid those things. So from a technology perspective, there's a lot of stuff out there. If you look at like LiveRamp, they have their ID resolution, their network. Cleanrooms are this new thing, the technology is getting dangled out there. So they're privacy compliant. And so Google has their Ads Data Hub, cleanroom technology. Amazon's got a cleanroom, so there's lots of cleanroom technology out there.
I'm a little bit of a cynic, so I'm not saying I'm often wrong about things too, but my grandfather, I have to get a family reference in, he was a very avid fisherman. And I remember one time he made a statement that stuck with me for many years, which is the fishing lure has to catch two suckers, and the first one has to walk into the bait shop. And that's what I feel about a lot of these technologies. Who are they selling these technologies to? It's marketers, it's data people like, oh, I just have this ID resolution, I can perfectly track my audience. But a lot of these things, do they actually work or not? Are they providing incremental value? We don't know.
The fishing lure has to catch two suckers, and the first one has to walk into the bait shop.
We want to think they do, so we set up experiments, we run models that show us like, hey, Amazon ads have this great incrementality and this super high row ads. But if someone's on Amazon, they're already in market, they've already typed in, searched fishing lure. So were they going to buy if you ran the ad? Not anyway. Amazon's not going to make it easy for you to figure that out because you're paying them money for that slot. So I don't know what the future there is. And perfect targeting, I think, often actually just captures somebody who's already super low in the funnel. They already made a purchase decision, and maybe you're capturing them at that decision point.
Career advice
Yeah, so I think I partially gave one earlier, but I'll reinforce it, which is in business, I've never heard someone talk about like adjacencies, right? Like a business strategy always look like, what's that adjacent space to move into? So if you're 3M, you have scotch tape, and then they have scotch guard and all these scotch things, right? So as an individual, you should do that as well. You asked about when I moved from academia to being a data scientist, like I somewhat joke, but I think it probably a little bit seriously, like the thing that nailed me the interview is knowing how to use GitHub and version control, because I was at a company that had one data scientist, a bunch of software engineers. And like I was the person that showed up, I was like, oh yeah, yeah, version control. I know how to use this thing.
I used it because I like majorly screwed up an analysis, and I couldn't recreate some code. And so I was like, how do I prevent this mistake from happening? And so I found this thing I could use, and it took a little work and it's a little bit weird, but you know, like was my postdoc advisor happy that I spent probably more time than I want to admit learning how to use it? If he knew it, probably not, but it benefited me down the road, right? So there are these things you can do. Today, frankly, it's probably playing with some different AI tools that are out there. Learning how to use something like Cursor. Cursor is amazing. Use it. Learn how to use that, even if it's not directly relevant to what you're doing.
So I think learning those adjacent skills to what you're doing and keeping out of the space, personal connections, meeting people, figuring out what else is out there is still invaluable. I did see some research about what percent of jobs come through recommendations. So building those networks is really important. And you can do this today with LinkedIn and online and some, you know, referencing back to what I said earlier, you can do that. And, you know, I'll first like those passion projects, like find those things you're excited about that you can work on.
Yeah. And join the Data Science Hangout and come hang out with people in the chat. Introduce yourself. And my biggest suggestion is to go find the people that you are connecting with in the chat here on LinkedIn and actually message them and talk to them and ask them questions. Get together on Zoom. There are so many people that I'm looking at on this call that I've met on Zoom just before, you know, before I even hosted the Hangout. So go do that.
All right. Well, I'm so excited that everybody joined us today. Thank you so much for hanging out with us. Thank you so much to John for sharing all of your experience with us, to Rachel and the crew behind the scenes. And I wanted to let everybody know that next week we're going to have Blake Abinanti, Director of Analytics and Data Science at Suffolk Construction. So join us next week. Same time, same place. Have a wonderful day, everybody. Bye.