Resources
Oct 12, 2021 |
1h 23min
|
668 views
|

Data Science Hangout | Óli Páll Geirsson, City of Reykjavik | Data Science is More About People

We want to help data science leaders become better.

The Data Science Hangout is a weekly, free-to-join open conversation for current and aspiring data science leaders.

An accomplished leader in the space will join us each week and answer whatever questions the audience may have.

We were recently joined by Óli Páll Geirsson, Chief Data Officer at the City of Reykjavik.

► Subscribe to Our Channel Here: https://bit.ly/2TzgcOu
► Add the Data Science Hangout to your calendar: https://www.addevent.com/event/Qv9211919

9:44 - Providing value to stakeholders in data science
12:35 - Why data science is more about the people
15:06 - Communication with stakeholders = Crucial
17:21 - The value of building up your data science team
18:08 - Why you need a diverse data science team
20:15 - More efficient data science teams by breaking down goals
37:46 - Active listening to better identify needs for your data science projects
43:55 - Prioritizing the projects your data science team works on
49:37 - The best way to approach key stakeholders
1:07:06 - The importance of visualizing data products

Follow Us Here:
Website: https://www.rstudio.com
LinkedIn:https://www.linkedin.com/company/rstudio
Twitter: https://twitter.com/rstudio

image: thumbnail.jpg

Transcript#

This transcript was generated automatically and may contain errors.

Welcome, everyone, to the Data Science Hangout, and welcome back to all the familiar faces and to those joining for the first time. We really want to use this time to focus on questions that are most important to you all. So no agenda, everyone's welcome to join in live or put any questions you have in the chat. We do also have a Slido for any questions that you want to ask anonymously, too. So, Rob, if you wouldn't mind putting that in the Zoom chat, that would be great. And just a quick note that we will record this session, so we'll post it up to YouTube. And as you may have heard us just chatting about, also take little snippets of really important points as well and put them there. But for today I'm joined by my co-host, Óli, Chief Data Officer at the City of Reykjavik. Óli is focused on bringing value from data to residents of Reykjavik. Óli, I'd love to have you introduce yourself and maybe share a bit about your team and the work that you do.

Yeah, of course. I'm happy to be here. I think this is an important discussion framework for us to have. This is an interesting field. And I think many of us are doing these things for the, some of the things are being done for the first time or second, like they're relatively new. We need to learn from each other, at least in my case. So, as Ash mentioned, I'm the Chief Data Officer of the City of Reykjavik. And data is there, it's playing a central role recently. And it's relatively new. I think that's the reason why we need to keep the discussion going and learn from each other. It's like, we're kind of inventing things that haven't been there before. So, I'm going to go out on the line and claim this.

So, just a few words on my story. And while I'll switch from Icelandic to English, it takes some time to get into gear. You know how it is, two Icelandic speakers. So, a few words on me. My background is in mathematical statistics. I did my PhD seven years back, finished seven years back, which was on basic hierarchical modeling. And all of my modeling was carried out, of course, in R, like the same person. And so, when I was doing my PhD research, I was also doing a lot of statistical consultancy, both for start-up companies, for big companies, and also within the academy. So, once I graduated, I spent a year in postdoc, but I felt that I needed to be somewhere else. And so, after the PhD, I worked for a company called Quisup. It was a computer gaming company here in Iceland. And after that, I went to Ellis Retail, that is also an Icelandic company, specializing in POC software, POS software. After that, I was with Lundspanken as a senior data scientist, where I laid the groundwork for the use of machine learning at the back, which was extremely interesting because it wasn't just risk-related. It was data science in general, and what can we do with data science. And that was the point that I was trying to get to earlier, is that using data science as a part of your operations as a whole, that is relatively new. I'm really familiar with these things aren't new, but including them and incorporating these aspects into your business. So, it's not just a nice lab, it's a must-have. That's nice. That's fun. And that is, I feel like, gives me a sense of purpose at work.

So, but off to Lundspanken, it continues. Olli, I hate to cut you off here, but I'm getting a little bit of feedback in the audio, and I don't know if it might be from AirPods. So, I don't know if we could maybe try just using the computer audio. Yeah, of course, of course. Let's try this. Yes, there we go. That's better. I think they're starting to break down.

But that's another note. So, my journey at the city started two years ago, and the city council decided that, hey, we want to be a modern city. We are nothing without data. And that is just something that city council has been discussing. And I find that fascinating. They realize that data plays a central role in the operations of organizations these days, and are aware of big companies like Google and Netflix and all of these things, Amazon, and they're data-driven. And being data-driven is something that we need to breathe and feel. And this is along the lines that it isn't just nice to have, it's a must-have in your business. So, but it needs to have some point. It needs to bring some value. And this is something that I would like to emphasize on this discussion today, is that being a part of an organization as a data science team or a data science leader, or a data leader, or whatever, isn't about algorithms or data science in that sense, or cool structures, or cool inference methods. It's about seeing where you can bring value to somebody in your business. And this is, of course, obvious when you discuss that, but you need to be okay with letting go of this amazing model you spent weeks training, and realize that if it isn't bringing value, then you're doing something that is pointless for the organization. It isn't pointless for you, but it's pointless for the team as a whole.

It's about seeing where you can bring value to somebody in your business. And this is, of course, obvious when you discuss that, but you need to be okay with letting go of this amazing model you spent weeks training, and realize that if it isn't bringing value, then you're doing something that is pointless for the organization.

And I feel like if you don't see the value and you're not okay with having these discussions, then you won't make any progress within the organization. So, we, the data science team at the City of Reykjavik, we have this momentum from the City Council, but we need to bring the action to the other departments in the sense that we need to show them how we bring value. And I feel like this is for at least the team or the team leader, or somebody has to be responsible for this aspect of the data science operations.

And bringing value isn't only your responsibility. It's your and your clients' or your stakeholders' responsibility. And it can only happen in a dialogue and in an open discussion with your stakeholders. Let me give you an example. Like, we have various stakeholders at the City. The residents of Reykjavik City are the largest stakeholder. But let's demystify this. Let's take an easy example. Like, one of our biggest stakeholders that is within the city are all departments like school and leisure. And they know their business. They know what they're about. And we can't, like, do data science for them without working with them and having a dialogue with them such that they understand the value they're bringing. Like, how are we adding to your operation with data science? And so being in this role, you have to be okay with, hey, I need to explain things that are obvious to me and realize that not everybody is as excited about data as you. And be okay with that.

You need to have a discussion with all stakeholders. Like, hey, this is great. You can do this. Did you realize that? And this is awesome. And they need to fit it into their own framework. And I feel like in my experience, like, be energetic about this. Be mindful of their needs. And if they're, like, bringing up the wall and being difficult, there might be a reason for that. That has nothing to do with you or data science. It's all about them. And slowly but surely, you can bring value. And realize that people are different. And what I find amazing in this job is, like, data science is more about people than it is about AI. Isn't that weird? You can't just sit in some corner and train AI models and expect things to be amazing.

data science is more about people than it is about AI. Isn't that weird? You can't just sit in some corner and train AI models and expect things to be amazing.

Building a data science team

So, you have to have this in your team. So, a few words about my awesome team. And I feel so privileged to be a part of it. So, I have thoughts on how to assemble a team like this. And of course, there's certain categories and certain domains that you want to make sure you have checked in some box. And of course, you need to have, like, these data engineers. They are responsible for getting data into some sort of domain, like a data warehouse or data lake or whatever technology you're using. Their responsibility for the underlying framework. And I often have, like, this metaphor for using a data pyramid. And the first layer of the pyramid that you're trying to build are the data foundations. Getting data, cleaning it, and maintaining it. All of these things. I'm assuming that everybody here is relatively familiar with that. And the skill set there is a bit different than the skill set that you're looking for when you're looking for state scientists. But there is an intersection between those two domains. But they aren't necessarily wholly the same.

So, once you have these strong data foundations, and then you need to bring in the data scientists, the talent set that can go into the data and find patterns and bring value. And I think it's incredibly important that the data scientists are okay with maintaining an open dialogue with the stakeholders and explaining things that are obvious to them. Like, hey, you know, this is the mean and this is the median. And just be okay with that. Put on your teacher's hat and enjoy the ride. And I think it's much more about that than training extremely complicated models, at least in some cases. And I guess this varies between business domains. But at the city, most of the effort is in the dialogue. But we sometimes hit problems that are difficult in some sense, are involved. And that's fantastic. But that's just part of the game. That's not the whole game.

And that is the second layer in the pyramid. And I think the third layer is what I call and I think in the literature is called data ops. And that is operationalizing some of your products, your data products, and your data products might be APIs, APIs that you expose to your clients. It might be connections into Excel or Power BI or whatever. It might be like shiny dashboards. And we use a lot of shiny dashboards. And in our case, the operational side is maintaining and designing and operationalizing dashboards and also access to models and access to plots. So in our particular case, it's an RStudio Connect.

And so all of these things are aiming to, you're trying to build this pyramid to get to the top part of it, which is bringing value. You can have all these nice things. But if you aren't bringing value to somebody, and I'd like to emphasize that it's not necessarily only monetary value. It can be value in the form of better services, in a better experience for the residents of your city, like in our case, you need to define it. And you need to be able to measure it. And then you need to live by it. We brought value through this product in this sense. I feel like when you are, when everybody in the team realizes that, you get this positive feedback loop into your team from your clients, that you and your team belong to something bigger than yourself, and bigger than your team. And then, at least in my case, and I think I can talk for the team as well. And then you're more, let's say energetic about going to work the next day.

And the team consists of data engineers, it consists of data scientists and data ops and product manager and me. And there's some extra obvious things that I'd just like to say out loud. Everybody knows this, but I'm still going to say it out loud. You need to have a diversified team. So embrace diversity. It's so obvious, but it's so true. You need to be diverse in age, you need to be diverse in gender, and in all of these things that you can be diverse in. Because then you bring different mindset into the mix. And I also feel like when you're not surrounded by people that are just like you, you automatically become more professional and try to bring your A game all the time. That is just something that I think is true. I have no idea whether it's true or not. But it's the thing about having different mindsets and don't discriminate towards age. Like somebody who's been 30 years in the game is not too old for data science. And there's something that I had a discussion on this recently. And I just, in this particular

field, which is this crazy mix of science and thoughts and disciplines and discussions and being humane, then I think it's more important than ever to diversify.

That's just like how I think about things. And one other thing that I'd like to mention is that so it's so important to get the sense that you belong to something bigger. And there are various ways to do that. But I think you mainly do that through your stakeholder relationships and building up a good team spirit. But I have some tricks that I'd like to share for building up my fantastic team spirit and belonging to something bigger than yourself. First off, I'd like to emphasize the importance of being agile in your development. And I'm not saying that because Agile is cool or the main thing I take from an Agile approach is that you have dailies and you have opportunities to set these and take these big goals. Like you want to run a marathon, but you're going to break it into pieces. And when you do that, you realize that, hey, this overall big goal is not something that I need to think about all the time. I only think about individual small goals. And that is nice because when you reach them, and then you can celebrate with your team. And then you get this positive feedback that, hey, I did something important. I didn't complete this whole marathon, but we've reached our short sighted goal or like short term goals. And on daily or planning, I share that with my team and got this positive feedback that, hey, I did well. If I didn't, you know, break this goal into pieces, you're losing these opportunities to reenlighten the spirit or the sense of purpose in your team. And I think during COVID times and you work from home, I think that's actually important when you have team members sitting at home and they kind of think like, who am I? Why am I here sitting in the same seat all day long? I need to belong to something. And yeah, so I'm totally sold on these tricks.

And the second trick that, these are just tricks, you can't just do these things and hope for the best. These are just extra flavor. But I'm really keen on allowing the team to have their own time on Fridays. Just work on something that I think is interesting, not necessarily projects, but it has to be data related, data science related. It can be something that hasn't been like, can be a project that, like in our particular case, one of these awesome team members decided to play around the data on garbage collection calendars. Those weren't in a digital format here at the city. So now he built up this interactive garbage calendar, which is now a hit product at Reykjavik City, which is something that was totally unknown, but he just rolled the dice and we thought that was cool. And now this is our most popular product.

But these are also the times and opportunities for you, for the team members or for your team to upscale themselves. And don't be too strict about it. And just like, hey, go play, but go play in data science, right? Because that's like having like this one lottery ticket, you never know what comes out of it. Worst case scenario, your team members know something more. And I love it. I already have a product now from that lottery operationalized. So yeah, it was totally worth it in my case. There's so many more things. Just please ask away. I'd love to share them.

Worst case scenario, your team members know something more. And I love it. I already have a product now from that lottery operationalized. So yeah, it was totally worth it in my case.

I love that point you made that data science is more about the people. And I see, Rob, you just put that in the chat as well. But Oli, I know I have the advantage of knowing a few more of the use cases from the team, from our past conversations. But I think it'd be awesome to also maybe share a few other examples of what the team is working on to provide value to residents.

I have slides ready. I can show those slides. I can also spin up some of the task boards. Oh, it's okay. Just kind of like a few. I mean, I love the swimming pool example, but I'm sure Yeah, okay. Okay. I love the swimming pool example because that is one of these surprisingly complicated process.

The swimming pool project

Okay, okay. So let me take you back. So Icelanders, they really like swimming pools. And that has something to do with that, that Iceland sits on this geothermal spot. For you guys who are not aware, we have an active volcano now. And if you don't believe me, I can send pictures. I can almost see it from my windows sometimes, just this right there. So this is a geothermally active island. And so we have these geothermal plants scattered around Reykjavik city, and they provide the city with abundance of hot water and electricity. So as a result, we have a lot of swimming pools, like crazy, crazy many swimming pools. And this is, I think, the most favorite pastime activities for Icelanders to go to a swimming pool. But since it's so popular, and it's also popular with tourists, is that we have been asked frequently, like, hey, can't you supply us with data on how many guests are in the swimming pools?

And so I've got so many requests from various people. When are you going to bring us a dashboard on swimming pools? Like how many guests are currently in the swimming pools, like live data? We thought, like, hey, that's easy. So there are these gates that count in guests in most of the swimming pools in Reykjavik. But the thing is, for some reason, so one would assume, like, hey, I just come in and the gate comes out. It's just plus and minuses and everybody is happy. And we can maybe query this gate and ask, like, how many have you counted out and counted in? So it turns out that these gates that are now in operations, most of them don't count out. And I have no idea why. It's just probably they were cheaper. I don't know. But they are now in place. They count in. So we know how many guests have gone into the swimming pool, but we don't know how many have gone out of it. So we have this accumulation, this kind of like an accumulated distribution that never comes back down. So how many people are in the swimming pool? A million. It's not a true answer.

So what to do about that? And in one of these swimming pools, Løvesløk, they also count out because they have more advanced gates. So we have data on how many guests have been counted in and how many have been counted out. So we have like a historical distribution on how long each guest is staying at the swimming pool, in the swimming pool, like, and so we can model that. We can build up a model that says, hey, so it's actually a temporal, it's a time series model that can predict like, hey, now I have this guest arrived at this point in time, according to my model, which is a linear spline model. And you can go into the details if you want to, but it's a linear spline model that can predict, hey, it's this time of the year and it's at this time. And I predict that guests that are coming now should stay on average 90 minutes. So then we count them out after 90 minutes. And so we count in and then we have a model that counts out. So we have an accumulated distribution with pluses and accumulated distribution with minuses. And then we can calculate some kind of estimate on how many are in the swimming pool during every hour.

But the thing is, here we need to think about, here comes, now we think about this project as a data scientist, or we need to zoom out a bit and ask, okay, where's the value? The value is knowing approximately at which capacity the swimming pool is at. I only need to know whether it's like 70, 80% capacity or 20%. I don't need to know as a guest whether there are 50 guests or 52 guests. Does that make sense? Like you might be discouraged to go to the swimming pool if you see it's at full capacity, or you might not. It depends on your needs. So then we are okay with errors, but we still need to have some way to predict within certain degree of some estimations and with some degree of uncertainty. So that's how we do it. We applied that model to the other swimming pools, and we are currently asking those swimming pool executives to evaluate whether this makes any sense. But there is a catch, and this is why it is one of our favourite projects.

But the thing is, here we need to think about, here comes, now we think about this project as a data scientist, or we need to zoom out a bit and ask, okay, where's the value?

In Lerpersloe, that doesn't count out, that counts out, sometime during the day, the staff says, to help with it, there are too many guests here, we're just going to open the gates and re-evaluate in the morning how many guests were. So we need to do some estimation on how correct the data is. I'm not going to go into those details, they're very minute and weird, but I really like that. So yeah, we have like this model running in our RStudio environment. It's like the data comes in, it comes onto our RStudio computer that's in the Azure cloud, so it's a virtual machine, and then it queries the model, like, hey, how many, like, for how long should this guest stay? Oh, he's going to stay for 90 minutes. And then it will do the aggregation of the pluses and minuses and then give the feedback back to the API. And I just think it's really cool. It's data science and some arithmetic.

Yeah, yeah. Not all of the projects are like that, but we'd rather do that than spend 10 million S&D kroners on new gates. Would you say that's your favourite project?

Yeah, it's the top 10, at least top five, because it's just, it was so laughable, it was so funny to see that, hey, this is extremely complicated, but can we actually solve that? And it turned out that we could. So that was fun. Yeah, so we have other super interesting projects, and I think that this swimming pool project is going to have a real value for the citizens of Reykjavik, because we really love going to swimming pools. And this is just something in our DNA.

I love that. It's interesting to think of, like, our own problems that we're solving, where maybe we have something, it's the same case, where, like, at a certain point, people just open all the gates. And so now all the data is messed up. Like, I wonder what those are in individual industries. Yeah, yeah, absolutely.

So, Frank, I see you had a question specific to the swimming pool example. Do you want to ask that live? I'd love to, as soon as I'm off mute. So, actually, two things that come to mind, Ali, like, really interesting example. I'm trying to visualize how your team worked on that over time. And I'm curious how many analysts slash scientists you had working on it, just roughly, like, I know sometimes it's tough to quantify, right, this person a little bit sometimes.

Yeah. So, I have a fairly accurate estimation of that. So, this has been, I think, not counting the summer vacation. I think it's been a four month project up to this date. The first hurdle was basically getting access to the data at that touching time. So, being like an official battery or entity like Reykjavik City, you need to make sure when you're doing contracts or when acquiring things that, like, in the case of these gates, like, you need to make sure that, hey, the data that these gates have generated, they belong to us, and we will not pay an ounce or krona for getting these data. And that particular contract

was not phrased well enough. So, that took some discussions back and forth. But once that was out, and they were taken care of. And I think when we realized that the gates will respond in this way,

I think it's been a month where we tried to figure out whether we could actually solve this. But that was not, the data scientist who was working on this one was not doing that full time. It was doing like maybe 25% on and off. But I think that is a good approach when you hit a problem that is new and unsolved, not like business as usual, to just let it sit with you and let your unconscious mind work on it. So, let's say 25% for a month. And now it's ready and in testing. But in actual time, it's been four months. And one data scientist and me as well, because I can choose projects. And I'm like, you get to pick on the best stuff.

Yeah, it was just so interesting.

Cool. One other thing, and this should be a quickie, but I'm curious, after you built the thing and built the model and built out the dashboard, how did you know, how did you validate that it was working? Did you like go to the field and say like, oh, yeah, great results, say 60%, looks about 60% full? Yes. So, almost like that. So, we asked the,

you know, the swimming pools are being operated as a like, like individual companies. So, they have executives. So, we asked them to compare, like, does this make sense with what you see and what you think is true in your swimming pool? And that is like, I think that is a good enough evaluation without actually going and asking, like, when are you going to go out of the swimming pool? According to my model, you need to go after 20 minutes. But if it's in, I would say, if it's within, if you're thinking about proportions, thinking about, like, capacity proportions, if you're within, like, 20% accuracy, I think you're doing fine in this particular case. Like, because it will not affect you as a resident or visitor, whether 70 people or 80 people in the swimming pool, but it might affect you if they're 10 or 10,000. Yeah, depending on your needs. I'll volunteer to go double check.

Yeah. I'll go hang out at the swimming pool. Yes. Rob, I see you asked a question in the chat, too. Do you want to ask that live? Yeah, well, you kind of answered it a little bit with Frank's answer. But you talked about collaboration with stakeholders. And, you know, you're in a sort of public service-esque position and citizens are your stakeholders. I was just wondering what that actually looks like. Yes, this is a tricky one, because I don't think there's a single answer here. And so I think we always need to bear in mind that ultimately what we're doing at the city are we are servicing the residents. And even though you are a data scientist, and then you just need to know this, you're a servant, you serve them.

and then you just need to know this, you're a servant, you serve them.

But I think the swimming pool example is an interesting example, because this is something that many residents were asking, and we're getting frequent requests. It wasn't to me only, but also to the department that is responsible for the operations of the swimming pools. And through the departments, like the Department of Welfare, School of Leisure, and Urban Planning, they get there. So what we're doing now to get cities going to digital transformation phase is that the departments have these digital leaders, and they need to be aware of what is the most important project for them. And that is ranked with respect to what is important for their clients. And the Department of Welfare has certain clients in urban planning, planning growth and stuff. And they get the needs from the residents through their operation that goes back to us.

Building a data product portfolio for residents

But I have a particular project in mind. So we are starting now to build up like this portfolio of data products that is supposed to be available on the web. And it's supposed to be some kind of data market or a data store or dashboard store for the residents of Reykjavik. And I think that's a really cool product. And I think we should build that. But we can't do that without understanding the needs of the residents. Do the residents actually need to know how many bikes were going around here every hour? Do they need to know that? That's something that helps them to go through the day. So when we are building something that is particularly for the residents of Reykjavik in a perfect world, that needs to be addressing some of the needs that they have. And we have done that with surveys and asking and having these open platforms for residents to get ideas in. And so some of these things have worked quite well. Others haven't.

But like for external data and external products that are like especially made for the citizens, you have to like keep an ear open, like listen. Like in the case of this calendar data product that is Trask calendar data product, that was a screaming need. Somebody had to do that. And we listened. So this is in an active dialogue and also by active listening. So I'd like to phrase it like that. But there's no one platform to decide that, if that helps. But the city is a large institution, a large battery. And you need to realise that the employees in all the departments, they know what their clients need. And they can give you that feedback. So you can build up these products and they go. So I have a dashboard here. So that I can share with you guys if you want to see it, which is particularly aimed for the Department of Welfare. And we built the dashboard with their needs in mind.

So the Department of Welfare has in operations a lot of institutions of hand houses for their clients. And they have employees there and they need to know a lot of things that I can't articulate about them. But they're important for their operations and for people with special needs. So they know what they need and they ask us to build that. And we have an open dialogue on how to represent that and how to get the info from the data. So that is another example of a data product that is not necessarily an external one, but is enhancing operation for a certain subset of the population. So I like that. I was just gonna say, I'm sure you get a lot of requests then if you're using all these different formats for coming up with ideas. I was curious, how do you like prioritise which projects the team works on?

Prioritising projects with the matrix

Yeah, that's a fantastic question. And with an interesting answer. So the data team belongs to a bigger department or is a part of a bigger department that is called Department of Service and Innovation. So we are tasked with the digital transformation of the city of Reykjavik. Basically enhancing and building the digital infrastructure. Not only data and data science, but the digital world as a whole. So I would argue that if you're building data products and supporting data decision, you're a big part, you play a big supporting part in the digital transformation journey. So with this digital transformation and changes and effort comes a lot of investment. So we are responsible for staggering amount of cash. In my mind, in my simple mind, it sounds weird, but it really isn't.

So the idea is our department is responsible for prioritising projects and person. And so we built up this, so we want to do this in a data driven fashion. So we built up this, we call it just the matrix. And the matrix has a certain factors in it, columns or whatever. And they are like one of the questions that we ask that belong to the matrix, how is this bringing value and to whom? And we have certain rating range or a numerical range that represents some value. The highest score would be brings a lot of value to employees and residents. And we need to be able to articulate that why it's bringing value. And that scores something. And then we have other factors that we score by, is it supporting the digital infrastructure or is it a hindrance in some way or is it alleviating some pain points or whatever. We also have more, so we have like these mandates and vision that we work by, like does this correspond to our green initiative. And also does this correspond to equality, gender equality and overall embracing diversity within a city. We're trying to score, I think if I remember correctly, it's seven categories. And they all represent it and the numerical values have some meaning behind them. So when we do this rigorously like that, we get like a list of projects that are scored efficiently. And then we are in a better position to argue for doing things this way. And since our, like we need to be responsible, we need to make the argument for our choices to city council. And so when we do this in this data-driven fashion, it's easier to defend your position, if you can think about it that way. But it isn't built only for that. It's built for us to understand why we're prioritizing projects. And before we built up this procedure, we chose projects into the data team, basically just like the first year was just being cowboys and just thinking something that was interesting. But now we are more rigorous in thinking about, hey, we need to build up a data lake. Hey, we need to build up this and data APIs and all of that, such that we can start to, that we can expand and grow efficiently.

So there's more to the story. So the other departments, they also need to prioritize their projects before they come into us. And we prioritize them. And this is not only data, this is the digital transformation effort as a whole. And how we are, the framework that we're building is that we are injecting these teams agents into the other departments and we call them digital leaders. And the teams agents or digital leaders, they work closely with the department heads and they have a certain mandate to work. And their task is to get an insight into the operations of their departments and see opportunities for digital transformation. So they are tasked with like getting their hands dirty and trying to find out, hey, here are pain points, we can do this, this, this. And then they bring that to us and that goes to the matrix and then we start to go and then we go to work. And so it's extremely interesting. And for the past half year, for me, it is more like being a part of this digital transformation journey than only data science. Data science plays a vital role, but it's just a piece of the puzzle.

Yeah, that was a really long answer.

No, that's helpful. Thank you. Rick, I see you asked a few questions in the chat that are related to the story as well. So I want to make sure to pass the mic over to you as well, Rick, if you want to ask those live.

You mean about getting what they really need?

Well, I see that you just asked about if a department's needs didn't match the required needs. Yeah, one thing that I'm really curious about is that, especially, I mean, it's really nice that you say you want to give them the value and you really want to give them something that is going to be helpful. But one thing that I'm always kind of aware of is that people don't know really what they want because they don't know what's possible. And it's really surprising when you talk to someone that's not involved in data science, like you can do that. And they're really

surprised. And it's something that's like, yeah, like 20 years ago, we can do that. And so like when they don't even know what's possible, how do they know really what they want? And how do you kind of tease that out of them? That's the trick, isn't it? Is to find this common ground with somebody who's an expert in their own business and their operations and knows nothing about data science. And I think the short answer is there is no one single way to do this. But the long answer is have many discussions and be aware of continuous development processes. So you can start out by doing something simple. And did you mean this? And have a dialogue on that. Then meet again and meet again and meet again and build something simple and build upon it. So I'm really fond of this concept of continuous development and building up some things like step by step.

And I think what happens in a process like that, if you have a good relation and open dialogue with your stakeholder, is they realize that, hey, what? That's awesome, right? And just to give you a super simple example, like one of these dashboards that we are building, and that has nothing to do with machine learning or anything like that, like we use a package in R called Plotly to build up graphs. And you can slice and dice on a dashboard and filter out the data. And then you can download the image. Like you can download it as is. Like how it appears. And that was just a game changer for somebody. And I had no idea that they didn't realize that. You know there's a download button, you can just press it. And this only happens because we talked frequently and you just, yeah, I'm not sure whether there's a silver bullet here. But I think the approach is, dare I say, humility and realize that nobody thinks the same. And most people have no interest in data. I just, if you're okay with that, like that doesn't affect me as a person or my interest.

I'm just okay with explaining, hey, a linear relationship, that's something that is, like looks like that. Hey, Simpson's paradox, did you hear about that? Like, if you're just okay with that and also curious about their business, then something can happen. I think the, when you're meeting especially a new client or a stubborn client or a stakeholder that has a complicated set of procedures and a complicated framework, then this needs to happen gradually and with humility. Like I honestly think that's the only way, because you can't come in hard. You can't come in and say like, yes, neural network, now we're going to do that. Like, yeah. Do you think that people do that? Do they come in and they try to oversell? Do you see the data scientists do that? Like they come in and they try to oversell and push their idea?

dare I say, humility and realize that nobody thinks the same. And most people have no interest in data. I just, if you're okay with that, like that doesn't affect me as a person or my interest.

Learning to meet stakeholders where they are

I can't, I'm not sure. I know I have done it over the years and I have learned this the hard way. So I have a story to share when I was at the bank. And so we were trying to push the boundaries there. Like we were trying to incorporate data science into many aspects of the operations of the bank. And I remember quite distinctly this meeting I had with the corporate department. And what we're trying to, I was building a model that can help them quantify and categorize companies that were not doing business with the bank. So basically a clustering algorithm for companies and clustering with respect to their portfolio, their economical size and all of this, all of these terms that I don't know in English, something to do with operations. But basically what I did there, so the algorithm I used to demystify these numbers was I normalized them and then I projected them on a normalized space. And then I did some clustering algorithms and projected them back in order to understand that, hey, this company belongs to this category that says, okay, it's a big fishery or a small bakery. And then why I would like to do that is because you approach these companies differently with different products based on which category they belong to. And the approach that I was adhering to is nothing new in this world. It's been done for decades.

And I came in hard in a way that I started to explain to them how the algorithm worked and how it is superior to the way they're working. So, and that was coming from a good place. Like I was trying my best to explain that was made probably from coming from a place of, hey, I need to convince them that this is correct. Whereas what they needed was why does this help me? Not, hey, they don't need to be convinced that the algorithm is this way or that. And so that what I mean is like, I just hit a wall there and I didn't understand like, why are they not super excited about this? This is awesome.

And the interesting thing there, like after that meeting and my old manager at the time just had this discussion with me, like you came in way too hard. And when you do that, even you were overselling, not necessarily overselling, but you're doing it incorrectly. It didn't find this common ground that they needed. You were playing basketball, they wanted to play football or water. And so you just didn't understand the game. And this was just so distinct in my memory, like, okay, I need to be the teacher. I need to be on their side. And when we went back to this, then we went about this differently and talked about, hey, this can help you sell products because we know what kind of company this is. Something like that. And then like, hey, yeah, now we're playing basketball, whatever they're playing. Does this make sense? I know this isn't the general approach, but... Sounds like your origin story to data science is more about people than about data.

Yeah, that's the insane thing. I'm super excited about AI. I teach AI at the University of Iceland, but employing it is so humane, like deploying it and operationalising it and getting people on board. This happens through discussions because I also had dialogues where the people I was talking to thought that I was trying to take away their job. And what I was actually doing was I was helping them. And because they, from their point of view, thought this, and then we couldn't have a discussion until later when we just started again. So especially for stakeholders that don't see how it works, are stubborn, have been there for a long time. You need to respect their boundaries and respect their needs. And that is not easy. That happens through continuous dialogue and trying to inject ideas, being patient. Somebody is totally on board, and that's just awesome. I think they're more and more like that.

Yeah, that's the insane thing. I'm super excited about AI. I teach AI at the University of Iceland, but employing it is so humane, like deploying it and operationalising it and getting people on board.

When stakeholders ask for data you don't have

Olli, kind of on a similar thread, there's an anonymous question from Slido. Do you sometimes meet with stakeholders or communicating back and forth with them where they're asking you for a specific product, but they don't realise or really accept that you don't have the data to create that? Yes, yes. I've been in this position. I've been in this position, I'm going to say several times, thinking about an interesting example. I guess, so in the case of the city, we have so much faulty data and data that needs to be cleaned. So we can't do the things that they're asking of us right now. But then I try to say that, but we will do it. It's bound to happen.

But what I often think that, so I'm going to go a bit back into my story, like to the origins. The dark side of data science is when your clients want a certain result, but the data is not supporting it. That is a darker version of your question. And I feel like when you're in, if you're in that position in your life, in your academic life, any professional life or academic life, never sacrifice your values for somebody else's needs. And that also is true for the more benign version of this. If you don't have the data, just tell them and explain, like, I can't do that. I have an example now. So and so we need to do a certain analysis for the Department of Welfare. So this is data on financial aid, but how the data is currently being served into our data lake and how we, how we accept the data. We accept it in an aggregated form. The reason behind that is the financial aid system was built this way three years ago. And also because there at the time, they didn't have the GDPR regulations correctly in place to get more granular data. So the data on financial aid is aggregated on predefined groups.

But when you have aggregated data, you cannot answer all of the questions that you're being asked on certain groups. Let's say it's aggregated on gender, then you of course can't answer aggregation questions on, on AIDS, or, or is there a difference in financial aids between neighborhoods or, or whatever. And you need to have this aggregation, this granularity to do the correct aggregations. And I feel this is something that I'm constantly in this discussion. If I get aggregated data, then I can't answer questions on granular data, but going into the granular data, then you're answering a more, let's say more difficult discussion because that, that is something to do with, with GDPR regulations and personal rights. So, so yeah.

Can I add a follow up question to this, if I can bite in, because the question of making the data say something that it doesn't, which is, it sounds like a very clear ethical question, but it reminds me, I think just last week, there was a really nice article. I think it was an economist. I was trying to find it, but I can't find it. And they had this report where, I mean, it's a phenomenon that we all know that if you change the metric, you can change the results. So there is a relationship. If you, if you consider this phenomenon, but there isn't, if you consider this, it depends like how you define it. And it can be like, oh, I define it measured by hour, or I define it measured by a minute. Yes. Like, like Simpson's paradox, right? If you don't agree. So if you, if they say to say something opposite of what it says, that's an ethical, but if they say, oh, use another metric that I'm going to be more happy with, then that's a bit of a gray zone. And then you'll do that. And then it's also a question of how they communicate and how they use it afterwards, which is out of your control.

Absolutely. And this is something that, I think this is difficult when, when you, you output a certain analysis, certain graphs and certain inference on the data, and that is out there. Then it goes to the city council or, or the media and it's misinterpreted. And, and, and, you know, that's dangerous. And I feel like going into the 21st century, I think one of the most important upskilling, upskilling skills or something that people need to upskill is definitely the data literacy. And just now during the COVID epidemic, like there's so much misinformation out there. I was just reading an interesting article based on the data from Israel on vaccinations, which is a super example of the Simpson paradox, and where, where you could assume that, hey, vaccinated people are, are, are, are getting more effective than unvaccinated. This is totally not true when you look at the data in, in, in a correct, granular form. And just these things, these are incredibly important for

enlightened discussions. Sorry. You hear this a lot that people need to be more data literate, but I, I would, I would argue against it actually to be the devil's advocate, because it's impossible to get millions or billions of people to become data literate, right? Nobody says you need to be auto literate. You need to be iPhone literate. You just get the iPhone and you do it, right? Nobody reads the iPhone manual. So I feel like it's the designer's job to make a product that's intuitive and simple that people use. And the iPhone did that brilliantly. And it's the data scientist's job to produce a data science product that is not going to confuse.

So actually just this week, I had a similar thing where I produced a, I produced a plot and I put a lowest curve and it should not have been on there, but I like it. So I did it and I showed it to, it was actually for a friend in a company and I said, okay, this is just fun, but don't, don't use it. And he goes, oh, but I really like it. And I went to the shareholders meeting yesterday and there it was. And I said, I told you not to use this plot, but it looks so nice. And I was like, so that's my fault because I should, I should just not have given it because I knew it was wrong. I should just not have ever showed it to him.

Right. Okay. I agree with this statement and, and I certainly agree with that. Um, when you're, when, when you're in a position where you need to share important data and important analysis, it is your responsibility as a data scientist or a, or a teacher or a knowledge sharer to do it correctly and, and, and in a way that it can be read and not misinterpreted easily. I, I, I, um, I agree with these, this principle or, and, and point of view that, uh, there is a certain responsibility on the shoulders of data scientists. I absolutely agree on that.

Uh, whereas I also, I think that, so that like here in Iceland, we have like this open course that everybody can take, which is on the basics of, of, of statistics or, or, or AI. It's super nice. It's actually super cool. Um, but, uh, but absolutely, um, with, uh, can I just please quote comics like with great power comes great responsibility.

Um, but, uh, but absolutely, um, with, uh, can I just please quote comics like with great power comes great responsibility.

It's true. It's just totally like that. I've been thinking about that a lot when, I mean, especially with COVID, but when you scroll through Twitter or LinkedIn and you see all these like plots where I can look at that and have no idea what it's actually saying. Um, but how do you, how do you train people to do that or to be able to communicate through visualization?

Communicating through visualization

Yes. Um, that's a difficult question. Um, I guess like when I'm thinking about the, the products that we've made and they've gone through so many iterations and with my new changes for color schemes and, and just color schemes in scale, something like that, that matters a lot because unconsciously you, you like, like blue is cold, red is hot. Like this is just something that we think about and all of these things, um, need to have their space, um, and need to have the discussion.

So what we have been doing and, and so we test this with our, our clients, but we also have like just an access access to awesome designers that are thinking about this design of digital products. And we have meetings with them when we're building a product and, and, and just comments on, on, Hey, what are you trying to convey here? Like these are not the correct colors. I just, one of my favorite, most favorite discussions. So, so I would say in one sentence, pay attention to detail and pay attention to us to aesthetics and because great looking graphs, uh, catch the eye and, and they have this, like they have their metal space, you know, these old school axle graphs, nobody wants to look, no person wants to like print them out and put them on the wall, but you want to build graphs that you could, you have in your living room almost.

But I, I'm, I'm really like, I'm really, uh, sold on this subject and I, I'd like to, um, just there's so much out there, like by actually working on, on, on the subject. Um, interestingly, he came to a conference here in Iceland in 2010, when I was starting out my PhD and I met him there and he told me like, uh, if you spent, uh, months doing research, spent at least a few days to make sure that your graphs look cool. And that's so true, like make space for it.

Connecting with Óli and closing remarks

Yeah. I, I know a few people have to drop as we just get past the top of the hour here. So I want to be sure to ask you, Olli, if people have other questions or want to connect with you, what's the best way to do so? Yes. So, um, you can find me on LinkedIn. Um, I'd be happy to talk there. I think that is the easiest way to contact, to be in touch with me. Um, yeah, it's right there. I think that's like, that's my more unofficial, like where I just share my professional thoughts, not ask the chief data officer, like more like, but what I think about data science as a whole and please reach out. I think this forum that we have here is incredibly valuable. Um, and so, so cool to see people from other countries and all the origins that will be thinking about similar things as I am.

Definitely. And I, I know I talked with your team a little bit about maybe having a meetup in a couple of months or so and getting other cities together too, and create a working group. Yeah, they're all in, and you know, I spent some time and I'm totally on board with that. That's wonderful. Um, if you have a few more minutes to stay on only, there are a few other questions. Okay. Perfect. I know Rory, you had asked a question before I wanted to make sure we got to around like shiny app usage analytics. If you're still on, do you want to add any, uh, additional context there?

Shiny app usage analytics

Yeah, sure. Um, I haven't really built many shiny apps yet because a lot of what I do is very heavily like research project oriented. Um, I have mostly background organizational psychology and so I do workplace behavioral analytics. Um, but I was curious with your, your swimming pool dashboards and all that. Have you started looking into like the, the user page or the use, the page use analytics to see, okay, is that another KPI I can use or method I can use to say this is adding value to people as opposed to just the qualitative input you get from folks?

Yeah, that's a fantastic question. Uh, and the short answer is no, but the long answer is, uh, what we are paying attention. So we need to, this is where something that we need to get more into are which features on our page on our dashboards is being used the most. Um, we haven't, but we have been paying attention to is how many visitors or unique IP addresses are coming onto the webpage as a whole, but not thinking about individual features. And most of our products are just a single page, um, like for our trust calendar. So we also have our voting dashboards and, and we have all other things. Um, and we can see that they're in frequent use. So that's how we rationalize keeping them in operations.

Um, but I'm thinking about, for example, the, I think, which is our coolest. So we have two really big dashboards. One is internal for the city council called CityWatts, and that is a huge dashboard aimed to see how the COVID epidemic is affecting the citizens of Reykjavik in socioeconomic means and, um, and all that. And also have this welfare dashboard that was mentioned earlier, but what would be super cool is to monitor which features are being used most so we can emphasize them clearly. Yeah. But, uh, next step.

Yes. Yeah. I guess, I guess a quick follow-up question, if you don't mind, because I haven't done as much of this. I also wonder if I would lose motivation at some point because generally I feel like data scientists are less, we might be a little less interested in all the aspects of the webpage development as opposed to the metrics that someone could go and find out of them. So I don't know if you've thought about that and thought, Hey, do I need a data scientist for this? Is any, are there any of them going to be interested or is that where you pull in a web developer who's going to be a little more nuanced than that?

That's a good question. Um, that's, um, so in my mind, when I'm referring back to the, this, um, metaphor with the data pyramid that was mentioned earlier, but I called data ops that are sitting on top of, um, data science, that can be web development. And, and I think that, uh, if, if, if you're, you're building data processes that are supposed to sit on the web or that aren't pure models or pure analyses or insights that are more like cool graphs that are in operations or dashboards that are being used to slice and dice. Um, I would argue a perfect team consists of a specialized data ops, um, talent that is, that finds enjoyment from work doing web development.

Um, so there are two in my team that is as one data engineers slash data ops and other data scientists slash data ops. And in my discussions with them, they get this job fulfillment from, from doing both. Um, so if, if in your particular case, um, if, if you're mainly interested in data science, there is a place for that, like, like in, in the team, um, there is one, one that does only data science. Um, but I think you need to be aware of consisting, like, it depends on the size of the company and, and, and, and, and whether it's a startup or not, like you might have to be aware of that you need to do some data ops. And, and then it's nice to be able to use tools that you're familiar with, like Shiny. Um, so the, uh, so you're still in R and so you could still like, um, speak your favorite languages.

Yeah. Yeah. But this is, this is a good topic. And, and I think there's a certain merger between, um, being a data engineer and being a data scientist in data ops. And I think, um, one needs to be okay with doing a little bit of all of those things. Um, yeah, yeah, yeah. But, uh, hopefully in the team, you, you can specialize in a certain domain. Yeah. Like, like, for example, if I had a data science problem that is involved or convoluted, I would think that we'd like, like build some cool models on top that I wouldn't ask the data ops people to do that. Right. Right. Cool. Thank you. Thank you for answering my question. Appreciate it. Yeah, my pleasure.

Data culture at the City of Reykjavik

Olli, I think there's one last question that was, was in the chat that I wanted to ask, even though Kevin, I think had to drop, but he asked about what was the data culture like before you were working for the city? And I think I'd also like to add to that question, like what advice do you have for other people maybe on the call or listening in to maybe start to create this kind of data culture that you have now?

So, um, when I came in, um, that was two years ago, I think that I was quite fortunate in the sense that we already had this backup from the city council and from the majority. So it wasn't, one of the fights that I didn't need to go into was convincing the city council that data matters. They, they are of the opinion and are still of the opinion that data matters. Um, but how it matters was not clear to them. And that was my role to explain and exhibit.

But changing the data culture, um, if you, so, so more about the data culture when I came in, I think everybody was, so at least in my, um, in my department, people were relatively excited about data and, and thought about like, oh, data science is something like something from minority reports. Everybody has to see minority reports, like it isn't like that, but you can think about things being like that. Um, but the truth of the matter was the data culture was broken in some sense, um, meaning that, meaning that a certain architecture was already in place. I like, these are big words. I'm not, I'm not, I'm gonna, I'm gonna rephrase it. It wasn't necessarily broken. The culture was in its infancy and was ready to evolve, but the technology that was in place was, was not the ready to scale up. And I would argue broken.

So the, the technology that was already in place and took a year to, to unwind was all data that was supposed to be used to do analysis, to infer, to connect to everything. Everything was carried out on a single BI platform, everything, all the analysis, all the ETL, all the APIs, all the dashboards. It was insane. And, and some of the ETL loads that were implemented in the, in this, uh, in this BI platform were written in a language that nobody understands. And, and the culture was off in the sense that, hey, we, we will pay, uh, contractors to do our

data science. And I would argue, you need to do data science within your company, because then you learn about the data. It isn't about the technology as I've been thinking, been discussing. It's about the business as a whole, the other stakeholders within the companies and what the data is telling you. That has nothing to do with technology. And, and, you know, outsourcing that, the, the inherent discussions, the, the, the knowledge that the data brings to contractors, all of that, you can, you can outsource technology, but outsourcing the inference and the learning on, learning from the data, that is bound to fail, which it did. Because then the contractors can just say, like, they have all the cards, all of them. And, and that was, that's what happened.

Um, I know this is, I'm not generalizing this for all the, all companies in the world, but this just didn't make sense for regular city. The operations are different between the departments and the know-how is there. And yet you need to empower, uh, employees within the company that know about the business, what the data is telling you, not just shoving it up. And, and if I were to do this all again, I would do it again, uh, like this. Changing the mindsets was having this discussion is that like trying to point out in a friendly and non-confidential manner that bringing, uh, this work into the fold brings knowledge and know-how and value.

And, and I just did that again, again, again, again. Um, and switch there, I think is, if, if there is a certain project, a low hanging fruit, go there, then that can be your, your price that you're always pointing towards. Hey, we did that. Did you know we did that? We can do that with data and having an example, a success story helps, um, helps your journey.

Hey, we did that. Did you know we did that? We can do that with data and having an example, a success story helps, um, helps your journey.

Um, yeah, that was a pleasure. Thank you so much, Oli. And thank you. Well, I know we're a little bit over, so thank you so much for all of your time and the insights as well. I do want to just ask here, anybody that's still here, were there any questions that I possibly missed in the chat that you want to ask live?

One more quick thing I will ask you, um, someone asked me to keep making this a recurring question. Are there certain podcasts or resources that you recommend or things you, you listen to or read?

Yes. Um, so lately I've been more focusing on the human side of things. Um, so for a team leader, I would argue that this is holds true for, for everybody, not only the team leader or the executives or managers, but there is a book called five dysfunctions of a team. Um, that I would argue that everybody needs to read that is who is working within a team. And that has to do with realizing that you can't do confrontation. You can't go into difficult discussions, uh, on, on subject matters unless you have trust.

So if you're working with somebody that doesn't trust you on a certain instinctual level, you cannot have a discussion like that. And these, these discussions where we are, um, have different opinions, where opinions class, those are extremely valuable. Everybody's voices needs to be heard that, that, that coincides or is along the lines of having a diversified team. But that doesn't bring value unless there's trust within the team. Such that I'm okay with sharing my thoughts and I'm not going to get, um, harassed or bullied if I have different opinions than the team. So, so if I were to suggest one thing, then it's five dysfunctions of a team.

Um, I'm with respect to podcasts, I'm really mainly doing Icelandic podcasts, so I'm not going to recommend them to you guys. They're more about like having, I don't know, for the past year I've just really been thinking about how to be a good person and, and how, how to, how to, um, I'm trying to remember the name, like you can, I can maybe just post it later on, but there are certain rules that if you adhere to that makes sense, like confrontations, they are never personal, always bring your, always like be consistent and always try your best. But I think like when you're having difficult discussions that work on data, like why are they not using the model? Just remember it's never personal. It's never about you. It's always about you, always about them. But, uh, um, and so there is a podcast on data science that are, it's called standard deviation something. I don't remember the name. I'm horrible with names. Um, but I think that, yeah, that is pretty cool. Um, and that being a good person one was reminding me of the four agreements. Yes, that's the one. It was the four agreements. Yeah. Yeah. Read four agreements. It's short, but, uh, but it's just so comforting to, to think about when you're having confrontations and you will have confrontations at work when you're trying to move against the tide. It's like, Hey, it's never about me. It is nothing to do with me. And then it just feels easier.

Sums up nicely your quote from the beginning that data science is more about people than about data. And I feel like if we were going to name this, uh, recording something, it would probably be that human augmented AI or whatever. But, uh, so, um, what I'm listening to these days, I listened to a lot of sci-fi and I'm listening, listening to stories by Craig Allenson called expeditionary force. So highly recommended.

Sums up nicely your quote from the beginning that data science is more about people than about data.

Awesome. Well, thank you so much, Olli. I really appreciate your time. And I'm so excited to hear more about the team story and do a meetup with them as well. Yeah. Yeah. And thank you for having me. It's been a vivid and lively discussion and, and all the best to you guys. Thank you. Have a great rest of the day, everyone. Bye.