Machine Learning Zoomcamp: Free ML Engineering course. Register here!

DataTalks.Club

Articles Slack Events Podcast Books Courses

Building and Leading Data Teams

Season 5, episode 6 of the DataTalks.Club podcast with Tammy Liang

Links:

Did you like this episode? Check other episodes of the podcast, and register for new events.

Transcript

The transcripts are edited for clarity, sometimes with AI. If you notice any incorrect information, let us know.

Alexey: This week, we'll talk about building a data team for making decisions with data. We have a special guest today, Tammy. Tammy is the chief of data at Platanomelon. Platanomelon is a digital sex toy brand that empowers people to break taboos on sexuality. Together with her passionate and creative team, they infuse data analytics into the company’s day-to-day operations and decision-making. Tammy is also a co-host of a podcast called ‘Data for Future’. I was recently interviewed on the podcast. We talked about many different things, including DataTalks.Club, so check it out. It's a really cool podcast. Tammy interviews people who are in the field of data or sustainability. (1:14)

Tammy’s background

Alexey: The topic for today is ‘building data teams’. Before we go into our main topic, let's start with your background. Can you tell us a few words about yourself? (2:10)

Tammy: A few words? (2:22)

Alexey: About your career so far, yes. I also wanted to welcome you to the podcast, I think I haven't yet. So welcome, Tammy. (2:26)

Tammy: It's very nice to see you again and working together with you for another podcast. A brief introduction about my background – How did I get into data? I was studying in the US with an undergrad degree for business. Then carrying that, I worked one year in Silicon Valley, in the bay area where I really got to know a lot of people who work for big tech companies. They talk about data technology, AI – and I saw the sparkle in their eyes, how passionate they are about their job. That really got me intrigued. After reading some books and doing some research, that landed me to stay in Barcelona to do my Master's in Big Data. From there, my data career started. I first worked with a Spanish unicorn called Glovo. It's a delivery company, almost like Uber Eats. With the company growing, I was dedicated to analytics and supporting different departments. From there on, I joined forces with Platanomelon, starting from zero. I started by being the only one in the data team. Gradually, I proved the power of data analytics and showed the amount of insights we can contribute to the business and to the team. Then, little by little, we now have a growing team of five people. (2:37)

Being the chief of data

Alexey: You started as the first data person that's really cool. What do you do as chief of data? (4:07)

Alexey: Well, that covers a lot. As I mentioned, I really grew my position from the ground up. At the beginning, I was doing a lot of hands-on work myself. These days, I'm doing less and less, but our functionality really covers a lot of aspects of the business. One of our biggest parts is marketing. The nature of our team is communication marketing with e-commerce. So, communication marketing – whenever you do campaign communication, you always want to measure your performance, because what can be measured can be improved. Then among all the marketing channels that we are using, it's also important to compare them against each other to see which channel brings the most efficiency, the first CAC, the most/least CAC – this kind of marketing attribution model is something we're working on. (4:16)

Alexey: But I have to say, due to the nature of our company, when we do publish, for example, on Instagram, Facebook, and other social media, if a link reveals a sex toy with one click, then we cannot really do an analysis of the advertisement. With that being said, a lot of our job of doing analytics is quite tricky. We have to get very creative about how to identify different data sources and try to merge and mash them in different ways to serve our purpose and to make the best educated guesses. On the other side, we also care a lot about the business operation. Since you have e-commerce store going on and the prediction of orders coming in for each product every day, it's very important to have an estimate and then comparing it with our inventory in the warehouse. With this data combination, we also give business insights like, “Okay, we are going to be out of stock in one month. We need to start purchasing.” We also help this operational side. Then on the business side, day-to-day performance monitoring, we are very web/e-commerce driven – the performance of the traffic and different product performance conversion, our Shopify portfolio – we touch upon all these aspects as well. The last aspect that we are going to focus on-

Alexey: That's not all? You have even more? (6:42)

Tammy: Yeah. There's a lot. Just one more product than I’m really excited about is – we are very customer-centric, and we want to understand and listen to our customers better and better. Therefore, we try to do a lot of social listening and customer surveys. This kind of ‘non-Big Data-ish’ data analytics, we are also doing that. (6:44)

First projects as the first data person in a company

Alexey: Yeah, that's a lot. Marketing, then demand forecasting, and then social listening. Those are the three main approaches. What was your first project that you started with as the only data person? Was it marketing or something else? (7:07)

Tammy: It was more about business health monitoring and about building dashboards. If you think about helping an organization to start doing data analytics, the most low-hanging fruit is actually to deliver timely monitoring about the business’s health. What I mean is – we have various data sources that are very important to us, such as our web traffic: “How much traffic is coming in every day? Does it get better or worse one day or the other because of our marketing campaigns? Does our web traffic conversion change from day to day, because we change our web design, introduced new product?” So on and so forth. (7:22)

Tammy: I started collaborating with the CEO and the different departments. Instead of once a week or once every month where everyone struggled to find the information and put together a PowerPoint presentation and spent a lot of time on it, I showed them that by connecting the sources and merging them well, we can discover a lot of rich information. At the same time, we can check it minute-by-minute, on the go.

Alexey: That's cool. So people in the company realized that it's very difficult to find the information they needed, especially when they want to prepare reports, like PowerPoint presentations. This is how they came to realize that they needed to hire a data person who would help them connect all the different data sources in such a way that it was just one click away. Is that right? (8:28)

Tammy: I would say that it's a little bit the other way around. It’s more about the management realizing, “Okay, we do have the need. Everyone is talking about data and for the company to grow further it’s maybe something we need.” But people in the departments that are producing reports day-to-day, they didn't realize that data can help them. I remember at the beginning, I needed to approach each department lead or people in the job functionality and I said, “Hey, how about you introduce me to your day-to-day workflow, especially when you're doing reporting and performance measurement. What do you do?” Then I proposed to them, “Why don't you work with me and we can help you to produce automatic reporting. This will get more information and more in-depth analysis about what’s going on and how the performance is doing.” So I actually needed to convince them a little bit in the beginning. Whenever you want to set up a dashboard and connect the data sources, you need collaboration from the team in order to get the data and help them change their workflow and habits. This can be challenging sometimes, but people tend to be very collaborative once they discover that this actually brings a lot of value to them. (8:51)

Initial resistance

Alexey: I’m wondering – did you get any resistance initially? (10:06)

Tammy: Yeah. There are some projects that are for the bigger picture of the company – those that don't come to the pain point of each individual. You say “Hey, actually, we need each team to authorize or be careful about this data.” You require extra labor for everyone, but they may not necessarily see the benefit directly shedding light on them. So this kind of effort is a little bit difficult. (10:10)

Alexey: It's about making sure that the quality of data is good, right? (10:36)

Tammy: Yeah. (10:43)

Alexey: Because then these people – these teams – are producing some data, right? (10:45)

Tammy: Yeah, sometimes the world is not ideal. You cannot stream all the data and automate everything, unfortunately. We are trying to reduce the amount of spreadsheets we use as much as possible. But in many cases, it's proven that that's the best way to collaborate with such a business-driven company. (10:47)

Alexey: So the management realized that the way people are creating reports is not the most optimal way. They heard that data analytics can help – data in general. This is how they realized that they need to hire somebody. They hired you and you started by talking with every department, understanding what they are doing there, what kind of data they need, or what kind of data they create. (11:09)

Alexey: Then you convinced them – you showed how useful that was. The first thing was business monitoring, right? How well the company is doing, what happens after a change, whether it leads to drops in traffic or increases in traffic and things like that. So what happened after that?

Expanding the team

Alexey: You said, now there is a team of five people. How did you arrive at a point when you needed to hire the first person? (12:00)

Tammy: Yeah, it's been a journey. As I introduced in the beginning, the low-hanging fruit is the dashboard. Then you realize that the team understands “Oh! You can actually solve many things for us!” Then they come up with more and more questions and requests. Certain questions and requests definitely go out of the dimensions that the dashboard can cover. Therefore, we needed to create web apps or models in order to be able to help the business make decisions and solve their doubts. (12:10)

Tammy: One of the things is demand forecasting, which is something that is predictive. Then we need to start building models in-house. Building a model means you need to have historical data. Having historical data means you need to have a data warehouse, where you can store the information and stream it. Then you realize, “Okay, only having your historical sales doesn't help you with forecasting. To forecast, there are many factors to consider.” It may depend on the product or the promotion event of the company, and things like that.

Tammy: It is a company that is active every day – we have a lot of events going on. So you need to ask each department to give you data before they take action, like “Tell me what you're gonna do in the next week or a month.” I don't know about many other businesses, but I tend to see that people’s decision-making changes from day to day. Therefore, it's very difficult to convince the team to give you a forecast of one week ahead when they're not sure what is going to happen. But without that provision, we cannot forecast. This kind of balance is one of the efforts we're making.

Tammy: Going from there, there are some bigger projects that are about model development. “Once we have this model, how do we deliver it?” It’s a project that in the end, we want to benefit the business, right? So how can the business really use it and take advantage of it? We cannot give them a Jupyter Notebook or Py file for them to use – they don't understand it. So how can we build a web app, so that the user can enter and fine tune one or two filters or parameters and get the result that they want? That also includes web deployment and a lot of more things coming up.

Alexey: Were you trying to do this all by yourself at the beginning? (14:43)

Tammy: At the beginning, I was enjoying machine learning and model building, because I also enjoy coding very much. But then at the moment of delivery, I was like “I really have no idea. I totally skipped my data engineering class in school.” (14:46)

Alexey: You had a data engineering class. (15:02)

Tammy: Yeah. I didn't think it was important at all. Now comes the time that I regret that and I started seeking help. At the beginning, when I had my first hire, we had a lot of demand for the dashboard – this was still one of the main tasks that we are dedicated to. So my first hire is actually a data analyst, not an engineer. Then, I was seeking help from my previous professor or I looked for some existing solutions in the market that were kind of ‘plug and play,’ and then you can get your information. I tried, but in the end, the model just varies a lot and the data source and the data output were very different. It started to go beyond my capacity. Asking for help meant I needed to wait because the people I asked were not part of our team. It’s very fortunate that the management also grew enough trust in our team that we can have the capacity to hire a data engineer. This was totally a game changer. From there, we started moving more freely. The team – the analysts – could focus more on the analytic side and the backend, since we had an engineer who gave us a lot of care and information, which is really good. (15:04)

Alexey: Basically, the first person you hired was a data analyst to help you create dashboards. But then you realized that you need to have a lot of historical data that you need to process – that you needed to start somewhere. This was quite difficult for you, even though you like coding. You also said that you asked for help but this was taking forever, so that's why you hired a data engineer. From that point, you had your data warehouse, and building models became easier. (16:35)

Tammy: Yeah. I was actually able to configure the data warehouse. With Stage, I was using the options in the market and I was very proud that I was learning a lot. But it was not enough to keep pace with the growth. Then, just to finish up the flow, the next step was hiring another data scientist in order to go more in-depth and bring more statistical model building and more in-depth data knowledge to the team. Then we realized, since our company's not really AI/ML driven, it’s more about analytics and serving the business to provide insights. Having this side of business communication is very important. (17:11)

Tammy: As the data team we need to work a lot, because if we do not tell the team what we are producing, the tools are developed, but they just sit there and no one uses them. So we would just be wasting our time and energy. Therefore, the last person that we incorporated in our team is called a business analyst /data researcher, whose main job is to communicate from our team to the business side. We would publish something like a weekly, five bullet newsletter – a quick image graph to show business insights and the updates of the team and finishing with a meme. Then some workshops we're doing with a company to introduce our latest developments. That's the current project picture and we're looking to grow it more this year.

Role of business analyst

Alexey: That's interesting, especially the last role you mentioned – this business analyst/data researcher. The problem you had there when you were growing your team was that you realized you have these tools – you have these dashboards, you have these web services – but people were hesitant to use them? Or what was the problem? (18:41)

Tammy: Yes. A big part of our OPR of the data team is to build data culture. We realize that if we put in very advanced things in the data part, it is not helpful. We need to move forward with the whole team. That effort, since everyone is so busy with the day-to-day workflow, no one is doing it. When you're giving a presentation and diverting the data scientists from their daily workflow, it's something they are not 100% interested in. That’s how we realized that we had a need for this role. (18:59)

Alexey: So data scientists should be doing this but that's not their main focus – they would rather spend time doing models? (19:36)

Tammy: Or if data scientists do pass on knowledge or share information, it could be more within the team. But to communicate with the business side or answer quarter questions, it's a different profile. (19:45)

Alexey: Yeah, that's quite interesting. So you have a data analyst, you have a data engineer, you have data scientists, and you have a business analyst. And then you as leader of the team, right? (19:59)

Tammy: Yeah. (20:13)

Alexey: So, five people in total. Did I miss anyone? (20:14)

Tammy: No. But maybe soon. We’re looking for more profiles. (20:17)

Alexey: Can you already share what kind of profiles you are looking for? (20:21)

Tammy: We're looking at hiring a senior analyst. (20:26)

Alexey: Ah, so the analyst needs some help. (20:29)

Tammy: Yes. It’s more about growing the team and to sort of grow a bigger vision and have even more ambitious goals. (20:33)

Alexey: Yeah. You mentioned that you did your Master’s in Barcelona. Then after graduating, you started this position, right? Or was there something in between? (20:42)

Tammy: No, I was working with Glovo while I was on Master’s. I started as an intern and then went full time. Then for some personal reasons, I had to leave for a while. Then I joined Platanomelon. (20:55)

Alexey: Because I'm wondering – that sounds like a lot of responsibility being the first data person in a company. You really need to know a lot to be able to be useful, right? You need to know things about processes, how exactly you collect data, how you get insights from data. You already knew this from Glovo? Or you had to learn most of this stuff? (21:07)

Tammy: Yes. I don't think you necessarily need to know a lot. Rather you should be ready to learn a lot. When I joined the job, as I said, I had no idea – I needed to do ETL/ELT, data warehouse, engineering, web app deployment – I learned all these. (21:36)

Alexey: I hope you at least kept the slides from your data engineering course. (22:01)

Tammy: Well, I believe I was more doing research to search for new products. The space is changing so fast – materials from two years ago might already be outdated. (22:08)

Alexey: I guess that since you don't have a large team, most of the solutions you use there, you're not building them in-house. You use some off-the-shelf solutions. Right? (22:21)

Tammy: Exactly. (22:31)

Platanomelon’s stack

Alexey: What kind of stack do you have? (22:32)

Tammy: We started with Stitch and we are on GCP. In terms of the data engineering process, we're using DBT. Then in terms of visualization, BI, we're using Google Data Studio. Am I missing something? I think one thing that is very important is how we deliver to the businesses – we built an internal Wiki for data. We have a Notion where we do our analysis and track our workflows to present links of dashboards and everything to the company. (22:34)

Order for growing the data team

Alexey: Yeah, that's interesting. Thanks. When I was preparing for the interview, I took a look at your LinkedIn profile and then there was a line for your ‘current job’ that said that your responsibility is “building a leading data team to provide accurate, timely, and useful insights to empower the company to make the best decision possible.” That's quite a packed sentence. I was wondering if we could deconstruct it a little bit. There are many, many different parts and I think like every word here means a lot. I'm interested to know what exactly you are doing there. Because I think every word here sounds very interesting. Let's talk a bit about this and deconstruct it. (23:11)

Alexey: Maybe we can start with “building and leading data teams.” I think we already talked about this to a large extent, like how exactly you realized who you needed and in which order. Maybe the question I have is – say if you needed to do this again, would you hire people in the same order? Or what kind of order do you think is a good one?

Tammy: Very good question. I would say that I actually would change the order. First of all, when I did my first hire… I always participate in talks to encourage women to overcome the ‘imposter syndrome’. But I think I was suffering from the ‘imposter syndrome’ myself, even though I was very sure that I could build a team and that I can achieve many great things. But when I asked for help the first time, I was a little bit shy. I thought that I was still at the phase of proving my value to the company and I couldn't just ask for a very senior profile person to join me immediately. (24:34)

Tammy: My first hire was a junior analyst. This proved to be a very helpful and very talented asset to the team, but that is usually not the case. After this, I also talked with many data leaders and came to the conclusion that at the beginning phase of the team’s development, it's very important to have senior people on hand, especially when you're setting the foundation. In the future development, everything is built up from what you set initially. There are some changes, but many things are built up from here. You need to be willing to invest in bringing some more experienced people on the team in the beginning. That could be a good start. Although I think I'm very lucky in a sense. My first hire was still very talented and brought many good insights. Then, you also need to be brave about hiring engineers when they’re needed. It really solves a lot of problems and headaches – things that took me days, but for them, it's a matter of hours to resolve.

Alexey: In which order would you hire? First, a senior analyst or a data engineer? (26:20)

Tammy: This really depends on the company. If in your company there’s a lot of data infrastructure, then the engineer would be first. In my personal case, I can do analytics myself. So for that part, I'm not worried. But if it's a business driven position, I would actually hire them both at the same time. Engineers are important. (26:26)

Alexey: So the only challenge now is to explain to the management that you need two positions, and then ask for the budget for two people. Especially if you're asking for two senior people, then maybe it's a tough one to solve. Yeah, interesting. So you would hire a more experienced analyst and maybe somebody who's a senior data engineer – maybe at the same time? Then a data scientist, right? Or would you hire maybe a business analyst? Or only at the end? (26:48)

Tammy: So if I'm hiring an engineer first, since the technical side is very well covered, I would hire a business analyst. Given the property of our business – we're very stakeholder-oriented. We talk with different departments and business operations. So it's important for us. (27:23)

Alexey: Did you say that you would hire a business analyst even before an engineer or did I misunderstand that part? (27:42)

Tammy: Together. Yeah. (27:49)

Alexey: So that means you would hire three people at the same time, right? If possible. [laughs] (27:52)

Tammy: So, it would be me in the team and I would hire a business analyst and an engineer. (27:57)

Alexey: Oh, okay. Let's say somebody wants to build a team. The plan is – they need to find a company that doesn't have a data person yet. They get into this company, they work hard, and they show value. This shows that, like you mentioned, there are some low hanging fruits such as building dashboards, and then you need to do a lot of talking. So you do that – you build dashboards. Then the management sees that this is very useful. Then you start convincing them to give you the budget for more people. Then you hire a business analyst and a data engineer. Right? And then you hire a data analyst. Then data scientists at the end. (28:01)

Tammy: I would say that this all depends a lot on the type of business you're in and what projects you have in mind. If all your projects are about building dashboards, they you don’t have a dire need for engineering. But if you're developing more in-depth analyses, publishing web apps, or there's more data from different sources coming in, then yes – an engineer is necessary. (28:47)

Alexey: The data scientist is first because there’s probably a lot more work that needs to happen before a data scientist can join. Right? (29:08)

Tammy: Exactly. (29:18)

Demand forecasting

Alexey: But for you – I think you mentioned that this demand forecasting project appeared quite soon enough, right? (29:20)

Tammy: Yeah. That's actually one of the first things that I worked on when I joined the company. There were pain points of the business regarding “How much are we going to sell? How do we manage the workflow of the warehouse?” I started with that but it's proven to be not easy at all. Up to this day, we're still in the iteration and further developing the model. One of the components, as I mentioned, is how to get the provision of data from all the business components. It's proven to be a little bit of a challenge. (29:27)

Alexey: Yeah. This demand forecasting sounds like a machine learning problem – is that right? This is something you solve with data science – with machine learning. But I guess in many cases, this is something that an analyst can solve. You don't need a data scientist who works with a lot of machine learning, but more somebody who is more on the analytics side. Someone who can analyze the data and then build a simpler model, right? (30:06)

Tammy: Yeah. When I was looking for data scientist profile, it wasn’t about hardcore data scientists, but more about who can do a little bit of basic analytics BI and has a passion for developing models. That has proven to be sufficient for us. (30:35)

Should analysts know machine learning

Alexey: So would you require – or would you consider it a nice quality to have – for your data analysts to know some machine learning or to know some time series modeling? (30:57)

Tammy: I would say so. It’s a test, you know. People who go into the field of data naturally know machine learning and they grow some interesting development models. So it's like a proof of passion for me almost. It doesn't mean that they need to have done 100 projects on it, but it shows that they are willing to learn and develop. That type of curiosity is good. (31:09)

Alexey: I don't work as a data analyst, but my impression is that analysts deal a lot with time series data. All these dashboards, most of the time – this is time series data like profits or losses or some KPIs. Then sometimes, somebody comes and says, “How will this KPI look in three months?” And then you have this time series problem, right? So did this happen to you as well? (31:37)

Tammy: Well, I think our business stakeholders have been more understanding because they know we are not the ones who decide that. Our job is really to help monitor and look at historical information. For example, we had a free shipping promotion. We understood from the historical data that free shipping helps increase a certain percent of our conversion or web performance. So the next time when the business has a goal of increasing conversion or web performance, we can propose that as a solution and give them an estimate about how much difference we expect to see. But we never tell them, “We're gonna be here or there.” It's more about giving advice. In the end, if we run a marketing campaign, it's going to be different every time. The success of the different campaigns varies a lot. It depends on our creativity, the timing, how we communicate. (32:10)

Qualifications for the first data person in a company

Alexey: Yeah, thanks. I have a lot of questions that I didn't send you in the list of questions that I have prepared. So the question I have is, “What kind of qualities do you need to have as the first data person in a company? Should you be more on the analytics side? Should you be more on the engineering side? Or does it even matter? What do you think?” (33:09)

Tammy: I think to start… I have more of an analytics profile. I think the most important thing is to align with the business stakeholder – what we want to do with the data and how we can really help the business. With that said, I would argue that business understanding is actually the most key component, especially when you're leading the team. They also need to learn, and learn fast. There are just so many things that are changing all the time. If you're growing a department and finding talent – I needed to hire engineers when I had zero engineering knowledge. If you can imagine, there are a lot of things I needed to dig into and prepare myself for. (33:36)

Alexey: So as you said, the most important thing is to align with business stakeholders, which is more about your communication skills rather than anything else. I think analysts are, in general, better at communication than engineers. At least this is my observation from my current and former colleagues – that analysts are more into communicating than engineers. Is that right? (44:26)

Tammy: Yeah. (34:52)

Alexey: So probably somebody with analytical and communication skills will be a better first hire as the first data person in a company, right? Just to really understand what the business needs from the data team before starting and building things. (34:53)

Tammy: Yeah. Imagine if you want to start a data team, usually it starts with one person, and you need this one person to lead and grow the team. They need to have these leadership assets rather than technical assets. It is more important for them to know “What profile and what technical skills can I bring into the team in order to achieve our mission?” Rather than having all those skills themselves. So having an understanding of the overall big picture is very important. (35:07)

Providing accurate results

Alexey: Yeah, thanks. So let me again repeat the sentence from your LinkedIn. We already talked about “building and leading data teams.” Next is “to provide accurate, timely and useful insights,” and then “to empower the company to make the best decisions possible.” So I want to ask you about this, “to provide accurate, timely and useful insights.” What do you mean by this? Why does it even matter that the results are accurate, timely and useful? (35:38)

Tammy: Yeah. So there was an error I committed while I was developing the data team. Sometimes you're just very happy that your dashboard is working and you can finally show something that helps the business. But what you don't realize is, the seemingly working dashboard is showing wrong data and that is very troublesome. This could be that some of the data sources got broken. Or when the team is updating information, instead of a dot, they put a comma, and then the reading becomes 1000 times different from what it should be. That maybe could change our product price or product cost and that totally messes up the whole dashboard information. (36:12)

Alexey: What kind of consequences are there when something like that happens? (37:02)

Tammy: Luckily, all of the errors that we committed were discovered by our business departments first, instead of them using it to make business decisions. For example, the product cost or our profit margin all of a sudden shrunk a lot because the operational team put the product costs as a very big number due to the comma-dot difference. Then the business says “What did we do wrong? From this day, all of our profits are gone. Should it be a business emergency?” But in the end, we figured out that the fault is from our side from that incident on. I can't say it just happened once. I got different errors that I wasn't able to foresee. (37:09)

Tammy: I realized how important data accuracy and reduced data downtime is. From there on, with the engineer and different analysts, we really focused a lot on our information – updating, maintenance, and regular checks. Before, it was an area that I would ignore a lot. But now I realize that developing things is cool, but keeping things working and delivering correct information is even cooler. It's even more important. So that’s the importance of accuracy.

Alexey: Did you have any trust problems from the business people after they saw that dashboards had errors? Perhaps they said “Okay, maybe I shouldn't really trust this dashboard. I'll just do my own thing without relying on data and make decisions without using that.” Or you didn’t have to deal with this? (38:34)

Tammy: As this happened, there definitely was frustration. The business initially thinks that you're great and that you're offering them great tools to make decisions on. Then they realize it's not reliable. Of course, there was frustration, but it happens – everyone makes mistakes. We were also in the process of developing a team – there are things that we're learning on the way. We really took it very seriously from there on. We did research, consulted people in the field. We internally developed a white paper – a playbook – for our team to ensure accuracy and to improve data governance. From that point on we also tried to be very open about the errors we've made because of our ignorance and then what policies and solutions we're going to present for the next steps so that we are aligned with the business. They were also very understanding. (38:59)

Alexey: Basically, you tried to be transparent. You said that “We made this error. There are some mistakes in the data. We're working on fixing it.” Right? (39:59)

Tammy: Yeah. (40:08)

Alexey: Okay. You mentioned a playbook. I don't know how sensitive the information about this playbook is. But if it's not, maybe you can tell us in a few words – what’s in this playbook? (40:09)

Tammy: We are updating it here and there. As in the case of the error I mentioned, we need to educate each team how to input information. There are multiple dimensions to this – from the team who is offering us the information, we need to create guidelines and communicate with them. But that is not enough. From our side, when the data gets input to our data warehouse, we also need to do testing and checking. So that's also one of the reasons we started using DBT, because we can do a lot more testing to ensure that the data is reliable, that there are no outliers, and that there is no non-functionality or downtime. (40:24)

Alexey: So you rely on DBT for that, or there's also something else you use for data checks? (41:04)

Tammy: Mostly DBT, and in terms of downtime on the dashboard – before we never had regular checks from the team, but now, on a regular basis, we will have team members dedicate a certain amount of time to go through dashboards and identify if things are working. (41:09)

Alexey: Basically, it's more like a manual process right now. (41:25)

Tammy: Yeah. I believe I haven't found a better solution. If anyone has a better idea, I'm all ears. But so far, that's the most effective and efficient way that we found. (41:29)

Receiving insights in a timely manner

Alexey: I think we only talked about one word “accurate,” right? But there are two more “accurate, timely and useful.” So let's talk about “timely.” Why should insights be timely? Why is this important? (41:42)

Tammy: Well, time is very keen on information transmitting. In the stock market, when you have information 0.01 seconds ahead of everyone else, you can win the game. This is true, especially if your business is fast growing and from many different areas – for monitoring the business house, you can quickly identify the error. If there is a conversion job from ‘add to cart’ to ‘transaction,’ maybe there are some technical problems going on, because you should test the stable metrics. You can fix it before customers start complaining and you start losing profits. (41:57)

Alexey: Like if there is an error in the ‘buy’ button. If it doesn't work, you want to find this out as soon as possible. You don't want to find this out two days after it broke, right? (42:39)

Tammy: Exactly. Another thing, just to give an example, before – when the business didn't have timely information and visibility – say we are outsourcing our warehouse and they don’t have enough staff to deliver all the orders we receive every day, the customers will be unhappy. Then after five days, maybe seven days, ten days – everything is still quiet. But then customers suddenly start canceling their orders, they start leaving bad reviews on Google. This is very passive if you, as a business, don't have visibility about what is happening in terms of operation in a timely manner. So this is another thing – business health monitoring and operational monitoring. (42:53)

Alexey: So, for this warehouse – you said that maybe you outsource this. So you would have a partner for dealing with this and you can’t always control how long it takes. For you, it's important to have visibility that sometimes things take longer than five days. You do this monitoring by collecting the data and when you see that something takes 10 days or more, you see that you have a problem. Is that right? Or when you see that there are bad reviews coming up somewhere. (43:45)

Tammy: That would already be too late. So you need to monitor it before that happens. (44:20)

Alexey: Okay, yes. I understand. (44:25)

Tammy: I think just in general, for example, say you launch your campaign. It’s a campaign that lasts for one day and you couldn't really wait for another day or after a week to see the result. So timely measurement here also means, “Once you air this ad, does the traffic really start coming in? Is it effective?” One of the things we have is campaigning on TV. I don't know if anyone listening is in the attribution model development area, but you have to understand – TV's not measurable. Well, at least not easily measurable. To track the performance, you have very specific spikes in your time series data. If you don't have this timely measurement, you cannot discover. (44:29)

Alexey: Do you also have banners like in real life? Like on bus stops or somewhere? (45:24)

Tammy: Yes. (45:30)

Alexey: I was wondering – how do you measure that? How do you do marketing attribution in this case? (45:30)

Tammy: Actually, we benefit from that. We created a brand that has a lot of customer engagement. Whenever we launch surveys, we have this community who are very happy to give us feedback and information. After the customers make a purchase we usually ask them, “Where did you hear about us?” And whenever we have different campaigns going on, we tend to add them as one of the options. So it's sampling our customers for some ideas. (45:39)

Alexey: So the options you have there are TV, banners on bus stops, and... (46:06)

Tammy: ...Instagram, YouTube, social, all the channels. (46:11)

Alexey: You're not the one doing this – I guess this is your marketing team that’s doing this, right? But your role here is to have a good overview of the dashboards that show the performance of this. Is that right? (46:16)

Tammy: This actually goes much further beyond dashboards and is one of the biggest projects – about attribution model development. (46:29)

Alexey: Ah, right. So it involves a lot of models as well, right? (46:35)

Tammy: It involves a lot of data source identification, merging, cleaning. It's a big project, and we're delivering it in a web app. So people under the page in the HTML, they can choose and select the time and other information and they get the attribution. (46:39)

Alexey: Yeah. That must be quite a challenging project, right? (46:58)

Tammy: Yeah, but it's a lot of fun. Especially the moments when we figure out some ideas about how to tackle it. (47:02)

Providing useful insights

Alexey: Okay, cool. Then the last word. We talked about “accurate” and “timely” and then there’s “useful”. So why “useful”? Why does this matter? (47:08)

Tammy: I think when you're in a technical area of the business, you could be very detached from the business itself. Like “Oh! We should use NLP! NLP is very cool. We should get all the text information we can get from the company and then develop some fancy algorithm.” But then you realize the business doesn’t really need it and you're just spending your efforts developing something that's going to be hidden in the corner of the website. So I think this alignment with business is super super key. We have a very big product mindset. Every product we're doing, no matter if it’s a very simple task or a big project that lasts months or even years to develop, every step of the phase, we deliver a product that the business can interact with and use. Then they can give us feedback. So it's very key for us. Our mission is to serve the business and what we deliver has to be useful. (47:18)

Alexey: I guess it also goes back to having this business analyst role – a person who makes sure that the team is working on something that matters. I guess there is also an educational component. Now you have a dashboard, how do you show that it is useful? So how do you actually show that? You build the dashboard, you know it's useful. You know that you can make a lot of useful decisions on top of that. How do you go about convincing the business operations team and other teams to base their decisions on these dashboards? Or on your services? (48:19)

Tammy: Very good question. To encourage a team to use your product when it's something new – that’s difficult. Not everyone's an early adopter that’s eager to see more things about data. So to tackle that, we do a lot of workshops and informal sessions on data to the company in order to build data culture. At the beginning we were doing it more as a lecture, like “Oh, here is our newly developed dashboard. Here you can find A. There you can find B. And if you've got a question about C, you can find it there.” We've realized that even after the session, people came back and asked us for information that they can find in the dashboard. So we were like “This is not effective. People are not really listening.” So now we changed this session to a Q&A format, which is more like “Okay, now we have this question. Where should we go?” You give people some time to answer and then people can guess whatever, or they find the answer. And then we show them the pass to find the solution. That is proven to be much more effective. You catch more attention and people engage better. From this step-to-step we're building data culture. Many things like that. (49:00)

Giving ownership to the team

Alexey: So this is something you really need to know well as the chief of data: You need to make sure that the results are accurate. You need to make sure that there are timely quality checks. You need to make sure that there are no delays in data. You need to make sure that there is adoption of your data product and your data services – that people not only use them, but also understand when to use them. And that they're eager to use them. These are all things that you and your team need to take care of. To do this as chief of data, that's a lot of work. (50:18)

Tammy: Actually… I would say that it’s nothing that tires me. There are a lot of things going on within the team. But one of my leadership mottoes is that “I need to give ownership of all the projects to everyone who is doing it.” That really helps. As a leader, especially when your team is growing, you cannot micromanage and hold hands with everyone in the team – that really drains you. But if you give more of a direction and give everyone ownership of the project and give them enough resources or encourage them to find whatever resource they need to complete it, then my job here is only to be at their service – or whenever there is trouble that they cannot solve themselves. Then I help them and step in to help them troubleshoot. In the end, everyone is doing much better than I would have done it myself. We are all contributing as a team. That way, we are moving quite fast and I don't feel exhausted or tired at all. (50:52)

Alexey: But at least you know that these things are important and you also communicate this to your team. (52:02)

Tammy: Yeah. Totally. (52:07)

Alexey: Then of course, you don't go there and implement DBT yourself. Somebody else is doing this, right? (52:08)

Tammy: Yeah. I do spend a lot of time doing research to keep myself updated with industry trends. And I talk with professionals to know, “Okay, we have this problem.” Because, in the end, I'm communicating with my team whenever there is a blocker. Say we have this blocker, we cannot really solve it at the moment. How should we do it? That's more my job. (52:16)

Starting as the first data person in a company

Alexey: Yeah, I think we managed to cover only half of that sentence so far [laughs]. And I see that we don't have a lot of time left and there is a question. So I thought maybe we go to a question from Bayram. The question is “What resources do you recommend to be successful in the path of starting as the first data analyst and leading the team? Books, courses, programs, tools, etc.” (52:39)

Tammy: I actually think select communities are so far my favorite source to get input. For example, DataTalk.Club and where I met Alexey is in this community – Locally Optimistic. There are many groups. Personally, if you have time, get a course and the follow-along book. I like more module information. I also like to see in real life, how it's happening. With communities, there are people asking questions, seeking advice all the time. Whatever is on your mind, you can just publish there and there are for sure some people who have gone through the same thing or who know some information that they will share with you. It’s in this kind of ‘chat’ setting. You can even write to people, like, “I'm very interested in this kind of content, can you have some guests that can talk about it?” Or you can reach out to someone on LinkedIn. So I'm more interested in the community aspect. I like to reach out and talk. (53:05)

Data For Future podcast

Alexey: Yeah. Actually a few people from the DataTalks.Club community reached out to me saying, “Hey, I have this problem. Can you invite somebody who can talk about this specific thing?” And this is really cool. So please do that. If somebody is listening and has a problem that you want to solve or get some inspiration – please do reach out. I think for you, it was also interesting, right? Because you also have a podcast, right? Maybe we can spend a couple of minutes talking about your podcast. So what is it? (54:09)

Tammy: Yeah, sure. My podcast is called “Data For Future” and we're doing interview-style talks with founders and professionals in the space of data and sustainability. I do believe in ‘data magic’ – it helps with so many things. Sustainability is one of the biggest challenges that we're facing right now. I'm actively looking for people to spread ideas, get inspiration, and more than anything – to curate a community in order to be able to talk about this subject matter and maybe to inspire more action. (54:43)

Alexey: What did you see in this space? Why did you decide to create this? I guess there are not that many podcasts about sustainability and not many communities about that? (55:17)

Tammy: Yeah. To be honest, the community about sustainability is pretty big. But the cross-section between data and sustainability is not. Actually that poses a challenge for us, either to find more profiles in the space or to find a bigger audience who will echo with us. But we do believe there's a huge potential and for now, really, the mission is to learn. For myself, the biggest gain is that I've learned so much and get to know many cool people to build this community and a network. (55:30)

Alexey: Okay, I see a comment in YouTube Live Chat from datadev saying that they follow your podcast, Data For Future. (56:08)

Tammy: Cool. Awesome. Thank you. (56:17)

Supporting team members that are stuck

Alexey: Please, everyone else also do that – Spotify, Apple podcasts, whatever you use to do that. We have another question from Bayram. “How do you support your team members if they are stuck on some problem? Let's say that they're working on an NLP model and they get stuck – they have a problem – and you yourself do not have the knowledge to help them. How would you support them?” (56:19)

Tammy: I think that's one of my strengths. I don't know the answers to everything, but I know where to find them. (56:45)

Alexey: Google? [laughs] (56:55)

Tammy: Yeah, Google is always the first source to turn to. Then, as I mentioned, community is another one. These days, a lot of the questions for things I don't know, I post in, for example – Locally Optimistic – and then people with a lot of experience give me really good resources and things like that. So it's one of my go-to areas. Then, for everyone who is learning, maybe you come from a boot camp or a Master’s background – your professors, your network, can help. Or just read books. Because at our level of expertise, unless you're like super, super top-notch, all the questions you are facing, someone has already suffered from it and probably found a solution. There’s likely already a very well-established solution out there. As I always tell my team “You need to be resourceful and know what questions to ask.” The same applies here. Everyone has the same access to Google. But some people know how to ask the right question and they get to the solution very fast. (56:57)

Alexey: I’m wondering how to train this skill. (57:57)

Tammy: Actually, for example – say I want to start with NLP. I will Google it, I'll find a community. If they have events, conversations – I join. I will find my professors for NLP and I'd go talk with them. Any events, I am all in. Really, by talking with people, I discover so much. (58:00)

Alexey: Okay, so your learning is more driven by the community rather than by textbooks and courses. (58:21)

Tammy: Actually, courses – for sure. But it depends on how much time you have and how much dedication. Taking courses is actually one of the best things to do if you want to develop more in-depth knowledge. Books are a big time commitment. Sometimes you get bored if no one is explaining on the side. But some books are really well-written and they can be really helpful. (58:30)

Alexey: Some textbooks have a lot of good materials, but they are boring. Indeed. I remember some books about machine learning that are very heavy. I remember when I was studying machine learning – that I had to carry them to class [laughs]. Okay. We should be wrapping up. Do you have any last words before we finish? (58:57)

Tammy: No, I’m just happy to be here. I hope this is helpful for everyone. I appreciate the questions. (59:21)

Finding Tammy online

Alexey: How can people find you? (59:28)

Tammy: On LinkedIn. You can search for my name. My podcast as well – dataforfuture.org. I'll be there. (59:31)

Alexey: Yeah, I think you have a website and there is some sort of contact form, right? (59:40)

Tammy: Yes. (59:43)

Alexey: Okay, I remembered correctly. Then I guess that's it. Thanks a lot for joining us today and sharing your experience. For telling us how you went through being the first data person in the company, to growing this into a team of five people and now hiring even more people. So, thanks a lot. Thanks everyone else for joining us today, for watching, for asking questions. Yeah, I guess that's it. Thanks, Tammy. (59:44)

Tammy: Thank you, Alex. Thank you. Bye bye. (1:00:20)

Alexey: Have a great weekend. Bye-bye. (1:00:24)

Subscribe to our weekly newsletter and join our Slack.
We'll keep you informed about our events, articles, courses, and everything else happening in the Club.

DataTalks.Club. Hosted on GitHub Pages. We use cookies.