Paolo, thanks for taking the time to sit with me today. Could you start by introducing yourself?
I studied electrical engineering and I moved into data science about four years ago when I applied to Decibel as a Data Scientist. Prior to that, I worked for three years as a Research Scientist in another start-up called Soundisplay. At the time, I was doing a little bit of everything really because it was a proper start-up – it was 8 people and a very small company. We were developing a virtual music instrument, which was very cool. But that meant working both on the hardware part and on the software part. I learned a lot and my duties spanned from running optical and electrical measurements to programming FPGAs, which are sort of micro boards that can handle digital and analogue signals. So, that was kind of close to what I studied at university.
At Decibel, I started working with Python, so I transitioned into more of a software-oriented career. I’ve always loved numbers and I’ve always loved data and graphs, even when I was studying. I’ve always had this kind of analytical mindset that wants to make some sense of numbers. So, I think that data science was a natural transition for me. I mean, if I had known it before, I would have probably started even earlier because it’s something I really enjoy, and it can be applied to so many different topics. It can be applied to digital signal processing (topic in which I worked at Soundisplay), it can be applied to web analytics (as I’m doing right now), it can be applied to sports events and it can apply to many different things. If you like numbers, you can easily associate pretty much anything you like with numbers. That’s what got me to data science.
Amazing. Similar to yourself, a lot of people I speak to don’t necessarily ‘plan’ to get into data science, but rather, end up in it (for one reason or another). Why do you think that is? PT: It’s a mix of things. Surely there’s a lot of demand for it, and it’s a good salary. So, some people want to move there for the money – although it’s not something I really recommend because you should love what you do first thing. There’s also a lot of demand, so if you have to choose a job based on the market, it’s good to become a data scientist. And as of now, although times are changing, there’s still quite a generic demand for Data Scientists. There are still not very specific requirements when looking for a Data Scientist, for example, a Data Scientist who specialises in NLP, or a Data Scientist who is very good at coding. Roles like Machine Learning Engineer, or Machine Learning Scientist are starting to appear, but we’re still at a very early stage. So, anybody with a minimum scientific background and some coding skills can actually start a career in data science at different levels. You may not be a senior, but you can get into the data science world. And then it’s up to you to go up the career ladder.
And what was the big appeal for you to get into data science then?
In terms of data science in general, as I said, it’s something I have always done it, just in different ways because the data science term was not there at the time. I also did some research at university here in London, and I was working with electromagnetic waves but, at the same time, I was working with data. So, I was trying to figure out algorithms, predictive models and so on, and trying to automate things because eventually, that is what you will do with data. You want to use the data so that you can predict things, you can classify things, and you can automate a lot of processes. So yeah, it’s always been the passion for numbers and the passion for data.
So, Decibel was your first as a ‘Data Scientist’. What made you join decibel, and what’s made you stay?
Yeah, that’s a good question. I think I liked the company from the beginning because, unlike other companies, they were very fast in terms of recruitment and the recruitment process was very clear. In retrospective, I can see why they were fast in hiring because it was a growing start-up and they needed to hire people. From the moment I applied, I was contacted very quickly and they gave me a task, I did the task at home, I came back, presented the task, and then shortly after they offered me the job. The communication during the interview process was very clear, I asked very specific questions and they were very honest. For example, I remember asking about the release process and they told me they worked with agile methodologies: they could release new features quickly and if it doesn’t work they would just revert it. And I really liked this kind of attitude.
Also, they do web analytics which is something that I wanted to do. At the time, I had a small website and I was writing on a tennis blog, so it was interesting for me to understand what’s behind websites and apps in general because it’s something that we use on a daily basis. I was really interested in getting more into the web technologies from that point of view
You’ve been with Decibel since 2016, how has your career evolved in the last 4 years?
Yeah. So, I joined as a Data Scientist and there wasn’t much hierarchy at the time. Then I was promoted to Lead Data Scientist because we scaled up the team and we had more people. This meant there was a need for somebody leading the team, deciding how technical projects will come life, what will be the priorities etc. Then I got promoted last year to Staff Data Scientist. I now manage the people in the Data Science team, as well as liaise with other departments like marketing, product or the account management team. In a way, I have much more of an overview of the projects and I do far less coding then used to do. So, it’s not a matter of building the features anymore, I’m now over-looking the features and someone else builds them. But that came with the scaling of the company.
How has what Decibel use data science for evolved as you’ve scaled as a team?
A lot! We collect large amounts of data here at Decibel, and when I started there was a huge appetite to progress our data science capabilities so that automated analysis became core to the product. When I joined, we’d just launched what we call ‘behaviours’, which are patterns within a digital experience. For example, when somebody is visiting a website and exhibiting some frustration, a behaviour will be identified. Now we score every single session with an algorithm which calculates the digital experience score, which is a comprehensive score for the experience. So, we move from analysing small sets of data to automatically tagging all of the data that we have. We have also moved from a simple Python script, which we weren’t even deployed, to a complex micro-service, which is deployed automatically. So, both in terms of technology and in terms of scalability, we’ve evolved a lot because our service can now scale and can now serve far more requests. The coding standards have also increased a lot – we’ve had brilliant people on board that have helped us through the journey.
Our code base is much bigger now. We provide six different services (besides the main one). Back then at the beginning, there was one (which wasn’t even a service, it was just an integration of a service from the processing team), so it was a little extension of the back end. We now have Data Engineers on board helping us building and maintaining new services. We’ve also transitioned to a cloud system – we are with AWS (which are very nice partners) – that we didn’t have at the time, which for us meant having to upgrade our skills and our knowledge about different services and technologies. It’s been an interesting journey!
Amazing. It’s impressive how you’ve scaled the team in the last 4 years. How have you gone about scaling your data science team?
We proceeded step by step. We started by hiring Junior Data Scientist. It was the first time I had to supervise someone, and I’d never managed anybody at the time. I wasn’t managing them directly (because we had our Engineering line manager) but it was the first time I had someone working with me that I had to look after. Shortly after, we brought on board our first Data Engineer. We had investments, so we used those resources to scale the team and there was a definite need for that – because the original members weren’t enough to supply all the features that we were required to do. And also, we had big plans to create all these scores and features that I mentioned before, and we needed more people to work on those.
The strategy was very simple, we had internal recruiters that were posting ads and trying to get as many people as possible, but we were mostly hiring Junior people.
After that, we started looking at it in a more refined way and also changed strategy by spending more time improving job descriptions and we started targeting specific people. Last year was a very important year, especially for me because I was the Hiring Manager. and I also had to actively search for people on LinkedIn myself (I basically worked as a part-time data recruiter temporarily). It definitely paid off because we managed to fill all our openings, including Senior positions. Hiring for Senior professionals is generally harder, not just because of their higher salary, but also because they’re much harder to find and you need to understand the right type of professionals you need to hire. As I said, we started to be more specific in our search because we are the type of data team which works on analytics and predictive models (we don’t just do big data). Looking for Data Engineers was especially difficult because we had to find the right ones, the ones that could work with our technology stack and they were able to come up with solutions to our needs. We don’t hire people to tell them what to do, we hire people so that they can tell us how to do things. This means I’m expecting a high level of seniority and a high level of expertise from the people I hire. It wasn’t easy because I was spending a lot of time interviewing a lot of people and rejecting candidates, not because they were bad or not experienced, but just because they weren’t the right type of people that we were looking for.
Okay, and so what were the main barriers that you faced whilst scaling?
So, the first thing is the availability of people. You have a lot of junior people available and lots of people who want to move into data science without previous working experience in data science. These types of candidates are either people coming straight out of university or people with different fields of expertise and they want to transition into data science. I see a lot of people with an academic background transitioning into data science because they love working on cutting edge topics and, at the same time, they want more permanent positions. Even for very experienced academics, it’s difficult to be hired as a senior data scientist straight away, because you need to master certain mechanisms that occur in enterprises and businesses that are very different from academia.
Budget is another barrier as you may not be able to offer a salary like, for example, Facebook. London is a very competitive market because very experienced people are very much in demand. Retaining people can be an issue too: we lost some people on the way, but I don’t think it was a major problem because it’s normal. Having two years/three years turnover is normal in this sector. When you are not able to offer a very high salary, you have to make sure you offer exciting topics to work on and you need to communicate this well in the process. The reputation of the company is also very important as you want people wanting to work with you. Often candidates did not even know Decibel existed.
Another problem we came across was candidates losing interest in the position after the first stage interview. The technical task must be well designed for two main reasons: the first reason is that it has to be appealing (so that candidates will do it) and the second one is that it has to span through all the areas of knowledge you require in a candidate.
How did you make sure you had the right tech test? Did you try different versions?
Yes, we experimented a little bit as we tried different versions of the task. We had an original task, which was good because it was very open, as you don’t want to constrain someone to do something that has a “yes or no” outcome. You want to give options because you want to see how candidates think, especially for a data scientist role. When we updated the task, as it became old and too simplistic, we decided to add more coding skills to it. After a few tries, I sat with the other Data Scientists (who were very, very helpful) and they came up with a mix of questions. The result was a task containing a variety of topics, statistics, data handling, machine learning, coding: in this way you’re not only rating candidates but you’re also rating their strong and weak points.
The other change was switching from an off-line task to do at home to a live task to perform on the spot: this change made sure that candidates would do the task, as we experienced a few cases in which candidates would disengage as soon as they received the task.
Nice. And what drove the data science team to scale? Was it your need, or the Founders?
It was a bit of a mix really. The business has always supported data science, they’ve always put data science at the core of the business. This is because Decibel works with web analytics and thus collects extremely large datasets. With so much data available, we offer unique value to our clients by doing the bulk of the heavy lifting with data science to remove the time sink of finding insights. We’ve built algorithms that provide meaningful information to the website owners so that they know exactly what to do in order to improve their customer experience and their websites. For this reason, data science is a need from the business point of view as it’s our core differentiation point in the market.
What came from us was direction, for example, we requested data engineers. Since we had a few data scientists, we could use someone who could build all the infrastructures that are needed for the data to be stored and accessed. The data engineers were super important for our job and they still are today. Basically, the business decided to allocate resources, we decided how to use them.
Got it. And so what are the biggest lessons you’ve learned from going on this journey and scaling the team?
When I had to look for people on LinkedIn, I thought that would have been really difficult as there are so many candidates out there and people may not like to be approached directly. I was actually surprised to receive many replies, even if people were not interested. I used a very honest and simple message, describing in a few words the job and asking if they were interested in having a chat, without being aggressive. It worked really well. A lesson is that we need to remember, we’re all humans, whether we communicate from LinkedIn or in person. Another very important aspect is to be clear with responsibilities: we all feel more part of a company if we’re given decisional power in what we do. When candidates hear ‘you’ll be working with this kind of technology and you’ll be responsible for this kind of service. I won’t be telling you how to do it, you’ll be telling me how you will do it.’, that’s when they are attracted to a job. If you send out this message across the reply you get is people wanting to be on board. Then you switch from me wanting you in the company, to you wanting to work with me. Which is a big switch.
That’s always the struggle as there are so many startups in London now!
Yeah! To add on to the barriers I mentioned before is that people didn’t even know we existed. It’s difficult to try and convince someone to work with you if they don’t even know who you are, what you do, or how you do it and so on.
Another lesson I’ve learned is that being human also means being social. Which means attending meetups, hosting meetups, trying to contribute to the community and doing hackathons. Exchanging knowledge really works because you attract the kind of people that love to learn these topics and love to share their knowledge, which, ultimately, it’s what you want for your team. You don’t just hire people, you hire people you want to work with.
Do you think you got a better response from people because you were a Data Scientist, rather than a recruiter?
Probably, yes. A technical person can normally dive deeper into the details of the job, whereas a recruiter has better soft skills to get the first contact with the candidates. Nevertheless, I believe that, especially in my field, there’s a stigma about recruiters. We are lucky enough to be in demand and therefore we get contacted by many recruiters and some of them are either too aggressive or too misleading. For example, some recruiters may advertise positions in the wrong way, just to attract more people, ending up in either wasting our time or getting you the wrong jobs. And because of these recruiters, all the recruiters get the reputation of wasting people’s time.
So, what advice would you give to companies that were in your position four years ago?
You first have to think about what you really need. You shouldn’t hire a Data Scientist just because it’s trendy. I heard stories of people being interviewed for Data Scientist positions in companies that don’t have a data team and they get asked the weirdest questions. So, if you don’t even know what you’re looking for, think again. Do you need a Data Scientist or a Data Analyst? Do you need a Data Engineer or will a Backend Engineer do just fine? Think about the data you have and think about what you want to do with this data and start from there. If you don’t know, which is fair enough, try and get as much information as possible. Attend meetups, try to go and meet the people and talk to the people before you start hiring. Because you may end up with people doing what they don’t want to be doing and they’ll be leaving you.
Not all data scientists are the same, some will be happy to get their hands dirty and build up some infrastructure when there’s none. Very often start-ups don’t have any infrastructure, so what you mostly need is a person who can look after several disciplines. In that case, you don’t need a person who’s only looking at algorithms.
Definitely! Also making sure that they want to be doing that sort of work and they’re not just forced into it as well.
In a market like London, people will leave so quickly!
So, if you go back in time, what would you do differently?
In terms of the hiring strategy, I would probably focus more on getting the right job description earlier in the process. Planning the future of the data team, in general, is something you need to do as soon as you start hiring. We are lucky now because we have a team of Data Engineers, but we have been needing those Engineers for a long time. If we had forecasted this earlier in the past, we would have probably focussed more on hiring Data Engineers. Another thing – I think we improved a lot the task we give in the technical interview, but it could have been addressed before. Because it took us quite some time to hire people.
And what advice would you give to Hiring Managers that are writing job specs for the first time?
I would say, be honest. There are lots of off-the-shelf job descriptions, like the ‘typical Data Scientist’, which ask for a million things – don’t use them! Just ask for the things you’ll be using. So, if you use Python, don’t add R or Scala. If you use Scala, write down Scala. Are you working with cloud technologies? No? Then don’t write it – maybe add it as a preferred type of skill, but not an essential one. You don’t want to scare people away. Some people will apply nevertheless because they have the confidence, but other people will not apply because they’ll be either not confident (junior positions) or not interested (senior positions).
And do you think by having a job description with loads of requirements put’s candidates off appyling?
It could be the case. Broadly speaking there are normally two types of candidates, junior and senior. A junior would usually be scared, “Oh s**t, I don’t know this, I don’t know that, I’m never going to learn this in time for the interview”. Senior candidates, on the other hand, may think that the company doesn’t know what they want. Job descriptions with many requirements may be relevant if you’re hiring for a very big corporation and you may be working with very different systems. However, if you require some technologies which can be conflicting, like Google Cloud Platform and AWS, are you really sure you’re looking for a Data Engineer or are you looking for a DevOps Engineer instead? Don’t think that, just because your job description lists many cutting edge technologies, that makes it more appealing. I think that you can lose candidates because of poor job descriptions or attracting the wrong type of engineers.
Amazing. Do you have any final words you’d like to add?
For companies looking to scale. I would say try to be as human as possible and do not contact people with template messages, make the effort to personalise the message. Try to participate in meetups and to contribute to the community, it’s good to have visibility from that point of view. Also, try out internal hackathons or internal events and encourage participation from people from all departments, not just technical people. Everyone can contribute to creating a more data-centric culture within their organisation.
Nice. To finish off, I always ask these two questions. What’s the best piece of career advice you’ve ever been given?
So, a while ago, I was having a chat with a friend of mine, who is a Software Engineer, and I was stating that I didn’t really need to look for a job as I was in a comfortable, stress-free job. He then asked: “Are you feeling very comfortable? Maybe it’s time to change”. That got me thinking at the time and I realised that, when you feel too comfortable, you’re probably not in the right job – you should always have some challenge. Even if a job is hard, if you like it, you’ll overcome the difficulties. Always be honest with yourself and ask yourself: “Am I too comfortable?”, because if that’s the case, you’re not going to be happy for long.
Amazing. And what’s one book that you’d recommend?
There’s a lot of books! But I would say, one that I really liked is ‘Factfulness’ by Hans Rosling. He was definitely a good inspiration for me to get into data, but not just data per se because he used to talk a lot about Social Sciences. It’s a great example of how you can use data to understand the world. ‘Factfulness’ is about showing that perhaps things are not as bad as we picture them. It also gives you more of a logical approach to the world. It’s a good book even if you’re not into data science or you don’t have a technical background.