Tom, thanks for sitting down with me to discuss your journey at HeadBox today. Do you want to start with an introduction to yourself?
I’m Lead Data Scientist here at HeadBox and I joined about 14 months ago where I started in sales and then moved across to join the Product and Tech team. HeadBox’s mission is to reinvent the events industry through technology, we focus on both guests and hosts and try to optimise their experiences.
I’m a Biochemist by education and studied computational biochemistry in my master’s at uni. I thought about doing a PhD but was feeling a bit stifled by the research world and a little frustrated that it’s very slow-moving and you don’t really see the output in what you’re doing for years afterwards.
During my final year of university, I was also involved with a few start-ups. One was building an app to track intramural sport and then pivoted into a company that would track sports matches and automatically generate highlight reels. At the time, I wasn’t particularly technical, but I got bitten by the start-up bug and coming out of that slow research environment made me think, ‘wow, you can make something this quickly!’. Those ventures didn’t really work out but after those experiences, I wanted to go and do something at a startup where you can have a tangible impact immediately.
So, after joining HeadBox in Sales, I quite quickly realised that the interesting and exciting problems were being solved in the tech team, and there are some really smart and impressive people in that team here. The CTO is fantastic and gave me a shot to switch across. I started as a Frontend Dev, did a bit full-stack work and then found a few problems that could be solved by data, and it’s built from there.
Was a career in data science always the plan?
Not at all! I think if you had asked me coming out of university if I wanted to go into data science, I would have said ‘definitely not!’. But at the time, I didn’t quite realise how having that scientific grounding can really help you in an industry like this one, where you can identify problems and frame them differently, based on the data that we have as a company. There are so many complex problems in this industry that really require us to be creative and learn quickly, which means the work is constantly varying and challenging.
What is it about data science that interests you so much?
It’s probably a couple of things, I would say. Particularly in a company of this size, there aren’t many other roles where you would get to touch as many different parts of the business and take such a holistic view of the strategy, the customer experience and the business’ outlook. Everything you might do in a start-up, a Data Scientist can be involved in it, and that requires such a breadth of skills that you’re always going to be learning. I really love that aspect of it. The second thing is being able to frame problems with cutting edge technology and research. It’s becoming so accessible that you can set a meeting and discuss a solution, that 18 months ago, it would take a year to implement! It’s an exciting time to be in data science.
We’re actually hosting a Fast.ai study group here in the office and it’s just great. We’ve got a great group from the office involved as well as Data Scientists from around London. I think the possibility of seeing what other people are doing and it makes you think ‘wow, they built a model that can do that, and there are so many areas that our business that can really benefit from something similar’.
I had been through it briefly in the past but was primarily focussed on Keras. I think that group setting just demonstrates the benefit of getting people in the same room, learning together and pooling knowledge. I’ve found fast ai to be so much easier to use as a library.
You were the trailblazer for data science at Headbox. How did it all start?
I guess it started off when I was working as an engineer on our new Brief Builder, which replaced our search functionality on our homepage. We wanted to move slightly away from your bog-standard search and more towards something more guided, like ‘tell us about your event and we’ll suggest to you the best venues for it’, and create more of an organic flow, similar to how we might have a face to face conversation about it. As we were going through that flow, it just became really clear that there are loads of points where we can start injecting little bits of data as guidance for the users, in that sense we are really trying to mirror what a human events expert might offer. Things like an ‘event like this would normally cost X much’ or ‘if you held it on a different day, it might cost less’ or ‘you might want to look in X area for more results’ and things like that. So, as part of my work on that project, I built a few models to start to predict event pricing for different venues (which is really tricky in this industry). From that point, we saw that there’s real value in being able to offer on-demand insights to our customers and from there it has really grown to be part of the business’ strategy.
How much has data science evolved since you’ve started this initiative?
I think a lot! In terms of our processes, we’re obviously still learning all the time. I’m really inexperienced still and the team is still really young. I would say the main area that’s evolved is how data is thought about by the rest of the business. I think that was a big hurdle in the first three to six months of talking about data. Everyone sort of nods along, but until you really deliver value and start making their processes easier, it can be quite difficult for people to picture the vision. I think we’re starting to get there in terms of getting real buy-in from the rest of the company and the CEO really loves it – which is exciting.
What’s the plan for data science at Headbox?
That’s the big question! I think that there are three things we want to deliver. There are lots of companies that are really good examples of using machine learning and data science in industry, but in terms of the events industry, it’s virgin territory for anything like this. The marketplace model has been laid out there and it doesn’t necessarily work that well. One of the key things that data science and machine learning can do is, in effect, redefine that ecosystem and make it a more valuable place for both the guest and the host. The way we’ve been trying to frame that is around three key themes; personalisation, guidance and feedback.
Feedback is where we can start to think about, particularly for hosts, what data can we gather about their behaviour on our platform in terms of who they’re messaging, what they’re putting in their messages and how their messages are suited to the guests. This blends in a little bit of personalisation already and if we can feed that back to them, the Headbox ecosystem becomes a valuable place to be for them and effectively makes them better at their jobs.
For personalisation, it’s all about the user experience when you come to Headbox. That can centre a lot on displaying events that hosts want to host or venues that are the perfect match for a guest. Recommendations are actually a really interesting problem because there are so many nuances in the events industry. If you look at just London as an example, there are these cultural hubs that each have their own stylistic identity. There are different locations which appear far apart, but they’re both near Central Line stops and that works well for a guest. I think really that venue finding piece is something that people come to an expert Account Manager to get and we would like to be able to model that experience as an automated one, so as many people as possible can have it.
The last aspect is guidance. On the host side, I think it’s very similar to feedback, but we want them to gain more from being part of our ecosystem. So, when they’re filling in an events proposal, what can we deliver to them to make that proposal better? And that can be fed in from what we know about the guests and the history of all events on HeadBox. If you’re part of Headbox, you get all this feedback and you get to learn about the events industry yourself. I think that the guidance piece is really crucial, we ultimately want to help hosts get the most from their event spaces.
It’s a similar thing on the guest side. When you have a dedicated person as your event planner what you’re expecting is to have your handheld through that process. Things like price negotiation or adding careering to the event and finding somewhere fun that will suit your guests after the event. All those little guidance pieces are things we’re not trying to make the individual choice of the guest away, but we suggest to them things that we think they’ll really like and might have never thought about themselves.
What are the biggest lessons you’ve learned?
I think we have really learnt the value of bringing stakeholders on board from an early stage if you are looking to change a process. In our case, that meant helping people understand that we were looking to make their roles more interesting and take some of the mundane and repetitive work away. Ultimately that means they can focus on building great client relationships more, which is ideal for everyone as we get better quality feedback to the tech team.
How receptive were the different teams when you first started to propose using data science?
It varied! But I think that’s fair enough because at one point it probably looked like we were selling snake oil. It was a tricky balance between being creative and telling a story about where we could be, but then also being realistic about how long it would take for us to get there. We had to explain that it might slow you down for the next six months because you’ve got to input all these extra fields, but those fields will turn out to be useful in the data sets. Those sorts of things can be challenging, but now I think that we’re starting to deliver on things that we were promising.
Do you think your experience in sales has helped that data storytelling piece?
Yeah, without a doubt. We’ve actually had the Junior Data Scientist in the team go to a couple of the sales training sessions. The ability to know how to structure your narrative and focus on the needs and interests of your audience was very valuable. We do a Friday afternoon demo session to the whole company here and if I’m demonstrating some research or a solution, it’s very similar to how we might demo Headbox as a salesperson.
What advice would you give someone who wants to get into data science that might have a similar background to you?
I think I would say you need to have a real passion for data and telling stories through data – it’s a bit corny, but I think it’s true. Do side projects but do it with datasets about things that you really like. When I was just learning and doing data science work in my own time, I did it with stuff that was interesting to me. I think that’s really valuable because it can feel like a long slog sometimes when you’re first learning. Once you find work you’re passionate about you start to develop those data storytelling skills, which I think are really undervalued in Data Science. If you learn how to do it on stuff you’re passionate about, then hopefully you can work somewhere where you’re passionate about the datasets and do well.
What non-tech skills do you think are really important for someone that wants to get into data science?
The classic ones that you see a lot are really important. So, the ability to explain a complex concept to people who are non-technical, in a non-condescending way, is super important. But also being able to take a simple problem and break it down into chunks and say ‘okay, you describe the problem to me in a sentence and I can break it into modules of data science work, and then I can break those down into more theoretical modules’, that’s a really powerful skill.
I think, particularly for Data Scientists in small companies, you need to be creative about what you’re doing. You don’t work at an Amazon with all the data on anything you can imagine. So, it’s thinking about where your best data is. You have to work closely with Product Designers and Product Managers to think conceptually about how you could deliver your findings in the best way as an insight. That’s really important and I think it has helped a lot in getting us off the ground as a team.
Finally, on a similar note (particularly as a small team) is the ability to multitask and to be a bit of a Data Scientist, a bit of a Machine Learning Engineer and a bit of a Business Intelligence Analyst. It’s all well and good us mining some data in a silo, but if I’m no good at building a really coherent dashboard for the relevant team, then that work has gone to waste. Understanding the full process, from being creative about what you’re going to do, finding the data and having the skills to do it is important, but then also the skills required to make those insights as valuable as possible, as quickly as possible.
What is one thing you wish you’d know at the start of your career?
I think when I left university I was a little fed up with the pace and environment of research-based work. So just knowing that environments can exist where you can do really interesting theoretical work, but also see it delivering value fast.
What is the best use of data science you’ve seen?
The funniest one I’ve seen is a really good article by a guy who built a classifier over Donald Trump’s tweets and it was basically to say whether Trump or one of his aides had written it.
What’s your best piece of career advice?
When I first started on the tech team here, the CTO just said to me that you should just eat, sleep, drink, code and gather as much information as you can. I think that was pretty valuable at the time. If you want to get into tech, just spend a considerable amount of time just thinking about building models or working with datasets as much as you can. I think if you do something you’re interested in it is a lot easier.
What do you think data science or machine learning will have the biggest impact on in the next decade?
I guess generally speaking, just the way people work. I hope that it will make people’s jobs more fulfilling. But I think that’s the big challenge. I also hope that it’s climate change – that’s probably what it needs to be, doesn’t it?
What’s one book that you would most recommend?
A really good book that I read recently is called ‘The Madhouse Effect’ by Michael E. Mann and Tom Toles. It’s basically about US politics and climate change denial, which is really interesting – although not particularly data related!