Alinea Meets Blair Young - Data Scientist at Triptease

HB: Morning Blair, thanks for taking the time to chat with me today. Could you start by giving us an intro into yourself?

BY: Okay, sounds good. I'm Blair, and I've been wrestling with data science, machine learning and deep learning for about four years, and I currently work at Triptease.

Triptease exists to stop technology from being a barrier to fair competition for hotels. We're on a mission to disrupt the intermediary booking sites and establish the direct channel as the dominant method of hotel distribution. In essence, we're trying to give power back to hoteliers through data-driven tools.

HB: And how did you get into Data Science?

BY: I studied genetics at Queen's University Belfast. When I was there, I noticed a lot of the researchers were crying out for scalable biological research, and that's when I started looking into doing a Master's in Computational Biology - which is a mixture of biomedical research, statistics and coding. When I finished that, I basically Googled my skill set- scientific methodology, statistics, coding, documentation and communication, to which the word “data science” popped up. I didn't really know back then what data science was or how much it would blow up in the past four or five years, but I thought building something that drove insights through data seemed interesting and was highly employable.

Before I could jump into it, I had to upskill myself before I could even approach the field effectively. I spent the first six months part-time teaching myself the absolute basics. That good old chestnut of Andrew Ng's machine learning course, which I probably wouldn't even recommend to-do now., as I think it's a bit stale and too math-heavy to get going- especially when it's taught in MATLAB. I was also learning Pandas and transitioning from R (which would have been the de facto language in life sciences) to Python. At the same time, I was doing a few odd jobs trying to save a bit of money to move to London in search of a job. It took about two months of full-time searching to land my first data science role. Full-time searching involved hounding anyone and everyone while still upskilling and learning from unsuccessful interviews (always remember to ask for interview feedback, you work hard for interviews, you deserve to know how to improve the process!)

HB: After having a couple of Data Scientist positions, what made you join Triptease?

BY: I left my previous job, and I'm pretty sure every Data Scientist has felt this before - I felt that I wasn't getting enough of my work in production. It’s kind of a kicker when you put so much effort into a piece of work just for it to get dusty on the shelf. I think that it was quite an acute pain point over the last couple of years for the field.

So, by that point, I was itching to move somewhere to get this experience. Enter Triptease. They reached out to me stating they’re starting their new data science wing of the company. They wanted to drive more business through large amounts of data they've been collecting over the past couple of years. Now, at the same time, I'd been to enough interviews where I realised when companies say a 'large amount of data', it turns out to be 300 rows on Excel, but that wasn't the case of Triptease at all. I think at that point they were collecting and curating about 15 million events per day. They really doubled down on the collecting and curating before they got their hands-on data science, which I think is probably the best thing to be doing really. When I came to Triptease, there was so much high-quality data, interesting problems and great people. I was blown away by how many people are so passionate about each part of the product.

HB: That's amazing. You mentioned that they didn't really have a data science function before you started, how has that evolved? And how have Triptease become a data-driven business? 

BY: The core of Triptease's product is the parity data- our ability to see how a hotel’s rates are appearing around the web, we use this data to help hotels improve their online performance. To begin with, we were just using that data to provide a price comparison tool on the hotel website. Then we realised how powerful the data behind that tool would be to hoteliers themselves, so we built a platform for them to review and act upon that price comparison data.

That was the start of us realising the value inherent in this data and joining it together in ways that are incredibly useful to a hotel. Technology across the hotel industry is very fragmented, so the fact that we’re able to bring together that central pricing data along with user behaviour data proved very powerful and well, unique.

Our real differentiator is the fact that we can join up all these data sources that would be quite difficult for an individual hotelier to access, and then crucially provide the tools for them to act on that data. That could be by changing their website content based on certain visitor types, changing their bidding strategy on metasearch adverts based on their pricing status, or spending more to attract a certain kind of user through display advertising. We know how valuable our data is, so we’ve made a conscious effort to invest in data across the business.

HB: And what are Triptease currently doing with data science?

BY: We're big fans of incrementality and being able to prove this relies heavily on A/B testing. We’ve invested a significant effort into A/B testing, everything from upskilling teams in understanding what a test is to developing streamlined frameworks for deploying experiments.

We've also dabbled in multi-armed bandits which I led last year- It's kind of like a turbocharged A/B test that leans on reinforcement learning. They're quite useful actually, traditional A/B testing can be quite slow as you have two separate phases, exploration and exploitation. You must first explore, which is basically running the test then once the test is finished, then you can deploy the better variant, that's when you start reaping your reward- the exploitation phase. With the multi-armed bandits, you explore and exploit at the same time. If you find one of the variants perform better, the model shifts more of the traffic towards that variant allowing the business to make more money and explore in parallel, so it's a win-win!

Another application of data science that we're working on now is building an end-to-end machine learning pipeline that drives our programmatic and metasearch advertising products. We essentially collect, curate and analyse hundreds of millions of advert auction bid logs in order to automate and streamline our bidding strategies for displaying adverts to the right audience at the best time.

It’s an understatement to say that the advertising data landscape fluctuates. A significant part of our work is understanding the bid landscape at any point in time. From this, it's important to know when to re-train models and identify shifts in day to day, or even hour to hour trends. We focus heavily on monitoring these shifts in data, whether that's encoding intuitions about data in a schema or building models to detect general changes in the landscape.

Another exciting application on machine learning within Triptease involves modelling visitor behaviour on hotel websites. By mapping onsite behaviour to intent to book, we segment and optimise the experience for each visitor. This segmentation based on propensity to book is a game-changer across our visitor facing products.

One of the most important ongoing projects is improving data literacy across the company. By educating through internal talks and pairing, we have empowered a number of non-data scientists to think about how data could help them in their daily workflow. Some of the teams have actually gone off and realised "Oh, we can apply this data science technique to solve this problem". They'll go off and try something and every so often will have a check-in with the data science team for advice. A great example of this is within our integrations team. Every hotel website is different and leads to more manual work than desired when integrating our platform. The team has built a model to identify specific types of web pages across multiple languages. It’s a super cool little model, it’s very lightweight and deployed browser side. I think their approach and execution was fantastic and such a great addition to their workflow. It’s easy to only think about externally facing machine learning tools, but sometimes internal tools are just as valuable.

HB: Yeah, I completely agree. Most of the time, it's usually only larger businesses that invest in internal data science tools because of their apparent lack of return on investment.

You mentioned about up-skilling members of the business in data science. Is it just technical people, or do you teach everyone?

BY: There’s no criteria to be met; it could be from any discipline within the company, engineers, customer success, design, product managers, anyone. Each team is fully autonomous, and we double down on pairing quite a bit, from this we definitely cross-pollinate skills and angles at which to tackle problems.

It's important to note that this culture doesn't just reside within product and engineering. In order to keep everyone in the company up to speed with our data-driven products or decisions, we document our experiments, mini whitepapers, analyses and pipeline data flows on Notion (big shout out to them!). A lot of care goes into the language used so that anyone at any level can understand the reasoning behind all our efforts.

HB: Sounds like a lot of work has gone on for the last 18 months! What's the plan now for the Data Science team?

BY: Currently, quite a few of the machine learning products are isolated. We intend to link up the data streams and model interactions to create stronger signals for our clients. And like I said, we want to have this data-driven culture throughout the company. This encompasses more data evangelism and improving data literacy across the company.

HB: Amazing! It sounds like you have an exciting journey ahead of you.

Moving back a little to your career side of things. You've always worked in startups, why is that?

BY: Its cliché, but it's the mixture of the fast-paced environment, full autonomy within teams, seeing the work you're doing going live and making a difference and the fact you're not just a cog in a wheel. Also, the greenfield space - which is the place where I can live and experiment and really let my hair down. So yeah, classic cliché of working in a startup!

HB: Haha nice! You didn't purposely get into data science, like a lot of people. What advice would you have for someone that's potentially looking to get in data science that's either unsure or doesn't have the 'typical' background?

BY: Previously I regretted not going down the traditional engineering route of computer science simply because I'm working with engineers every day. Then I realised it's more around the way of thinking, that scientific process.

I can imagine the market has definitely shifted since back in my day - and it's mad to think that 'back in my day' was only four years ago. So much has changed in data science over that time. Previously companies wanted data science because it was trendy, to a point where it was becoming a bit of a wild west. This idea of want rather than need has trickled down to over the last couple of years which has led to a serious over-saturation of junior data scientists looking for roles making it very difficult to stick out in the market.

HB: Yeah, that's a really big thing. You've got so many courses now that are data science/machine learning specific from universities like UCL, Imperial and Edinburgh University. These courses are now churning out huge amounts of students all looking for their first junior role.

BY: I definitely think that universities and boot camps have seen the dollar signs just as much as the need for data-related jobs. It feels like they are getting as many people through their doors as possible. I think it’s much harder now for someone to land their first role in comparison to my experience. However, I can give you some tips- I just hounded every single company I knew. I don't mean cover letters. I mean going on LinkedIn and finding people who looked like they would work with me in a data science or engineering role and just messaged them directly and selling my wares - "This is what I've been learning to do", "this is what I want to do for your company" and "this is how I can improve your bottom line". It's a networking game, so just get out there. I wasn't afraid to pester people and you shouldn’t either.

HB: Totally agree. It shows much more than just a personality-less application. For example, you can now just one-click apply to roles on LinkedIn.

BY: Oh, definitely. I think it's now so difficult for people to stick out in general now. You got to put in the effort.

One of the other things I've noticed over time is to never be afraid to tell interviewers the horror stories or blunders that you've experienced or made yourself. I think a massive differentiator between junior, mid and senior levels is the ability to have scepticism.

Always worrying about how and where your data is growing and really understanding that it's not a perfect world! You can sit there and do your Kaggle competitions and you can create a portfolio the length of your arm, but all that means very little if you can’t get the data in a reliable way.

A lot of juniors will focus on the model aspect of data science and understanding the different techniques, but that's so far downstream from anything that will drive a business. You got to know the origin of the stream. I’d reckon if you bring up these thoughts during an interview, even if it’s just questions for the company, it will put you head and shoulders above other candidates.

HB: Very wise words. What do you think the biggest things you learned early on in your career that have helped you get to where you are now?

BY: No one cares about your coefficients. Always tailor to your audience and find out what they want to get out of that meeting. Don't bury the essentials in what you think is important; just get the information they want to hear out there. Basically, tailoring to your audiences is a huge one. There's no point in boasting about fancy feature engineering or finding outliers. No one cares, just show them the info that they want and everyone's happy.

Don't spend your whole time studying low-level nitty-gritty techniques. Get out there and build something, even if it's not for a portfolio. I think you learn more when you're on the path to completing something, rather than learning it for the sake of learning. Especially for someone like me, my head's like a sieve!

The other thing is don't try and force machine learning on every single thing you do. Chances are, you can get a really good starting base from just basic analysis - like a pivot table. There's no point in using something fancy when you can whack something simple out in a number of minutes.

HB: Very insightful, thanks for that Blair.

 Moving on to our final quickfire questions. What is one thing you wish you'd known at the start of your career?

BY: Understanding your domain is more valuable than data science techniques. The first couple of things you should be doing is talking to different teams, seeing what their problems are and then going and investing a good amount of time wherever the data lives. Look at the columns, ask questions on how things relate, what does it truly mean for an event? Chances are what you think is happening isn’t actually the case. Ambiguity is dangerous.

HB: And what is the best use of data science/machine learning you've seen?

I like the small interactions that affect me on a day to day basis, like Google Voice Assistant. Everything from putting on timers for my dinner, to reading me the news when I wake up, I think it's seriously cool—also, Instagram ads (which is where my whole wardrobe comes from!). The targeting is very impressive. Even the smaller applications like when your phone knows what sort of apps you're likely to open up. People don't really think about those things as being big data science projects and they're not crazy big, but those seconds saved really add up when you think how many users it touches every day.

HB: I’m a big fan of my Alexa in the morning!

What's the best piece of career advice you've ever been given?

BY: I can't stress this enough - tailor to your audience. Like I said, no one cares about the nitty-gritty stuff. Just tell them what they want to know.

HB: And what do you think data science/machine learning will have the biggest impact on in the next decade?

BY: I've been reading up on precision agriculture recently. In short, we’re going to struggle with food demands in the not so distant future. Our population is expected to hit 10 billion by 2050 and to feed that amount of people, we have to increase our agricultural production by 70%. We're stretched enough as it is, and who knows how many chemicals are keeping us afloat. Precision agriculture utilises sensors to give us topological insights into field composition. This’ll allow us to take advantage of every inch of space to grow crops.

HB: Finally, recommend one book that's changed your life?

BY: Yeah, so it's called 'Aliens: Science Asks: Is There Anyone Out There?'. It's a collection from different scientific and arts disciplines detailing if there's life outside Earth. I went into it like 'aliens one hundred percent exist', but whenever they actually lay out the facts, it makes you feel very small and very lonely. It's easily the best book I've ever read, and if I ever get someone a book as a present, it's that one.

HB: Amazing - I'll definitely have to check that out!

Blair, thank you again for taking the time to chat with me. It’s been great to learn more about the awesome work you’ve been doing at Triptease.

You can learn more about Triptease here, and about Blair's data science meetup here.