Mahana, thank you so much for your time today. It’s lovely to meet you. It would be great if you could start by introducing yourself.
Of course. So, I’m originally from New Zealand and moved to the UK in 2007 to do a PhD in Maths, as well as become a professional athlete. I wanted to train to do Judo in the London Olympics and do a PhD all at the same time. A bit of a weird story, but I decided on a university where I could do full time Judo, rather than where I could do any particular maths topic. So, I went to the University of Bath, which has really good sports facilities. I went there and said to the Maths department “give me any maths PhD topic and I’ll do that” while I do my Judo. That was super cool as I was training full time and they have a good international Judo team there, along with members of the British team, all while doing my PhD at the same time.
I’d always thought I would be a Maths Professor as I love maths, pure maths – the more abstract the better! But during my PhD I realised that there were only about three people in the world that knew what I was on about. Not because they’re not smart enough, but because it’s such an obscure topic and it didn’t have much real-world impact. I was also, for some weird reason, sitting in the computer science department as the maths department at Bath is more physics and the computer science department is more maths. I was with a bunch of computer scientists who were working on different things and I felt quite isolated in the lab and didn’t have anything to talk to the people about – which meant I was spending a lot more time with the Judo people.
In 2011, when I was finishing my PhD, I was also retiring from my Judo career due to injuries. I then thought “okay, I need to get a job – any sort of job!” and having a PhD in Maths from Bath, I had quite a few recruiters contact me with a lot of options (although I had no idea what I wanted to do!). It wasn’t anything I’d ever really thought about, as I always thought I was going to be a Maths Professor. It was really weird for me as I was so far from home, but I knew I wanted to stay in the UK as I had loads of friends over here.
I was offered a job at Ocado as a Software Engineer which was great as I’d done a bit of coding, so I thought ‘Great! I’m going to be a Software Engineer’. In my first week there, I met a guy called Vince Darley (who actually works here at Deliveroo now as VP Growth) and he was leading up what was the equivalent to the Data Science team. He heard that I had a PhD in Maths and asked if I’d like to join the team, so I thought why not, it sounds like fun!
A lot of people in that team had a similar background and I remember it wasn’t called data science at the time, it was called Algorithmic Software Engineering, or something on those lines. Vince was my Manager there for a little while, but then moved on to King.com and then as I had a bunch of other Managers there, we quickly changed the team’s name to Data Science, as it was the cool thing and that was what we were doing.
‘Data Science’ is a relatively new term, isn’t it?
Yeah! It kind of came up after I actually started doing that type of work and it was super cool. I worked on everything consumer facing there, so things like personalisation. I must have been there for about 4 years and by the end of my time there, I was in a semi-management role as well as doing day-to-day data science work. So that was really the start of how I got into Data Science!
I then wanted to get a job in London (as Ocado is in Hatfield, which is north of London) as I’d been with them a while. So I moved to the Guardian which was really, really cool as I did loads of interesting work there. I was Head of Data Science and they were going through a phase of trying to get into the digital age and work out how they can generate money going forward which wasn’t through selling print newspapers. I was lucky enough to work on a load of really interesting projects there, which included presenting to their Board and the Scott Trust (who are responsible for the Guardian) and helping define their top-level metrics strategy.
I was there for a couple of years which was really interesting, but I wanted to get back into more of the tech space and so I made the move to Deliveroo, which was just over two years ago now! Since then, I have had so many different roles at Deliveroo, jumping all over the place. I’ve worked in pretty much every business area and I had a baby in the middle of it all, which had me on maternity leave for 6 months. I’ve now got a one year old, so at the moment I’m still figuring out the balance of being a mum and working in a very fast paced environment!
Apart from Deliveroo being in the tech space, what else made you want to join the company?
I think mainly it’s growth. I was at the Guardian, which was a shrinking company that had redundancies as the newspaper industry has a very unknown future. Deliveroo has been one of the fastest growing companies in Europe – one year it was the fastest growing and the next year it was the second fastest growing (which is impressive for how big we are).
Another reason is that Vince (my old boss at Ocado) reached out to me and said “Hey, I’m now VP of Growth at Deliveroo – are you interested in coming to work for us?”. So, I had a coffee with him and thought ‘yes, very much so!’. All of the stuff we’re doing here at Deliveroo is really cool and forward thinking, especially about the future of food, not just food delivery!
Awesome! In terms of your role now, you mentioned you’ve moved around the company quite a lot – what has your career been like at Deliveroo so far?
I think when I joined, there was a lot less structure than there is now. I initially joined the Growth team to do email personalisation. But then, for various reasons, I moved to the Consumer team and then to the Network team, Pricing and then back to Growth. My scope has grown as well in terms of the number of people I manage, initially I wasn’t managing anyone and now I have about 35 people across the various teams I run and 5 direct reports. We’re also looking to significantly increase the size of our Data Science team in the next year.
What are Deliveroo currently doing with Data Science at the moment?
What aren’t we doing!
We recently merged our BI and Data Science functions because we had a bit of a weird set up. Our Data Science team was serving our Tech teams and our BI team were serving everything outside of tech. Another thing is these two teams have very different skillsets. The Data Science team was really good at building algorithms and running experiments and the BI team (although they did a little bit of that) built really good data solutions – so giving our business partners the data they need, when they need it. This might be a dashboard, or maybe sitting in a meeting and getting a number at the right time, which isn’t something the old data science team was good at! For example, if our consumer product team needed a dashboard, we didn’t really have the skillset to build that and we didn’t have anyone from the BI team in the Data Science team that could do that. We’ve now combined both of these teams and are deployed across the whole of Deliveroo, which is really cool.
We have three main disciplines here – Algorithms, Inference and Analytics. Algorithms is where we do production machine learning. This is things like optimising our network, so we assign the right rider to the right order and that we give our customers, riders and restaurants and accurate predictions of what’s going to happen at any given time – which of course is super important. They also do rider fees, which is about asking ‘how do we fairly reward riders for the distance of each drop?’, which is super interesting. On the consumer side, how can we give our consumers a really interesting and relevant offering when they’re in our app, so if they hate fast food then they won’t see fast food restaurants, or if they do like fast food then they will see fast food restaurants etc.
How are you finding it now overseeing the data science function? Are you still a bit hands on, or have you fully moved away from that?
I still very much read work produced by the team, but I certainly don’t write code myself anymore unfortunately. Even though I love writing code, I just don’t get the time to do it anymore. I help steer projects, where I’ve done a lot of these things many times before, so I can help spot problems before they arise. A lot of it is ‘where should we be spending our resources?’ or ‘where should we be putting people to have the most business impact?’ and of course helping communicate work and sense checking and quality checking.
Great. I wanted to go back to your earlier point about how you got into Data Science, as it seemed a little unplanned. A lot of people go through a BSc, MSc and even PhD wanted to get into data science afterwards, what advice would you have for that?
This is something I find super interesting and am very passionate about. The thing I’ve actually seen that works the best to get into data science is to not directly try to get into data science. What I did was to get a job that wasn’t called data science and then the term “data science” became common, and then convinced my employer to change my job title to Data Scientist. I’ve seen things like this happen so many times, most of the people I know who are now in Data Science got that job title via a sideways move in their organisation (from engineer, analyst, merchandising or marketing).
Some of the best career advice I have ever had is to start doing the job you want to have. If you get a job doing something somewhere and you see an opportunity for this organisation to become better through the use of data science, just do that in your spare time (whilst doing your day job really well) and make a case to be called a Data Scientist. This is the most effective way to get into data science.
I’m aware there’s a lot of bootcamps and university degrees nowadays for data science, but I haven’t always had the best experience with these, because I think the skills they teach are not necessarily the right skills to allow you to perform on the job.
Are they usually less practical and more theoretical then?
Yeah, I guess it’s a lot about the intricate details of a model, or some amount of coding and how to train a model to be really precise – which is only about 5% of the day job. The hardest thing which I never see people be able to do really well (and when I interview people, I consistently see this lacking) is translating a business problem into a data science problem.
For example, in our consumer team, if we’re trying to algorithmically optimise what consumers see to be the most relevant for them, what kind of data science problem is that? What are we trying to optimise for? Are we looking to maximise or minimise some quantity? And what is that quantity? And how can we do that offline by building a model and deploying it? Or should we be doing this online in some sort of reinforcement learning model? That whole problem framing doesn’t seem to be thought about or taught in the course that I’ve seen so far.
Where do you see the biggest challenges people face in their data science career?
The challenges are all around delivering impact, especially as data science is a very nebulas field which isn’t really defined yet. It’s very easy to build a really fancy neural network, but to not have any business value. The biggest challenge I consistently see in data science is doing impactful and valuable work that will improve the organisation’s bottom line and things they care about.
Do you think there’s a way that this can be taught?
I think one thing that’s often missing is experimentation, which is one thing we’re really big on here. Whenever we do anything and we have to decide on whether we roll this thing out or not, we run an experiment and we have some very complex experiment designs here, not just a simple A/B test. I don’t think, to the best of my knowledge, that experimentation is taught in these bootcamps and university courses, although maybe it is in some – I don’t know all of them!
I think experimentation is super important, for example; power calculations, picking primary a metric, making sure that the primary metric is actually aligned to what the business wants to change. All of these things for me are very basic due to the way we do things here at Deliveroo, but are not that common from what I’ve seen at other companies.
Another challenge is framing a business problem into a data science problem – what are you optimising for? What are the incremental first steps towards solving this business problem?
Is this something that you learnt at university, or did you learn it once you got into the data science industry?
I think I picked it up in industry. For me, I’m always constantly reminding myself to think about impact and incrementalism. I ask myself ‘what does the business actually want to achieve here and is the work I’m doing helping the business achieve this?’ and ‘Am I doing the smallest next step to get towards that rather than something that might take two years to assess?’
Have you got any advice for anyone that’s new to managing data science teams to implement this into a business that doesn’t already do this?
I would start really, really small. Start with planning experiments very well and do the power calculation upfront. One thing a lot of companies do is they just start an experiment and continuously peek at results, which invalids your experiment design as you’re likely to get a false positive/false negative along the way if you continuously look at the results. Start with a very simple but rigorous A/B standard hypothesis testing methodology to evaluate what you’re doing. Also, for any piece of work, start with the business problem and turn it into a data science problem rather than coming up with some sort of data science model and trying to segway it into some sort of business problem. Always start with the business use case and then you do the data science.
So, when you’re hiring for your team at Deliveroo, what sort of background and experience do you look for?
For me personally (and this is a bit controversial!), I don’t really care about academic qualifications. I have a PhD in Maths, and I don’t think it’s at all necessary. I think what is necessary more than any background is a focus on impact and being pragmatic and incremental about getting stuff done, but also very strong numeracy ability and logically thinking about very abstract things. One thing we do in our interviews is we talk a candidate through a very complex interconnected system and ask them to reason about it.
For example, if we have all these riders delivering things, what would happen if X were to go wrong – what would be the knock-on consequence for our customers who are eating the food, as well as the riders and the restaurants? We would get them to try reason about this and ask them ‘how would we assess the knock-on consequences and run an experiment if we wanted to take a certain action?’
When you’re interviewing candidates, what soft skills do you see as massively important?
Communication is so important! I’ve seen people with amazing technical skills but can’t communicate the work they’re doing or explain the output of what they’re doing to someone who doesn’t understand data science. That is really impossible to manage. I’d much rather have someone who is weaker technically, but stronger on communication (although, obviously there’s a limit!). It’s absolutely essential that you’re able to communicate with the people you work with. This involves not using very technical terminology with people who don’t understand it and being able to explain a very complex thing (like a neural network) to the CEO, for example.
Do your data science team have quite a lot of non-technical stakeholder interaction?
Of course. Speaking specifically for our Algorithm Data Scientists (as we also have Inference & Analytics people who have different skillsets), they sit within a product team which includes everyone from Product Managers, Engineers, Designers & even Researchers working in their team, and they’ll have to work very closely with these people. They’ll also have to work with a lot of people outside of this. For example, our Logistics Algorithm team (who deal with things like assigning riders to an order) have to work with our operations team who manage the day to day operations.
For people who are relatively junior, who are maybe just in their first data science role, what’s one piece of advice you’d give them?
Focus on impact! And think about actually moving what the business cares about.
Is there any other advice you would give to anyone looking to get into data science?
One important thing is, if you’re interviewing for a data science job, ask the people who are interviewing you ‘what do you mean by data science?’. Because it’s not a consistently defined term and it means something completely different in a lot of places. Also, make sure you have a really clear understanding what type of role you want and assess whether the role on offer is actually the type of role you want. Because a lot of the time people think that data science is this ‘cool’ thing that everyone should have in their organisation, but don’t have a clear use for it. I again and again see people hired into organisations as Data Scientists but have no work to do when they start.
Definitely. I see that a lot with start-ups at the moment. Some Data Scientists start with them and realise they don’t have any data, or not the right infrastructure, and they don’t actually need a Data Scientist.
On that point, we have two other disciplines in our data science function (I’ve mainly been talking about our algorithms people) and one of them is inference. These are the people who run experiments and do offline modelling and complex statistical analysis which is super impactful.
The third team is Analytics, who focus on helping the stakeholders make decisions. Arguably, what is needed in the vast majority of cases is someone from analytics – someone who can build a dashboard that can give the stakeholders clear visibility on what’s happening in the business at the moment. Or if the stakeholder has a question, they can answer it for them with relevant business context. If you ask an Algorithms person a question, by the nature of the people in the role and the time frames they work on, they’re less able to give the business context and all of the nuance needed to answer that question.
Amazing, thank you for that – it’s been massively insightful. One question I always like to ask at the end of my interviews is what book would you recommend?
I would have to say anything by Peter Singer, who is a great philosopher and does a lot of work on effective altruism and doing good things for the world, which is great for people doing data science!