Meet The Team: David Lawton

“The storytelling side of data really appealed to me because it felt so much more tangible than just speculating about the impact of economic policy. I really like the firm answers and the challenge of finding a story in a huge dataset.”

What inspired you to go into data science, and where did you begin? 

I originally did a Masters in Economics and Trade Policy and then got a job in that field. That’s when I realised I was far happier working with numbers than with the policy side of things, I much preferred the hard world of data than the softer world of strategy and politics. I’d never really used Excel before so that’s where it all started for me. I soon discovered VBA (Visual Basic for Applications), which I used to automate a lot of processes at the company and that’s when I first started coding. I realised I really enjoyed that, so ended up taking a course in R and that’s when it all clicked. The storytelling side of data really appealed to me because it felt so much more tangible than just speculating about the impact of economic policy. I really like the firm answers and the challenge of finding a story in a huge dataset, and thinking creatively about how you can find that story. 

What are you working on currently? What are you enjoying most about it? 

I’m currently working on an econometrics project, helping our client understand the return on investment from marketing to make a business case to significantly increase their current levels. This involved taking a deep dive into spend per channel and how they differ, then building this up into a business model with plenty of controls and caveats built in to enable us to forecast the impact of increasing spend to unprecedented levels. 

We’ve also recently completed work on Fluency’s ‘Metaverse Valuation Tool’ which helps brands to size their financial revenue opportunity in the metaverse.  Here we used a two-pronged approach, firstly from the bottom-up where we look at individuals in the market and understand their propensity to participate in the market, which we then slowly ladder up to evaluate the whole market value. Next, we performed a top-down approach, where we look at data on the current market value of the metaverse, combine it with data from our $1mil+ stack to split it out into actionable segments. The two approaches then allow us to produce a middle-ground range to inform the final opportunity size for the brand. This process involved a lot of Excel modelling to allow the client to fully interact with the end-tool and allow for adjustable assumptions within this too.

The team are also working on an AI project that can predict the performance of social media posts. Here we are tapping into a wide range of cutting-edge AIs to really get under the skin of what matters on social media. 

What would you say has been the biggest development in data science in the past 5 years?

Generalisable models have been a huge development in data science over the past few years. They are enabling us to do things like the zero-shot learning. It’s a significant breakthrough when models go from being trained for a very specific single use case to being so good they develop a multitude of use cases. It provides us with so much more flexibility. If you want to go and tag a load of text for some really obscure things like bike parts for example, you can go ahead and do that without having to build an AI model from scratch. I think things like generalisable models are one of the main advancements in AI that’s helped it go from something quite niche to a powerful tool with relevance in everyday applications.

Where do you see the world of data science going in the next 5 years? 

There seems to be a lot of exciting developments taking place in things like GPT and being able to text generate. This would likely fit into smart-homes very well, making Google, Alexa and Siri much more conversational then they currently are. 

The area I’m personally hoping will make big progress is explainability. Currently a lot of AIs operate as a ‘black box’, where it can tell that the object in an image is either a dog but is not able to explain why, since it’s such an intricate machine. Explainability is trying to push away from this ambiguity. The SHAP values method is an exciting recent development in the area: it takes an AI and uses cooperative game theory to try and figure out the mechanics hidden in the model and is able to clearly describe how it works to humans. A lot of the fear behind AI is attributed to the fact we find it difficult to explain properly, so I think there should definitely be a push in making AI more understandable.  

Do you have any advice for those thinking about getting a career in data science? 

The advice I always give to those thinking about getting into data science is that you should make sure it’s a real passion of yours. If you’re not genuinely interested and you’re just doing it to try and get a better salary or because it’s job security etc., you’re unlikely to do brilliantly even if you’ve got a good mathematical brain. If I look at the stuff I use on a day-to-day basis now, maybe 60% of it didn’t exist five years ago. If you haven’t got that passion and you’re not constantly reading up on new developments, you’re just going to fall behind. 

To round off, what fictional character would you say you relate to the most? What traits do you share? 

I’m going to have to go with Hobbes from Calvin and Hobbes. He’s sarcastic and a bit of a sceptic, but ultimately willing to give new things a go and see what happens!