There’s no data science unicorn - building a data team at HSBC

HSBC is one of the world's largest financial institutions, serving more than 40 million customers globally. One of its largest divisions, Wealth and Personal Banking, supports individuals, families, business owners, investors and entrepreneurs. It provides products and services that include current accounts, credit cards, personal loans and mortgages, as well as savings, investments, insurance and wealth management.

At the centre of the Wealth and Personal Banking division is a data analytics group, which is responsible for providing data-tailored services to HSBC teams and customers all around the world. Ranil Boteju, Global Head of Data Analytics at HSBC, was speaking this week at the Big Data LDN event, where he shed some light on what it takes to build an effective data science team that can scale.

Key to Boteju's message was that fostering specialists and establishing domain knowledge are important, as well as seeking out soft skills. And that there is no such thing as a ‘data science unicorn' - someone that can do it all. But more on that later.

Firstly, Boteju outlined the five core areas that his team focuses on. These include:

• Using data to drive commercial and revenue outcomes. For example, using data within the bank to create lots of personal and relevant messaging and tailored offers to instigate new sales or acquisitions. Part of this is also using data to derive insights into how customers are using products and understanding customer behaviour.

• Using data combined with techniques like machine learning, natural language processing and OCR to build new experiences for customers.

• Looking at how back office, middle office and front line teams work and applying automation and intelligences to processes take strip out manual and repetitive work.

• Building algorithms to help protect customers from fraudsters and criminals

• And providing business teams with insights, dashboards and metrics

Boteju explained that key to effectively scaling these data practices across HSBC has been having a central, global team that can build a product once and then roll it out across all locations. He said:

HSBC is a global, international bank. Banking is reasonably similar across the markets. In general how our customers deal with the bank, it is consistent across cultures. Having a global team, what that enables us to do is build various machine learning and data services, build it once, prove that it works in one market, and then rapidly deploy that across other regions.

We have a philosophy of build once, deploy everywhere. For example, if we build a recommendation engine that helps our customers select the best rewards offer from their credit card in Hong Kong, that same algorithm can be recalibrated in the UK or the US. Different data, but the mechanics is very similar. It gives us a lot of scope to scale.

Boteju said that this approach has been adopted over the past six or seven years, after HSBC realised that it was repeating the same mistakes again and again. He added:

[We were] reinventing the wheel every single time in every single country. That's not very efficient. What we've done is create a global team and wherever you are in the world you have the same business objectives. We really standardised how we do things and we have what we call global specialisms, or disciplines. For example, the way we do fraud analytics is globally consistent - one global team that makes numerous local instances, but they're all joined together and they're in constant communication. We work seamlessly across borders. And even though we are in different locations, the team operates as a whole around these discipline areas.

Acquiring the right skills

In terms of building the right skill set for a mature data science team, Boteju has learnt to take a multi-pronged approach - focusing on soft skills, technical skills and domain expertise - and then layering these together to support the business.

Firstly, on the soft skills side Boteju said that there are three key things that he is looking for. These include:

The first one is being really curious. In the world of analytics and data, you need to be really curious about how these things come together, using different techniques, how you can work with a front-end DevOps team to embed those in an experience. So being curious and really thinking about things that don't exist. Part of being curious and using data and analytics to come up with insights is to always think about ‘where next?' And always thinking about the business problem.

The second trait is someone that is always continuously learning. I've been in data and analytics for 20 years now and...if I think about the day of a junior analyst today compared to when I started, it's completely different. What we've seen is that there are always new approaches and new techniques, so having people that are really open to continuously learning and try different things is really important.

And then the third thing that's also really important is that really proactive commercial drive. If you're curious, constantly learning and got all the skills, you also need to be always on the hunt and on the lookout for the use cases where you can come up with amazing experiences for customers or change a colleague's whole workflow. We aren't in a world where someone is going to come to you and say please build this for me. We need the data analytics person to do that and be really hungry for it.

In addition to these three core requirements, Boteju obviously also looks at communication skills, given that the data science team is working cross-functionally with a wide variety of teams and people, across a number of different locations.

Finally, Boteju said that he's realised during his time working in the field of data science that there is "no such thing as a data unicorn". He explained that four or five years ago there was a view that a single person could do everything - from data engineering, to building algorithms, to communicating really well and pulling it all together. However, this is not true and for real scale, you need to build a diverse team of talents with individual skills and knowledge. Boteju said:

That's not really worked out and if those people do exist, good luck to them. For organisation like HSBC we just can't find enough of those people. We need to massively scale. What we've landed on is a different approach, where we see different types of roles. So rather than the unicorn that does everything, we have some people with data engineering skills, we have people that do data modelling and data management, we have data scientists that love to play with data, create algorithms, solve business problems. There's also a different role type that we consider a business translator role - these are the people that can scope out and really own a digital machine learning service, or alternatively someone that can really translate this to business or lay people.

And then the new role that's becoming even more important is what we call machine learning engineering, who is the person that can take those algorithms, work with the DevOps team and create those machine learning pipelines. We've really thought about those roles as really distinct disciplines. We try and have people really specialise within their lane.

Then we've also really thought about different domain areas. So in the world of banking you do need to know a lot of domain. For example, there is a person in the world of price optimisation, who might not be the best person to build you a chatbot. It's a very different type of skill. So even within data science your domain understanding is also very critical. So what we've done is overlaid the combination of the disciplines I've mentioned with different domains. We try and get people to focus on those.