Interview: Tim Baker, Global Head of Applied Innovation, Refinitiv

AI and ML are a hot topic, but how much is hype and how much is reality? This is a big area of interest and concern for our customers and the future of finance. I’m excited to hear from the practitioners and hope we’ll get to the bottom of where the real use cases are emerging, so we can further provide real-world context as to how customers can utilise AI and ML to access clean, usable data, create efficiencies and lower total cost of ownership.

The biggest challenge we see is around finding/sourcing, testing and on-boarding data. Roughly 80% of a data scientist’s time is spent wrangling and cleaning the data. Once tested and built, the next challenge is moving the resulting work into a production environment where data scientists can further test and ultimately scale across their organisation.

Actually, to better understand their challenges and barriers of ML adoption, we recently conducted a global survey, of over 450 interviews of data scientists and C-suite technologists. What’s the importance of machine learning techniques? What’s the current adoption of ML and subsequent content usage in their firms? We’ve been compiling our research and will be releasing our findings later in March. You’ll be able to download the full report on Refinitiv.com – stay tuned!

Previously, you may have known us as the financial and risk business of Thomson Reuters. However, as of October 1, 2018, we are now Refinitiv. While our name has changed, our commitment to our customers and partners has not. With Blackstone as our new investment partner, we are in a unique position to leverage greater financial flexibility in order to invest across a range of high growth opportunities. 

We are one of the world’s largest providers of financial markets data and infrastructure, serving over 40,000 institutions in over 190 countries. We provide leading data and insights, trading platforms, and open data and technology platforms, which connect global financial communities - in trading, investment, wealth management, regulatory compliance, market data management, enterprise risk and fighting financial crime.

Elektron Data Platform is the data platform which supports all our businesses and delivers data into our customers. We are working to make it easier for data science teams to find, test and leverage our diverse data assets through advanced APIs and delivery of large data sets into the cloud. Our goal is to deliver data science ready data.

One of our global asset management customers is using our NLP solution, Refinitiv Intelligent Tagging, to deliver semantic search and alerting to their global team of portfolio managers. Integrated into their Office 365 stack, Intelligent Tagging identifies and tags unstructured content such as emails, research and news, and delivers it into the portfolio manager and analyst workflow. All of this makes content more accessible and valuable via the automatic generation of rich metadata.

An understanding of the subject matter (financial services domain), the data (and its shortcomings), and be able to select the appropriate tool given the problem being solved.

Data Science tools and platforms are becoming more accessible and easy to use – platforms like H20 and DataRobot, to those made available by the big cloud vendors means the emergence of “Citizen Data Scientists” is becoming a real possibility. For Refinitiv, we attract talent because we have world-class data to drive performance and innovation – the key ingredient for a great career in data science. 

We don’t really have exposure to this type of risk as our focus is around market and trade data.  That said, badly designed trading and predictive algorithms can suffer from different types of bias or sources of error.  Common problems we see in finance are: 

1) Look ahead bias: models are tested using data not available at the time period being analysed

2) Over fitting: when insufficient data leads to an overly complex model in an attempt to explain the shape of the data

3) Sampling bias: when you use an unrepresentative sample to build your model

The key first step is to fully understand the problem, which usually involves helping the business define the problem statement or business hypothesis. Next, build a simple prototype and iterate with the business to develop the solution. Avoid black boxes, and anticipate how the solution will move to production.

Consider your audience. Make it tangible through real life use cases. How have others in your industry harnessed a similar opportunity? And, in some cases, analogies from other industries are needed to better provide a complete picture of what’s possible.

Certainly some data sets (e.g. Point of Sale) are less to do with finding unique sources of alpha and are now just contributors to the efficient market hypothesis – so you can’t not look at this data for certain industries (e.g. Retail). However, with that said, there are still lots of opportunities in other industries.

No matter how you look at it, our customers are adopting cloud at an accelerating rate. And, the biggest adoption is coming from pre-trade and post-trade activities. From our perspective, we are meeting our customers in whichever cloud they have chosen. We have launched various cloud native capabilities such as our real-time data on AWS, QA Direct through Microsoft Azure (which is our quantitative analysis platform), and Tick History through Google Cloud (available later this year). Additionally, based on customer feedback, we have a wide range of data which our customers can take as APIs and directly feed to their cloud platform.