How Bloomberg uses machine learning to create a competitive advantage

Profile picture for user Mark Samuels By Mark Samuels December 27, 2018
Summary:
Emerging technology is at the heart of the financial data company’s ongoing transformation efforts, says CTO Shawn Edwards.

bloomberg

Bloomberg is using machine-learning technology to exploit its vast data sets of financial information and create game-changing innovations for its staff and customers.

According to Chief Technology Officer Shawn Edwards,  his office is harnessing the power of emerging technology to develop new products and services. Edwards says the potential for change is significant:

I think the application of machine learning to financial data is exciting. It’s something that we’re focusing on in the CTO office. We need to think about our how our customers are changing how the marketplace is changing. A lot of what we do is centred on internal infrastructure, but we also do some product incubation – and some of those ideas and concepts eventually become products.

To this end, Edwards says his team is currently leading an effort to build a data model. This model includes metadata frameworks and database ontologies. While the use of models, frameworks and ontologies is nothing new to Bloomberg, Edwards says it’s the first time the company has allowed people to programmatically access its data:

It’s the capability to query and run compute on Bloomberg across all of our data sets. It’s about being able to do a cross-domain query, scan millions of bonds and join that with something about people, such as trading history. It's a game changer in terms of all the kinds of things you can build. It's pretty exciting.

Helping customers make the most of data

Edwards says the new data model is, in short, about bringing together all the financial data the firm holds, exposing that to internal and external customers, and then building tools on top of that information so that people can create new insight. Machine-learning tools will be a part of the suite of capabilities that the firm offers to its customers. Edwards expects those developments to bring benefits to Bloomberg’s clients:

They can explore more ideas. Someone might, for example, have a hypothesis that they can run price prediction on retail sales. That was hard to do before – they might have had a great idea, but they couldn't do it. In the near future, they'll be able to do that experiment, explore, and see if it has validity in relatively short order. And then, if something does have value, put it into production.

Bloomberg is currently building a data science platform for its customers, which Edwards says is “really exciting”. The development – which is a scientific computing environment based on open-source software – is being led by the CTO office. The platform includes all the maths and scientific libraries that are required for high-level computation. The platform also includes a rich visualisation layer built using open-source technology bqplot, which provides a way to use and build charts and graphs:

The value we're offering to our customers is the fact that you can have access to all this data programmatically. All the analytics that we have in Bloomberg will eventually be put into this environment. Right now, you can go and optimise your portfolio using our portfolio analytics system, which already exists. We’re putting ties and hooks into all of our analytics systems and all of our features. It’s really powerful and potentially game-changing.

Edwards says the firm uses machine learning internally to build products, suggesting it has been pioneering developments in this area for many years. This experience will help the CTO office to develop new capabilities, says Edwards:

Our initial use of machine learning was about extracting unstructured data from structured documents. So, when we get an earnings report from the customer, we used to have a clerk look at the document and then type the values into a database. Now we have these machine-learning models that can pull out the headline number. They actually read tables and extract the data better than humans; they perform with less errors and do all this in milliseconds.

Augmenting capability with emerging technology

Edwards says the firm is seeing benefits from machine learning in other areas, too. He refers to the automated production of reports. Bloomberg is using artificial intelligence (AI) to help create content – such as news stories – from the financial data it pulls together, says Edwards:

We build a lot of analysis around those kinds of documents, so news and analytics, and sentiment analysis. We've been using this technology ourselves but we're also exposing some of these capabilities to our customers. Customers want to be able to extract data and value out of their documents.

There are fears that the widespread use of AI and machine learning could lead to the replacement of humans with robots. Like many of his digital leadership peers, Edwards prefers to see automation as a form augmentation. Rather than replacing workers, he says the use of emerging technology at Bloomberg will create opportunities for staff, such as those in editorial, to work in new areas:

It’s not in our best interest to replace our journalists. We’re not writing news stories with AI to reduce our headcount – that’s not what we’re doing. We’re using our technology to free up our journalists to look behind the scenes. We want them to focus on colour and the investigative side of things. We actually use some of these tools to help journalists. We augment the writers; we give them some really interesting tools internally, so while they're writing stories and breaking news, we give them context. We use machine learning to pull out data points quickly, so our writers can produce breaking stories.

The potential impact of AI at Bloomberg is not just restricted to pulling together data and creating content. Edwards refers to the impact of automation in sales and the potential for emerging technology to support a performance boost:

If you look at some of our customers, a salesperson might have chats with 40 or 50 clients. We're helping them look at this information structure, helping them extract information out of their own systems, giving them the API so they can gather all this information. Instead of typing into five different internal systems at Bloomberg, such as inventory or pricing, our computers can do that. That automation makes our salespeople way more efficient, so that they can cover more customers or miss less trades.