AFTAs 2018: Best Analytics Initiative—Credit Suisse

Artificial intelligence (AI) may be all the rage at the moment, but less-often mentioned is a subset-of-a-subset on the bleeding edge of technological development—deep learning. Here, Credit Suisse shines for delivering actionable insights based on its work in natural-language processing and deep learning, via its analytics platforms, and wins the AFTA for its efforts.

One of the major problems facing equities research analysts isn’t necessarily information shortage—it’s too much information. The volume of emails and notes both generated and received by staff on these teams is immense—in Credit Suisse’s case, its Global Markets Equities Research Team has to deal with over 3 million emails and 500,000 meeting notes—but finding an efficient way to analyze these client communications has always been challenging.

Enter AI. The goal of the project was to deliver insights around four main areas: topic modeling, in which potential sales ideas could be generated around stocks or bonds that may be potentially of interest to clients; ticker tagging, in order to identify said instruments for future reference; contact-level sentiment analysis, to determine overall perception of products and services; and auto-summarization, which would enable the machine-learning algorithm to generate commentary on a client.

Implementing this required an element of trial-and-error at first, partly because this is breaking such new ground within actual implementations of machine-learning technology. “We tried various models, such as random forest and a few others, and we obviously wanted to give deep learning a shot because we had enough data to try this out,” says Paras Parekh, head of the predictive analytics team within the global markets technology group at Credit Suisse. “It happens that the deep learning model works fairly well compared to random forest.”

The work so far has resulted in the development of two dashboards which, the bank says, have generated “actionable insight.” The first displays an email or meeting note which has been through the algorithm, cleaned and tagged appropriately, along with sentiment. The second displays time-series data at a contact level, and the top entities by mention, as well as trends by sentiment score. All of this, too, while the platform is still effectively being built.

“We’re still fine-tuning the model. It’s giving us enough insight, but there’s still some work needed, especially when we’re looking to find cross-selling opportunities,” says Parekh. “But if we had to look for investment ideas for stocks that the client may be interested in, that’s working quite well. We’re also able to do a fair bit of entity recognition and trying to understand all the various entities involved for an individual or within a group at a client level or a macro level.”