Technology Needs A Catalyst. How Does Survival Sound?


By Adam Sussman, Global Head of Market Structure, Liquidnet.

Artificial Intelligence is not necessarily about predicting price, but predicting what kind of content, products, or services a user would find valuable.

According to a recent survey, 71% of US respondents believe that Artificial Intelligence (AI) will eliminate more jobs than it creates. Would the number be much different within the financial services industry? Almost everything I read or hear about on AI and automation is focused on the wrong issues. The discussion on automation versus augmentation is unhelpful; it minimizes the fear of job losses by downplaying the promise of massive efficiencies. Artificial Intelligence, like many buzzwords, is as popular as it is misunderstood. A recent Liquidnet-commissioned Greenwich Associates study indicates that only about 10% of Portfolio Managers and analysts use AI in the investment process. Yet, anecdotally, aren’t we hearing that EVERYONE has embraced AI? I bet a survey of those same firms’ marketing departments would show different results.

My point is that a lot of people who say they are using AI are not, and some people who are using AI don’t even know it. The latter is the more interesting case because when technology is put to its best use, we don’t know we are using it. I doubt that the folks that use 3rd party tools that leverage AI (6%), within the same Greenwich Associates study cited above, know exactly where and how AI is being deployed.

There are a few reasons why AI has not been more widely adopted. First, many alternative data providers tout their use of AI to generate a trading signal. They take unstructured content, such as written content and satellite imagery, and use various techniques to translate that raw content into structured data. AI can accelerate and improve the process because there are millions of data points (words and pixels) to train the model, and a robust training set with which to do it.

Even when the above is done well, the challenge is that the fundamental analyst doesn’t typically benefit from a data set that tries to predict price action in days or even weeks. They are called “long-term” investors for a reason. They care less about what the number of cars in the retail parking lot means for next quarter’s short-term revenue miss, and more about what the number of cars in the employee parking lot means for hiring trends.

Another shortfall in the use of AI is among investment research tools. According to the same Greenwich survey, alerting functionality is being used by just over half of the 56 respondents across US, Europe and APAC. The use of these alerts is for relatively simple events such as a new company filing or earnings announcement.

One use case that resonates with fundamental analysts is using Natural Language Processing (NLP) to help uncover when a company has made a notable change to some part of its filings. There is a substantial amount of academic literature in this space; the more interesting ones look at the words, phrases and changes that suggest future corporate events such M&A, litigation, or fraud. How that might translate into price movement is not necessarily a useful step.

In other words, one value proposition of Artificial Intelligence is not necessarily about predicting price, but predicting what kind of content, products, or services a user would find valuable.