Democratization of AI, NLP and Text Analytics: an opportunity for smaller and medium sized Financial Institutions

Democratization of AI, NLP and Text Analytics: an opportunity for smaller and medium sized Financial Institutions
The future successful organisations in the FS industry will be those who harness the power or text analytics and NLP. Those who ignore it, do so at their peril. The democratization of Artificial Intelligence, however, has demonstrated that the challenge for FIs is not the technological capability but rather the identification of value generating use cases.

NLP libraries like Spacy, Hugging Face and models like the Googles Switch Transformer or GPT-3 from Open AI allow technologically skilled teams in Banks to leverage this innovation. The application of transfer learning significantly lowers the barriers to entry such as data availability and computing power for training, thus making sophisticated use cases for smaller FIs possible. On the other hand, the proliferation of no-code or low-code platforms for Machine Learning & AI bring the technology to the business users. Medium and smaller FIs can thus generate a competitive advantage instead of following the market. With this article I demonstrate several use cases from the industry to serve as ideas for these FIs to help in this mindset shift.

Several Techniques can be used in navigating the capital markets by leveraging unstructured data.       

a) Key Phrases

Extract key phrases from large text to generate a summary.

b) Sentiment Analysis/ Text Classification

Analyze the sentiment of text based on topics, entities, or temporal direction of statements to generate more granular insights.

c) Temporal Analysis

Identify the temporal direction of the statements to evaluate the relevance.

d) Topic Extraction/ Topic Modelling

Extract statements regarding a topic of interest for your analysis or model the underlying topics in large text.

e) Q&A – Semantic Search

Intelligent search based on semantic similarity rather than exact keyword matches.

Use Cases in Due Diligence


Be it mergers, acquisition or private equity transactions, due diligence is a complex process that requires hours of professional work across different parties. One of the activities here is the redaction of confidential information from certain documents that can at times be high-cost, expensive manual work. NLP and Text Analytics techniques such as Named Entity Recognition, semantic similarity and image processing now offer a strong set of applications in this context. Confidential information, terms, and topics such as commercially sensitive information, sensitive personal data (GDPR) and others can be intelligently masked and unmasked, depending on the audience viewing the data. Below is an illustration on the possible steps to take for developing a similar model.




Vector representations of the sentences, with models like BERT, GPT and others, facilitate search within documents based on similarity. This means that functionalities such as finding similar words or expressions within a document to redact are also possible.

Contract analytics are another group of use cases powered by Text Analytics and NLP used in due diligence processes. Be it for a buyer reviewing existing contracts of a potential target, or the target carrying out an initial diagnostic analysis, a streamlined and automated process can save significant amount of manual work. Contracts keep information in an unstructured manner but include important clauses that may fall under the radar during the due diligence process. The thorough analysis of the contracts naturally involves exorbitant legal fees. Semantic search, intelligent tagging, and summarization of important information from contracts can significantly reduce the number of hours lawyers need in analyzing the contracts and thus save costs.

Research and analysis of the investment climate


Buy-side firms have already seen several uses of analyzing unstructured data for research or investment management purposes. Along with text data, there are also use cases available for leveraging foot traffic information, online search, and other similar data for more refined forecasts. During the start of the pandemic, many of the analysts were slow in updating their numeric targets for the companies although many of the companies have withdrawn their earnings guidance. During the financial crisis a similar phenomenon was observed. During the times of uncertainty numeric indicators are updated less often whereas the current challenges and the existing investment climate is discussed at length in text form. Investment managers that leverage this data in decision making processes have a holistic view on the current climate. Normally a human analyst would review these reports. But in a fast-paced environment manual review of all documents is not feasible. As such NLP and Text Analytics techniques can help in doing an initial screening of the reports and highlighting the areas to have a look at.


Nowcasting and Central Bank research


During the significant volatility driven by the lockdowns and the pandemic, central banks were struggling in adjusting their monetary policy. Conventionally used figures for monetary policy committees such as GDP, consumer sentiment indexes, inflation numbers and others are backward looking. Major central banks have thus seen greater use of so called nowcasting. The need for immediate data in stressed scenarios is rather pressing. As such nowcasting, by leveraging more high frequency data such as text and newspaper articles, provides an immediate insight into the market environment. Nowcasting is not only used in making monetary policy decisions but also gauging the effect of the decisions made and maneuvering where necessary. The trick is constructing the dataset into a timeseries that can be aligned with corresponding numeric indicators. Researchers from the Norwegian Central bank use LDA (Latent Drichlet Allocation) for this purpose to extract topics from news sources and forecast US GDP, investment growth and consumption.
 

3 Recommendations to smaller and medium sized FIs


Alternative data sources such as text demonstrate a wide set of advantages over conventional numeric indicators. These are more frequent, tell a wider narrative about the economy, incorporate qualitative information, and speak about expectations. With this in mind, there are 3 things that smaller and medium sized FIs should do to gain a competitive advantage over the larger incumbents in the market.

  • Technology and business teams inside the FIs should do regular workshops together to identify business focused use cases for leveraging AI, ML and NLP. Speaking to consultancies helps in identifying use cases given the wider industry footprint and the continuous research that consulting firms do.
  • Consider deploying no-code AI platforms to empower business users to experiment. Its during experimentation that ideas come to life. With closer collaboration with business focused teams, a much stronger business case behind the technology initiatives can be made.
  • Prefer finance industry specific over generic solutions that are aimed at any industry. A sustainable competitive advantage can be achieved when solutions leverage industry expertise and are configured to individual organisational needs.

George Karapetyan

Manager / Data Science & Machine Learning

LPA Group Joins IDC FinTech Rankings by IDC Financial Insights

LPA Group Joins IDC FinTech Rankings by IDC Financial Insights

LPA Group hires new partner to expand Swiss consulting practice

LPA Group hires new partner to expand Swiss consulting practice

The end of the transition period is in sight. What particular challenges does the PRIIPs Regulation pose for real estate fund managers?

The end of the transition period is in sight. What particular challenges does the PRIIPs Regulation pose for real estate fund managers?

Go back to all news