Blog image for In Conversation With Series Showing Elena Chatzimichali

Exploring the Opportunity and Functionality of AI and ML in Debt Capital Markets

Following the successful release of DealPro Data, our CEO Sotiris Manderis chats to Elena Chatzimichali, our senior AI advisor. With Finsmart’s focus in data analytics and AI/ML, Elena explains AI in financial services, and outlines why AI and Machine Learning is so critical for debt capital markets participants.

What Are The Differences Between AI, Generative AI & Machine Learning?

In the simplest of terms, Artificial Intelligence (AI) is the science of simulating the functionality of the human brain, the most powerful and versatile human organ; AI aims at machines (computer programs) being able to ‘think’ for themselves, and make actionable decisions based on the data they are being fed; the data represent prior knowledge and stimuli in order for the algorithms to “learn”. More specifically, Artificial intelligence (AI) is the field of Computer Science that enables computers to perform a variety of advanced functions, including the ability to see and derive meaningful information from digital images, videos and other visual inputs, understand and translate spoken and written language, make personalised recommendations, and more [1][5][10].

For the past year, Generative Artificial Intelligence (Gen AI) has been making the headlines. Gen AI describes algorithms and applications such as ChatGPT [11][12] that instead of just forecasting, they can be used to generate new content, including audio, code, images, text, simulations, and videos, among others [9]. The widespread fascination with ChatGPT made it almost synonymous with AI in the minds of most end-users. However, it represents only a small portion of the ways that AI technology can being used today.

Finally, Machine learning (ML) is one among many branches of Artificial Intelligence that enables a system to autonomously learn and improve without being explicitly programmed (no explicit instructions are given). ML algorithms are fed and trained on large quantities of historical data (“learning by experience”) in order to perform complex tasks, generate predictions or identify hidden data patterns. While both of AI and ML go beyond basic automation and programming to generate outputs based on complex data analysis, ML has a narrower scope and focus compared to AI [7].

Overall, AI is a fascinating technology that has the potential to revolutionize the world as we know it.

How Can AI Deliver Real Return In Financial Services?

AI is already in our everyday lives and is the backbone of innovation in modern computing, unlocking value for both individuals as well as businesses [1].

Most end-users interact with AI on a daily basis without even realizing it. AI is changing the quality of products and services that the end-users get accustomed to and therefore expect. Financial services will have to adjust and evolve in order to meet this ever-growing demand.

With the availability, rapid development and growing recognition of AI technologies, data has become the most valuable asset (“data is the new oil”) in any organisation, including the financial services. The main aim is to turn data into valuable and actionable insights. These, in turn, create more opportunities for innovation, product development, organizational efficiency, improved customer service and, finally, profit. Now more than ever, banks and financial institutions are aware of the innovative and cost-efficient solutions AI provides [2].

ML and AI, are not only providing better methods to handle large amounts of structured and unstructured data, understand markets and consumers, and improve customer experience at scale but also help simplify, automate, speed up, and redefine traditional processes to make them more efficient. According to a post by Google, AI in finance can help in five general areas: personalize services and products, create opportunities, manage risk and fraud, enable transparency and compliance, automate operations and reduce costs [3].

AI is fast becoming the norm, and businesses that fail to embrace it run the risk of being left behind as their competitors innovate and grow.

Is AI A Black Box?

In order to address this question, we need to look at the main pillars of building an end-to-end AI pipeline:

A. Define the objective (business problem)

As a first step, a clearly-defined use case needs to be set. The objective signifies the task that we want the ML program to achieve. AI is in no way “one solution fits them all”. Therefore, before we even initiate any ML/AI pipeline, the most crucial and important step that we must take into consideration is understanding and defining the problem that we’re trying to solve and its scope. No matter how good the data and the model, the solution will only be as good as the problem you set. According to Riley Newman, “good data science is more about the questions you pose of the data rather than data munging and analysis”.

B. Data

The next step is data. Often, we may need to revise a problem statement (step A.) based on the data we possess. The data is the collection of examples that the program uses to learn from. They may take various forms (structured and/or unstructured, confidential, sensitive or public), derive from multiple sources (standalone or aggregated data), reside on various infrastructures in-house or on the cloud, among others. Where possible, the data should span a significant historical horizon and be diverse enough so that the model can adequately learn patterns and trends. Even with a large amount of diverse data, the necessary steps of exploring, visualising and data cleaning are necessary prior to any modelling to avoid the well-known “garbage-in, garbage-out” effect.

C. Algorithms – models

The third step is how the program would learn from that data. This is that stage where Machine Learning or AI techniques and algorithms come into play in order to create data models that can be of descriptive or predictive nature. There is an abundance of AI algorithms, ranging from very simplistic open-box models to extremely complex black-box models, each of which accomplishes different levels of functionality, accuracy, complexity and transparency. Relevant ML metrics and business KPIs need to be set to test the performance and adequacy of the models.

D. Explainable, responsible and trustworthy AI

Model interpretability deals with the level that we can question, decipher and trust an AI model; it is crucial for solving business problems, especially in regulated industries such as the financial services and healthcare. Interpretability also reflects our domain knowledge and societal values, provides scientists and engineers with better means of designing, developing and debugging models, and helps to ensure that AI systems are working as intended [13].

How Will AI/ML Enhance Finsmart’s DealPro Data Tool?

Our daily discussions with bookrunners and issuers have provided a number of use cases where AI/ML will be the best approach to use. The overall benefits we are aiming for are:

  • Ease of data mining. Ensure complicated questions can be answered instantly
  • Generating automated insights as/when required based on investor/market events
  • Maintaining clean and reliable sets of data
  • Combining data from various sources and visualising in one, user-friendly environment.

Here are a few of the concepts we are working on at the moment:

    A. Natural Language Processing (NLP) and Gen AI

By harnessing the power of NLP on written text, accomplishing human-level understanding and deciphering of complex search queries by end-users e.g. “Who has invested in the pharma sector but not in a certain pharma company within a certain currency and time period?”.

    B. Investor Pattern Detection and Forecasting

Using historical information and live order books to forecast price or market event sensitivity of investors.

    C. Recommender Systems (RecSys)

Given an upcoming deal and past historical activity from various sources a RecSys algorithm could help detect, rank and return the top N investors that are likely to be interested in placing an order of e.g. “product X of sector Y under market conditions Z”, detect which investors have not yet placed an order and, where possible, provide explainability (XAI) of the prominent features that led to this outcome.

For more information on DealPro Data or for a demo contact


The Finsmart Team