Enhancing Early Warning Systems with Gen AI

Adding “Gen” to the already successful use of “AI” in Early Warning of Credit Risk: What difference does it make? 

Recent improvements in Generative Artificial Intelligence (Gen AI) have prompted the financial services sector to solve issues with this new, efficient tool. One such use case is the identification of credit risk deterioration in an Early Warning System (EWS).

In this blog, we explore the efficiency gains and end-user experience enhancements from integrating Gen AI capabilities into an early warning system for risk identification.

Our starting point is Risk Alert, an already well-established Early Warning System driven by AI. Risk Alert works in real time and uses over four hundred thousand sources to identify emerging risks with accuracy reaching high 80 percentile.  Identifying risks 9 to 18 months earlier than traditional EWS, it allows extensive time for risk mitigation.  In this blog we look at Gen AI extensions in three areas:

1. Translating- Integrating Gen AI for translation of web-scraped articles into English, which increases efficiency for reporting and review of identified risks;
2. Summarising - Using Gen AI for summarising articles by leveraging its language processing and production capabilities.
3. Accessing Information - Using a Gen AI chat bot to enhance customer experience to EWS end users.

Introduction: Applicability of Gen AI to Identifying Risk Deterioration

Automated EWS is an efficient way for lenders and portfolio managers to identify forward-looking risks. Rather than analysing historic data to identify possible risks, EWS integrates forward-looking information such as news articles into traditional credit risk indicators such as delinquencies and financial performance reporting.  Automated EWS substitutes traditional methods of manual review of internal and external sources into automated triggers based on a wide range of internal and external sources.Gen AI can substantially enhance the efficiency of an automated EWS. Gen AI can create concise summary content based on existing data by employing deep learning algorithms to analyse and learn patterns. In the following sections, the integration of Gen AI into EWS is explored through three use cases.

Use Case 1: Translation of Articles after Web Scraping

Our automated EWS leverages global web news articles to identify potential risks through natural language processing and sentiment analysis. In this process, EWS combines forward-looking prediction using news and traditional risk indicators into one rating.

Here Gen AI (see Figure 1) can be leveraged to expand the breadth of articles processed. Because global news is written in many languages, translation to English enables easier and expedited reporting and review of risks. 

This enables the EWS to cover an increased number of languages, which may alert end users of specific risks that are only published in news articles of certain languages. Furthermore, given that numerous banks and portfolio managers cover geographies that do not primarily use English, this provides additional data points for EWS and enhances risk identification.

 Case 2: Enhancing End-User Experience and Access through Text Summarisation

In addition to widening the scope of global news available to automated EWS, Gen AI also adds text summarisation capabilities to the EWS and end users. In the text scraping process, millions of global news articles are scraped for information that can be linked to an increase in risk for a chosen target, which could be a geography an industry or even an individual company.

Here Gen AI can be leveraged as a text summarisation tool that enables an end user to access the main original points of an article by providing a generalised overview, without the user having to read the full article.

 Figure 2: Summarisation Capabilites of Gen AI

Furthermore, as shown in Figure 2, end users may be interested in concise versions of articles given the amount of scraped information. Gen AI further enhances user experience by extracting the main points, directly summarising the points of the article that underpins a company’s risk rating.  

Case 3: Fast and easy access to information through Gen AI Chatbot 

Having scraped global news across languages and geographies, as well as other external and internal data sources, we now have a very large amount of data. A single global client as part of a wider group could have up to a million datapoints recorded in a single month. 

The use of a Gen AI chatbot makes searching through this information highly efficient. .Using the chat functionality, the user can enter the desired search criteria in plain language and obtain an answer, that is generated using knowledge from all the structured and unstructured data stored.For example, “What is the current rating of company A?”, “What are the risks drivers for company B?”, “Why was company C downgraded by an external agency?”, “What concerns did previous review of company D raise?”. This removes the need for manual search through the reports and dashboards. 

 Conclusion: Leveraging Gen AI for Increased Efficiency and Improved User Experience

Through integrating Gen AI capabilities into our EWS, clients have an enhanced, simplified, and streamlined user experience in summarised text that includes global news sources and non-English language sources. This drastically minimizes manual effort while providing enhanced and more comprehensive risk identification coverage. 

Risk Alert

At Deloitte, client experience and early credit risk identification are prioritised in our proprietary EWS solution, Risk Alert.  In our EWS system, we are integrating Gen AI to analyse and summarise even more data for portfolio monitoring, as well as leveraging Gen AI chatbots to enhance end user experience.