Monitoring tool in Tableau for early risk signals

Knowing your suppliers and identifying their critical goods and services is the first step to a successful risk mitigation strategy. At Deloitte, we call this strategy as Third-Party Credit Risk (TPCR).

The main domain of the TPCR is an identification of early signals of distressed companies (third parties) from various perspectives. By our approach, we consider financial and operational stress symptoms tracked by various Key Performance Indicators (KPIs) to increase the chances of a successful and timely risk identification.

Moreover, these symptoms are complemented by scanning of Adverse Media for all included third parties. By our NLP technique AOsint, we can identify relevant articles (scanning in more than 17 languages and more than 4 mil articles per day), score them, and assign them specific sentiment scores. As a result, our clients can monitor alerts about their third parties by the interactive TPCR monitoring tool.

TPCR approach in a nutshell

1. DATA & PARAMETERS

Data collection and combination of various data sources (such as Dun & Bradstreet, open- source data of Adverse media etc.) to get as much information about each company as possible;

• Information was collected for financial and operational data, and together with Business, we defined a set of KPIs to be monitored:
  • About 20 Financial KPIs – e.g. Interest coverage ratio (ICR), Gearing ratio, Debt ratio, EBITDA etc.;
  • About 10 Operational KPIs – e.g. OPEX ratio, Quick ratio, Current ratio etc.
• Moreover, News was scanned by open-source Intelligence called AOsint (Deloitte’s asset – read more here) to find a number of articles per each company together with risk scoring of articles in various risk categories and sentiment information.

2. MACHINE LEARNING MODELS

Two Machine Learning (ML) models for financial scoring and Adverse Media scoring to get into a relationship and track a probability that the chosen company is or will be distressed.

Based on defined parameters (tailored by customers’ needs and requirements), the logic for each ML model is defined separately:

• For financial model, almost 30 KPIs were included for last 6 years and applied for gradient boosting learning process (XG Boost) and in the next steps tree models for decision making were designed ending in the final scoring 0-1 (the higher the score, the model predicts company may enter distress or is already distressed);
• For adverse media model, we used our Open-Source Intelligence called AOsint (read more here) and the result between 0-1 indicates a negative abnormality in media presence (the higher the score, the more alerted the user should be as the articles indicate the company has events that may lead to distress).

3. ALERTS IN TPCR MONITORING TOOL

Results from both ML models are in a form of alerts so our client can easily check which third parties are in “bad” interval over time. This interval indicates that a specific vendor is or will be distressed in the next period.

Afterward, the client can drill down to get more details about each vendor – from financial and adverse media perspectives.

Benefits that the TPCR brings to our clients

• Not only early signals of distressed third parties, but also predictions whether a vendor could be potentially distressed for the next period;
• Fully Automated NLP (Natural Language Processing) Techniques with minimal manual inputs
  • Adverse Media model (powered by AOsint) monitors News in more than 17 languages and can scan up to 4 million articles per day;
• Our clients get a monitoring tool to track all required companies and can find distressed ones very quickly (based ML alerts → financial and News)
  • Compatible with various BI tool such as Tableau, MS Power BI etc.;
• After initial screening, other details about various financial and operational KPIs together with all articles are available;
• Export feature of TPCR monitoring tool - raw data processed by both ML models and providing additional AI insights can be sold to other customers as a part of supporting of Data monetization.