Cortical.io Case Studies How a large bank saved thousands of hours of manual labor by automating credit agreement analysis

Edit This Case Study Record

	How a large bank saved thousands of hours of manual labor by automating credit agreement analysis Cortical.io

How a large bank saved thousands of hours of manual labor by automating credit agreement analysis

Cortical.io

Technology Category	Analytics & Modeling - Natural Language Processing (NLP)
Applicable Industries	Finance & Insurance
Applicable Functions	Business Operation
Use Cases	Fraud Detection Regulatory Compliance Monitoring
Services	Data Science Services
Challenge	The bank was looking for ways to improve efficiencies through automation, particularly in the area of credit risk assessment. The complexity in language and structure of wholesale credit agreements made automation difficult. These agreements are complex, 100+-page-documents that are very bespoke because banks use different lending systems. Especially covenants use highly specific, use-case related formulations. Some extractions are straightforward, and the values can be extracted “as is” (no inference necessary), for example: named dates, commitments, rates, parties. Others are more challenging and require fine-grained distinctions between options or covenants (eg redeem deal options, termination options). Here an automation system has to comprehend domain logic based on document structure and relations of extracted items. Pricing-related tables contain many different parameters (applicable margin, interest payment schedules, commitment tables) and it is extremely difficult to automate the extraction of pricing information. This is why the review of credit agreements was still done manually, which cost a lot of time and money to the bank, not mentioning the high error rate of such a mundane task. Read More
About Customer	The customer is a Tier 1 bank in the US. The bank was exploring ways to improve efficiencies through automation. The bank’s goal was to find an appropriate extraction tool to help automate the risk assessment process and, provided the quality of results is sufficiently reliable, reduce the operational risks of wrong covenant assessments. The bank's VP Strategy & Emerging Technologies, Martha, was responsible for identifying areas where automation can improve process efficiencies in the organization. She continuously scouts new technologies to keep abreast of developments on the market and conducts internal interviews with line-of-business owners to assess optimization potentials. One of the areas identified for improvement was the wholesale credit department, headed by Rick, whose team spent a lot of time thoroughly reviewing financial information in order to assess credit risks. Read More
Solution	The bank leveraged the meaning-based capabilities of Cortical.io SemanticPro to extract key information from credit agreements and automate the classification of covenants. After having looked at the different solutions on the market, Martha decided to test Cortical.io natural language processing capabilities and apply them to the analysis of credit agreements. The main reason why she kicked-off a project with Cortical. io was the easiness and speed with which the system could be adapted to Rick’s use case: only a limited amount of example credit agreements (less than 50) was needed for training the system. This greatly limited both the initial investment and overall risk of this project. After an initial base knowledge phase where the system had learnt from relevant source material in an unsupervised manner, two loan officers from Rick’s team provided input by reviewing a small set of documents. After that, the system kept incorporating feedback and created more accurate models (continuous learning, aka human in the loop). Read More Log in to view content
Contents

Technology Category

Analytics & Modeling - Natural Language Processing (NLP)

Applicable Industries

Finance & Insurance

Applicable Functions

Business Operation

Use Cases

Fraud Detection

Regulatory Compliance Monitoring

Services

Data Science Services

Challenge

The bank was looking for ways to improve efficiencies through automation, particularly in the area of credit risk assessment. The complexity in language and structure of wholesale credit agreements made automation difficult. These agreements are complex, 100+-page-documents that are very bespoke because banks use different lending systems. Especially covenants use highly specific, use-case related formulations. Some extractions are straightforward, and the values can be extracted “as is” (no inference necessary), for example: named dates, commitments, rates, parties. Others are more challenging and require fine-grained distinctions between options or covenants (eg redeem deal options, termination options). Here an automation system has to comprehend domain logic based on document structure and relations of extracted items. Pricing-related tables contain many different parameters (applicable margin, interest payment schedules, commitment tables) and it is extremely difficult to automate the extraction of pricing information. This is why the review of credit agreements was still done manually, which cost a lot of time and money to the bank, not mentioning the high error rate of such a mundane task.

About Customer

The customer is a Tier 1 bank in the US. The bank was exploring ways to improve efficiencies through automation. The bank’s goal was to find an appropriate extraction tool to help automate the risk assessment process and, provided the quality of results is sufficiently reliable, reduce the operational risks of wrong covenant assessments. The bank's VP Strategy & Emerging Technologies, Martha, was responsible for identifying areas where automation can improve process efficiencies in the organization. She continuously scouts new technologies to keep abreast of developments on the market and conducts internal interviews with line-of-business owners to assess optimization potentials. One of the areas identified for improvement was the wholesale credit department, headed by Rick, whose team spent a lot of time thoroughly reviewing financial information in order to assess credit risks.

Solution

The bank leveraged the meaning-based capabilities of Cortical.io SemanticPro to extract key information from credit agreements and automate the classification of covenants. After having looked at the different solutions on the market, Martha decided to test Cortical.io natural language processing capabilities and apply them to the analysis of credit agreements. The main reason why she kicked-off a project with Cortical. io was the easiness and speed with which the system could be adapted to Rick’s use case: only a limited amount of example credit agreements (less than 50) was needed for training the system. This greatly limited both the initial investment and overall risk of this project. After an initial base knowledge phase where the system had learnt from relevant source material in an unsupervised manner, two loan officers from Rick’s team provided input by reviewing a small set of documents. After that, the system kept incorporating feedback and created more accurate models (continuous learning, aka human in the loop).

Impact #1	The bank was able to extract the covenant structure out of loan agreements at the press of a button.
Impact #2	The bank got a much better and quicker appreciation of credit risk for their wholesale credit portfolio.
Impact #3	The bank was able to predict workflows and capital allocation for different business lines.

Impact #1

The bank was able to extract the covenant structure out of loan agreements at the press of a button.

Impact #2

The bank got a much better and quicker appreciation of credit risk for their wholesale credit portfolio.

Impact #3

The bank was able to predict workflows and capital allocation for different business lines.

Benefit #1	Saved thousands of hours of manual labor by automating credit agreement analysis

Benefit #1

Saved thousands of hours of manual labor by automating credit agreement analysis

Download PDF Version

Overview

How a large bank saved thousands of hours of manual labor by automating credit agreement analysis

Operational Impact

Quantitative Benefit