Case Studies

Customer Challenge

A leading global bank had multiple repositories with similar data. Disparate systems would connect to different inconsistent datasets for business reporting. Multiple copies of the same data or missing data would further complicate data management across the bank and result into inaccurate reporting. Automated data quality improvement was not possible through non-AI systems due to inconsistent attribute naming and required checks for regulatory compliances.

Scry’s Solution

Implemented ScrySSoT® (Single Source of Truth) Engines that use proprietary machine learning algorithms to identify and rectify data disparities.
  • Data Quality (DQ): Bring data from multiple sources into a single data lake using pre-built and custom connectors. Automated data cleansing and harmonization to achieve data accuracy of 95%.
  • Business Rules (BR): Enforced business rules to audit data compliance with internal and external regulations.
  • Collation: Implemented business dashboards and descriptive analytics on cleansed and harmonized data.

Business Impact

Accurate and consistent business dashboards for business and executive users. Improved reporting led to:
  • Correlated data sets offering broader view of business performance without requiring tedious manual effort
  • Reduced exposure to loan defaults through identification of anomalies indicating customer delinquency and customer behavior shifts
  • Higher products portfolio performance with predictive insights on product sales by geography and customer segments

Customer Challenge

A global bank with SMB and consumer borrowers found its customer acquisition cost increasing owing to lower approval rates and high manual effort for each application. Bank relationship managers would need to manually evaluate cash flows and other financial statements to establish creditworthiness of its customers. Non-AI based automation in loan approval would not include unstructured business performance indicators found in external data sources. Also, external structured data was difficult to integrate with internal datasets due to non standard attribute identifiers.
Scry’s Solution
  • Implemented AI based products to automate prospecting process. ScrySSoT® uses Natural Language Processing (NLP) algorithms to collect and analyze customer sentiment insights from Yelp reviews, feedback on social media.
  • Deployed ScryCash® and ScryCredit® to identify anomalies in cash flows and credit card transactions.
  • ScryProspect® captured and analyzed the dynamics and network effect among customers and their business partners using Scry’s graph analysis algorithms and databases

 

Business Impact

  • Reduced false negatives (customers with low credit rating scores but strong cash flow trends) and false positives (customers with strong credit ratings but weak cash flow trends) and increased loan approval rates increasing interest income
  • Reduced manual loan processing effort and duration
  • Increased breadth of data used in credit decisions

Customer Challenge

In insurance business, when a law firm fills in an invoice and submits it to its client, a reviewer on client’s side reads and validates entries with respect to Legal Service Agreement (LSA). This process is manual, time consuming and error prone. There are no insights from past similar invoices to expedite the process. Also, it is difficult to assess line items individually leading to erroneous payouts of items that should have been rejected or adjusted.

Scry’s Solution

Implement ScryAudit® to validate invoices against agreements. The solution offers:
  • Unsupervised Clustering and NLP algorithms - Identified and extracted important attributes from unstructured text and categorized narratives into categories and subcategories
  • Classification using Deep Learning algorithms - Automated approval of expenses and adjustments to narratives based on past approvals, LSA guidelines and NLP engine

 

Business Impact

  • Identify and reduce erroneous payments - Analysis on a small sample set of 4000 of flagged line items from expenses revealed approximately $70,000 could have been saved by avoiding human error.
  • Summarized extraction and improved efficiency - A structured summary of the expense, much easier and faster to review, leads to prediction of approval of expense with 92%+ accuracy and adjustment of narratives for the data with 83% accuracy