Company: Bank Of America
Company Description: Bank of America is one of the world’s leading financial institutions, serving individual consumers, small and middle-market businesses and large corporations with a full range of banking, investing, asset management and other financial and risk management products and services.
Nomination Category: New Product & Service Categories - Business Technology
Nomination Sub Category: Big Data Solution
 Nomination Title: Bank Of America's Sherlock
- Which will you submit for your nomination in this category, a video of up to five (5) minutes in length about the nominated new or new-version product or service, OR written answers to the questions for this category? (Choose one):
Written answers to the questions
- If you are submitting a video of up to five (5) minutes in length, provide the URL of the nominated video here, OR attach it to your entry via the "Add Attachments, Videos, or Links to This Entry" link above, through which you may also upload a copy of your video.
- If you are providing written answers for your submission, you must provide an answer to this first question: If this is a brand-new product, state the date on which it was released. If this is a new version of an existing product, state the date on which the update was released:
- 2020: August – Proof of Concept completed
- 2020: September – Application & Model Development Begins
- 2021: November – Model Risk Management (MRM) Approval
- 2021: December – Production Release
- If you are providing written answers for your submission, you must provide an answer to this second question: Describe the features, functions, and benefits of the nominated product or service (up to 350 words):
Total 312 words used.
Sherlock is a Machine Learning (ML) application linking incidents to change requests stored in the bank’s system of record. Sherlock was created to reduce Mean Time to Restore (MTTR) to help the bank recover from business-impacting incidents faster. After its release, Sherlock reduced the amount of time to identify the root cause of a change-related incident from 90 minutes to about 10 minutes.
Sherlock leverages ML and Natural Language Processing (NLP) in combination with Graph Database technology to develop a solution capable of rendering next generation infrastructure blueprints that automatically detect component-level interactions correlated to business and environment data. This provides operations staff with the ability to quickly isolate root cause and reduce MTTR by a factor of 10.
Previously, Incident Management functions were highly manual and involved a triaged question and answer process.
The Sherlock team leveraged model development data from six months of incidents caused by application changes. The team also built an application dependency graph illustrating the relationship between Application Inventory Tools (AITs) and their Configuration Items (CIs).
Each incident record contains a free form text description of the incident, the incident date, a list of systems affected, and a description of the change request and date. Using NLP, Sherlock processes the free form text, clusters these incidents via density-based spatial clustering with noise (DBSCAN), and contextual text similarity. The resulting clusters build linkage between similar historical incidents and the causal change request.
Given the new incident details which are in progress, the above model is used to identify and rank the causal changes and leverage this knowledge to provide the user with a list of top ten change requests.
A patent has been granted for dependency mapping design and root cause analysis. The patent is in use today and lays the groundwork for the future vision to auto remediate incidents caused by a change.
https://patents.google.com/patent/US11132249B1
- If you are providing written answers for your submission, you must provide an answer to this third question: Outline the market performance, critical reception, and customer satisfaction with the product or service to date. State monetary or unit sales figures to date, if possible, and how they compare to expectations or past performance. Provide links to laudatory product or service reviews. Include some customer testimonials, if applicable (up to 350 words):
Total 170 words used.
- Decrease in MTTR (90min – 10min).
- In 2021, more than 1,500 hours of Productive Capacity were added.
Sherlock is continually updated based on user feedback to improve the quality of the data presented, efficiency in the use of screen space, and overall user experience. Additional data are being added to the search form to further increase the precision of the results. Through Q1 2023, the team is focused on building a new proactive view within Sherlock in addition to current reactive search. This provides insight into Change Requests that have a high likelihood of causing a business-impacting incident and can be corrected and/or monitored appropriately.
The Sherlock application reflects technical excellence, real world benefits and transformative teamwork. By utilizing the latest machine learning technology, Sherlock has closed the gap in application down time by reducing the amount of time it takes to identify business-impacting incidents. Sherlock is a key part of operational resiliency which continues to remain a critical focus of the firm and helps best serve our customers.
- You have the option to answer this final question: Reference any attachments of supporting materials throughout this nomination and how they provide evidence of the claims you have made in this nomination (up to 250 words):
Total 92 words used.
The below shows Quarterly Model Risk Management Ongoing Monitoring accuracy metrics for the past four quarters. For example: In Q1 ’22, out of all incidents caused by change, 80.6% of the causing changes were identified in the top 10 suggested causes:
- Q1 ’22: 80.6%
- Q2 ’22: 69.9%
- Q3 ’22: 70.0%
- Q4 ’22: 70.1%
Usage:
- On average Sherlock receives 10 searches per day, with nearly 300 searches in January 2023.
- Top features according to end users include the lookback date range and the ability to filter by location.
| Attachments/Videos/Links: |
|---|
|
