Data Science in Legal Operations Research Paper: “Detecting Anomalous Invoice Line Items in the Legal Case Lifecycle” – PR Web

A chart from the research presented by InfiniGlobe Research Labs at ICDM 2021 displaying outliers in UTBMS codes, identifying anomalies detected and presented for review.
InfiniGlobe Research Lab, a division of InfiniGlobe LLC, a legal software and consulting company, published an innovative research paper that explores how Machine Learning models can be trained on historical and synthetic invoice data to produce algorithms that detect anomalies in invoices, improve accuracy, and reduce the human effort and cost of invoice review.
“Detecting Anomalous Invoice Line Items in the Legal Case Lifecycle”, published 2020, (rev. 2021) was submitted, accepted, and presented at the 2021 Industrial Conference on Data Mining (ICDM) by co-authors Mori Kabiri, InfiniGlobe CEO, and Valentino Constantinou, published data-scientist. The video presentation of the paper and research is available here and an open-access copy of the paper is available here.
Kabiri who also chaired the conference, shared: “Having had the privilege to work with corporate legal departments across the spectrum on adopting and implementing technology, I should admit… we’re behind our peer industries in research and development.”
This research was inspired by the InfiniGlobe team’s experience with clients who needed help validating invoices against their billing guidelines, or agreed terms with law firms and vendors. Clients reported that at least 20% of their vendors exhibited “poor billing hygiene.”
“Corporate in-house lawyers should spend less time reviewing and approving invoices with confidence knowing automated intelligent review and adjustment has already occurred, ensuring compliance with billing guidelines.” — Mike Russell – Head of Global Legal Operations, Expedia group.
Most legal invoices are intricate, containing hundreds of line-items with detailed information. The opportunity for undetected discrepancies and the cost of review are exponentially greater than other industries, making this research particularly interesting to Legal Ops teams. Errors cost the company, resulting in over- or under-payment, additional manual administration, lost time trying to rectify errors, and even the potential for damaged relationships.
The research in this paper focuses on analyzing UTBMS codes, a legal eBilling coding standard, which was created in the 1990s to aid law firms in organizing and submitting invoices electronically. A UTBMS code would be assigned to each line item, corresponding to the type of activity and the phase of the case lifecycle during which that activity was performed. When correctly assigned, these are useful filters for anomalous work and aid in reducing manual review time. When misused due to lack of knowledge, time constraints, or other reasons, law departments are left with muddled data, which negates many of the afore-mentioned benefits.
“Properly used codes are very important for our legal and financial benchmarking, performance measurement, and resource utilization analytics.” – Connie Brenton – VP, Law, Technology and Operations, NetApp Inc.
Armed with this challenge, Constantinou and Kabiri partnered to employ advanced anomaly detection techniques trained against characteristics unique to law firm invoice submission. The goal of this research was to develop and train machine-learning algorithms to reduce the effort needed for invoice review and approval. The research focused on invoice line-items with UTBMS codes that fell outside of the typical distribution of services billed in the legal case lifecycle, or in other words, line items categorized with a UTBMS code out of order per the defined case activity chronology.
Singular Value Decomposition (SVD) and T-Distributed Stochastic Neighbor Embedding (T-SNE) were used to reduce the dimensionality of the test data. Subsequently, DBSCAN clustering was employed to select data that could be used to generate labeled synthetic data for model training. Model training was performed using the Random Forest, Gradient Boosted Tree, and Support Vector Machine (SVM) models, and feature selection took place through SME knowledge and Shapley Additive Explanations (SHAP). Models were assessed according to: recall, precision, F-1 Score, Accuracy, and coverage.
The ideas presented in this paper create a pathway for using AI technology in the detection and primary verification of anomalies in billing. InfiniGlobe Research Lab is looking to further develop the ideas introduced in this paper in order to build increasingly reliable automated detection/verification AI powered systems for the legal industry.
While this work is presented through its application to the legal industry, it may be applied more generally to other similar types of data and in use cases where data patterns of the anomalies are well-known, as well as to improve existing e-billing systems to detect this specific type of anomaly (lifecycle anomalies).
Beyond its applied impact, this published research is a first peek into the opportunities that can reveal themselves when applying legal business data to leading machine learning models. InfiniGlobe Research Lab has hinted they have other research plans in the work and invite innovators to join them in advancing legal through tech.
About InfiniGlobe
InfiniGlobe LLC is a software technology and consulting company headquartered in Newport Beach, California, offering a broad range of professional services and software solutions for the legal industry. With decades of experience working in legal technologies and a prominent reputation of consistently and passionately helping clients solve their problems, the InfiniGlobe team enables Corporate Legal Departments and Law Firms alike to overcome the challenges and complexities of technology through simple, intuitive design solutions. At InfiniGlobe, we don’t believe in the finite – in what just works; we believe in the infinite – in purpose, collaboration, and achievement. To learn more, please visit
Share article on social media or email:
View article via:
Questions about a news article you’ve read?
Reach out to the author: contact and available social following information is listed in the top-right of all news releases.
Questions about your PRWeb account or interested in learning more about our news services?
Call PRWeb:1-866-640-6397
©Copyright 1997-2015, Vocus PRW Holdings, LLC. Vocus, PRWeb, and Publicity Wire are trademarks or registered trademarks of Vocus, Inc. or Vocus PRW Holdings, LLC.

Connect with Chris Hood, a digital strategist that can help you with AI.

Leave a Reply

Your email address will not be published.

© 2022 AI Caosuo - Proudly powered by theme Octo