Fri Jan 27, 2023 3:31 pm Bankruptcy Filings
5 Ways ML And SME Collaboration Can Accelerate Innovation
This article orginally appeared on forbes.com
When it comes to legal tech, the concept of AI has gained acceptance as initial concerns about “robot lawyers” replacing skilled litigators, or strategic negotiators have largely waned over the years. Key factors in a machine learning (ML) based solution’s overall success include the diligence followed in its development and the oversight employed by the humans who build and train it. As a disclosure, my own company, Reorg, uses ML to power our suite of credit intelligence, data and analytics products which are used by financial and legal professionals at leading investment banks, law firms, hedge funds and corporations. In this article, I will lay out five ways in which partnerships between data scientists and subject matter experts (SMEs) can accelerate innovation.
- Enjoy greater efficiency by leveraging natural language processing.
One of the most challenging assignments faced by our team of covenant analysts at Reorg is the distillation of an Offering Memorandum (OM) into slick summaries for our legal, buyside and leveraged finance subscribers to consume. OMs are typically hundreds of pages long and filled with complex descriptions of high-yield bond terms and financial information.
Our SMEs partnered with our data scientists to develop a model that compares new OMs with all U.S. and European high-yield bond offerings since 2020 contained in Reorg’s library. This bond similarity tool is able to produce a list of bonds that are most similar to the new OM being analyzed and provide a numerical “grade” to express that similarity.
Reading through and synthesizing 400+ pages of an OM is a time-consuming task that requires concentration and fastidiousness. It could easily take an expert analyst several hours to ingest and interpret an OM. Using the natural language processing methods employed by our data science team, the bond similarity tool can locate the description of senior debt and senior secured debt notes, identify all subsections and financial covenants, extract the relevant sections and calculate a similarity score, typically within 10 minutes or less.
This helps our analysts and clients quickly surface OMs that they can rely on to understand market trends and anticipate any changes that may occur between the preliminary and final stages of bond issuance.
- Witness the network effects of building at scale.
Models that can process large sums of data can quickly expand the scope of available information at an exponential rate. We at Reorg were able to increase our universe of OMs to nearly 1,000 in a matter of months. When the “control” dataset is smaller, it is challenging to differentiate dominant patterns from coincidental ones. Our bond library now sits in the 1,200 range, and as this denominator increases, we are able to better determine “what’s market” when it comes to specific provisions or industry trends.
For example, with the increased number of OMs available, we are able to isolate specific drafting language preferred by individual sponsor private equity firms. We can also make connections between those sponsors and the law firms they hire. As a result, once a new OM is announced, we can quickly ascertain how certain sub-sections might appear, given the sponsor and law firm attached to the deal. We can also compare those provisions to highly similar provisions in other OMs and anticipate the degree of pushback the new deal may face.
- Improve accuracy through the exchange of knowledge.
In our case, we have recognized how crucial the role of SMEs is in developing the model to verify raw data, as they help our data scientists both understand the meaning behind our data as well as evaluate its usefulness. SMEs can suss out nuances in language and the importance of phrases that might not be readily apparent to a data scientist. For example, experts are regularly called upon to develop a list of aliases to ensure that edge cases are captured by a model.
An AI model built on higher-valued data will have more accurate results. SMEs can identify key deal provisions, consider the controlling jurisdiction of a drafter or isolate components that should be ignored. For example, our bond similarity model weights subsections that are riskier for creditors above those containing merely boilerplate language because SMEs highlighted the provisions of greater value when collaborating with data scientists.
- Accept the upfront investment and save in the long run.
Developing data science models can be time-consuming and resource-draining. However, it is important to remember that the hours spent are a one-time cost that can soon be reimbursed by the hours saved in manual labor.
Additionally, depending on the type of model and feedback loop in action, the outputs can improve and evolve over time. Done well, the increased velocity and accuracy created by employing a data science solution should outweigh the original fixed costs associated with its inception.
- Develop a whole greater than the sum of its parts.
A data science model built in the absence of expert oversight will be clunky and likely rife with errors. Conversely, a solution operated manually by an individual without the assistance of a technologist will be glacially slow and challenging to accomplish at scale.
An aligned, collaborative effort from both the data scientist’s tools and the SME’s perspective will overcome these difficulties and arrive at superior outcomes. By discussing the overall goals of the project at the outset, SMEs and data scientists can propose and test hypotheses that neither may have thought of alone. They can collaborate on best practices for achieving data cleanliness. They can also institute a common vocabulary and employ constant communication to ensure that objectives are aligned.
Data scientists are technical experts with a deep understanding of how to develop AI solutions, while SMEs are practice-specific experts who appreciate the utilitarian applications of those solutions. The combination of these skill sets serves to generate outcomes that can improve accuracy, save time and drive innovation.
Follow me on LinkedIn.