Page 38 - Understandinging Forensic Technology Landscape

AI products leveraging NLP, such as the tools in IBM certain words in a PDF file). Language processing
Watson, now help attorneys draft legal documents and understands the identified text and manipulates that
respond to inquiries. These tools are far more advanced information for different purposes.
and useful than traditional keyword searching because
they allow the use of plain English searches and, Imagine a scenario where you are told to identify a pool of
consequently, provide highly relevant and sufficiently contract documents and ascertain consistency or points
technical responses. of disparity in redundant clauses between contracts.
Applying automation could assist in multiple ways:
AI models are used extensively in fraud prevention and • RPA solutions can be used to automate redundant
detection, particularly in high volume transaction flows. search techniques to find contracts, such as pulling
For example, AI is now employed to scan customer down contracts from web-based interfaces.
and billing data to detect fraud and automatically block
transactions without human intervention. Similarly, • OCR technology (typically built into RPA solutions)
companies are increasingly using ML to automatically can convert contract images to readable text via OCR,
identify high risk transactions connected to bribery and allowing for searchable text listings to aid in contract
corruption or flag miscategorized transactions. analysis.
• Running a small sample of contracts through a
Getting started language processing module can teach the solution to
Today’s RPA and AI solutions are typically licensed identify clauses by certain keywords or key phrases.
software products configured for specific purposes, so The solution can then be run against the full pool of
they require an upfront investment of time and money. A contracts to pull out relevant clauses and generate the
cost-benefit analysis is recommended to determine the master template for further analysis.
best solution for a user’s needs.
• Integrating this contract parsing exercise into an
RPA is often a great entry point as a low cost, low effort ML module would allow the software to present its
solution. Automating commonly performed, day-to-day analysis of the contracts (for example, a certain clause
tasks demonstrates its value quickly. Learning RPA will is trending a direction through time or is not present in
also show its limitations and highlight the evolving ways a certain subpopulation of contracts).
that practitioners add value in a more automated world.
Multiple companies present solutions that function with Getting started with ML requires training and experience
easily understandable user interfaces and full training in data analytics and computer programming.
courses via their company websites. Practitioners should develop a skill set in working with
relational databases and learn a programming language
RPA solutions come in many varieties. While experience such as R or Python. Python is the most common
with coding languages can be helpful (and sometimes language used for ML applications. Next, a student of
necessary) for set up and implementation, graphical user ML should learn the theory behind ML algorithms to
interfaces (programming wizards), code libraries and better understand data types, bias, learning models, and
cloud processing tools make such tools more accessible. how to interpret the results. Finally, ML platforms (such
AI Multiple has a useful list comparing vendors. as Amazon AWS, IBM Watson, and Google Tensorflow)
41
all have distinct features and methodologies, so one
When taking the next step into AI, NLP is the best first will need specific familiarization with whatever tool
step into the technology. RPA solutions generally use they are using.
OCR to identify text in unstructured data and convert to
readable text for various use cases (for example, finding

41 RPA Tools & Vendors: In-Depth vendor selection Guide (2020), AI Multiple, blog.aimultiple.com/robotic-process-automation-rpa-vendors-comparison/, accessed November 26, 2019

Understanding the forensic technology landscape | 34