Detection of Weak but Deceptive Signals from Structured and Unstructured Data [Project completed in September 2023]
This research project was successfully completed by Judicaël Poumay, who defended his doctoral thesis in September 2023.
The Context
AI and related technologies are expected to have significant ramifications for society in general, and is foreseen to bring about disruptions to classical professions, such as medical practioners, lawyers and auditors. For the latter, the business is based on the analysis of many sources of documents, which are internal or external to the audited companies, such as accounting documents for instance. These data sources provide insights into business risks, enabling auditors to highlight potential crucial issues. Furthermore, the data sources are relatively structured with clearly defined fields and columns, from which the required information can be easily extracted.
The Challenge
Most auditors overlook other pertinent data sources, such as press releases, web pages, whitepapers and social networks and media (e.g., tweets). These sources are a valuable repository of potentially pertinent information, which could complement the traditional auditing documents. Taking these complementary sources of information could lead to a more accurate and efficient auditing process.
However, data sources, such as press releases and social networks are less structured, and their analysis requires novel methods. Among others, natural language processing (NLP) and social network analysis (SNA) can help in accelerating the analysis process. Furthermore, these data sources abound in “weak signals”, which are information that is not overtly expressed, but is “deceptive” in nature. However, they could be indicative of frauds and management errors.
Key Question & Goals
Based on the aforementioned elements, the key question here can be formulated as follows:
“How to accurately detect weak signals from huge volumes of structured and less structured data in the context of an audit process?”
The main goals of the research project will be to:
Develop and implement novel algorithms for automatically detecting these weak signals from various sources of structured and less structured data;
Develop innovative approaches to incorporate these weak signals with other pertinent information obtained from traditional accounting documents;
Investigate whether these weak signals enable the detection of mismanagement cases, which were not possible with information available from traditional accounting documents;
Determining which types of weak signals are effectively indicative of mismanagement.
About KPMG
KPMG is a one of the largest auditing companies operating in 152 countries with more than 189,000 people working in member firms around the world. With a one stop one shop approach, their clients are advised by multidisciplinary teams of high performing people using their expertise and insight to give clients business advice and to guide them through the complexity of today’s local and international issues and regulations. In addition to tax, legal, audit and advisory services, KPMG in Belgium and Luxembourg offers specific advice to family owned business.
Being at the forefront of the digital revolution, KPMG is interested in investigating novel AI approaches, based on machine learning and deep learning, to enhance the auditors’ business. This is the reason why KPMG decided to become project partner of the HEC Digital Lab.
