How AI impacts forensic accounting

August 2024  |  SPOTLIGHT | FRAUD & CORRUPTION

Financier Worldwide Magazine

August 2024 Issue

Like many industries, forensic accounting is an area where artificial intelligence (AI) has the capability to make a significant impact. The field is a niche area within accounting that investigates firms or individuals thought to have committed fraud. The work often requires analysing vast amounts of data to extract relevant information, which is where AI can make a tremendous impact.

Within AI there are various subsets such as generative AI (GenAI), machine learning and natural language processing (NLP) that further analyse data and even have the ability to predict fraud. Although AI is a new and continuously developing field, several companies have already taken the initiative by incorporating it into daily practices and investigation methods, and are starting to reap the rewards.

AI allows for more efficient detection, reduced manual review time, cost effectiveness and the ability to work 24/7. It can also be applied to a whole host of new anti-fraud technologies to enhance their abilities. Firms will need to stay on top of new AI developments and invest in their own AI models in order to stay competitive and fulfil their clients’ needs.

However, AI is rapidly changing and it is unclear exactly what effects it will have on the forensic accounting industry. In just a year there have been enormous steps in innovation – for instance, significant advancement in NLP where models can now understand and generate human-like text. A common example is ChatGPT, which is supported by GenAI technology. Users can have conversations and get assistance from the chatbot.

Advancing forensic accounting with AI

Emotional AI. Part of a current forensic accountant’s job is to gauge the mannerisms of an individual during an interview, and now there is a new AI tool to help. Emotional AI detects and interprets human emotions through software that analyses facial expressions, voice inflection, body language or other nonverbal cues to assess the sentiment of an individual or group.

However, this software can be controversial, not only for its use of facial data but also in AI’s ability to ascertain human emotions and behaviour. We recognise that emotions can be expressed differently based on the person and his or her culture, among other things. Therefore, when applying an emotional AI model, one must consider reprogramming the model based on different languages or cultures to ensure accurate results.

Network visualisation. Another key technology is network visualisation. For forensic accountants, visualising a network of actors and their connections is instrumental in detecting fraud. Beyond solely looking at transactions, network visualisation gathers data from payment methods, addresses, account numbers and the customer’s information. This allows the technology to draw connections between different actors and present the links in a visual way that is easy for investigators to analyse.

Network visualisation can also help narrow down who committed the crime or flag if a user is connected to a known fraudster. Moreover, to prevent overloading the network with links, network visualisation should be applied to a smaller data set. While fraud is not automatically detected, network visualisation can be a useful tool as it can point out anomalies or risky connections.

Text mining. Text mining can analyse both structured and unstructured data and then present its findings in a way that is easily understandable. It combines other technology like information retrieval, web mining, data extraction, computational linguistics and linguistic processing. The software can be applied to pdfs, websites, emails, social media, online chats and text to email messages – all forms of unstructured data that are used in investigations.

Many current analytical models rely on structured data fields, through including unstructured data, we can increase efficiency, particularly in areas like integrity due diligence (IDD) where background checks are crucial to the investigation. IDD includes accessing information publicly available, such as social media, to get the sense of who an individual is. Through using text mining to analyse this information, investigations can be less time intensive.

A frequent application of text mining is variable extraction. This entails using certain variables as indicators. For example, seeing how often the word ‘football’ appears to gauge the sport’s popularity. The software can also be used to search for patterns that suggest potential fraud. However, it is important to remember the model’s language detection components if it is going to be applied to data in another language. The technology must be adapted to different languages and cultural connotations in order to accurately retrieve data and present its findings.

Predictive modelling. With the advances in new technology, we are now able to predict future fraud using software called predictive modelling. This model uses historical data to create statistical models that predict where and when fraud will occur in the future as well as identifying fraud patterns. The software is trained on a large, unbiased dataset that includes instances of fraud so the model can learn fraudulent behaviour. This includes a longer upfront training time as the data must be refined and the model must be repeatedly tested to ensure accurate predictions.

After initial training, the model learns and adjust its predictions as new data is added. Predictive modelling is especially useful since new ways of committing fraud are constantly emerging. The model learns to recognise a new technique and start flagging it. Beyond proactive fraud detection, the model can also identify risk areas and help businesses protect their reputations.

While this technology is a useful tool, it is imperative that it has routine updates in order to preserve accuracy. Further, compared to other types of fraud detection, there are higher rates of false positives and it will require manual oversight for the foreseeable future.

Conclusion

Despite uncertainty about what specific role AI will play in forensic accounting, we have no doubt it will be integrated into the industry. Companies need to embrace AI technology in order to stay competitive and deliver the best results for clients. Although AI is still in its early stages and still requires human oversight, professional service firms and corporates alike should start investing in it now to facilitate compliance and forensic accounting strategies.

Dr Tim Klatte is a partner at Grant Thornton Shanghai. He can be contacted on +86 (21) 2322 0580 or by email: tim.klatte@cn.gt.com.

© Financier Worldwide

BY

Tim Klatte

Grant Thornton Shanghai