Generally, a software can be described as artificially intelligent if a human sees its capabilities as human-like. The current state-of-the-art form of AI can be seen in machine learning, a technology that enables software to learn without being explicitly programmed.
According to the rule-based approach a programmer needs to “tell” the software where or how a certain information can be found or how it is expressed. The rule-based approach finds its limitations dealing with very individual document types, as the rules need constant adjustments.
The machine learning approach aims to teach a machine to identify information regardless of how the content is expressed or how the document is formatted. Machine learning software learns from ‘experience’. Furthermore, when the machine is confronted with unknown document formats and/or new content, it predicts a text’s meaning.
An accountant’s activities can be divided into creative tasks, the analysis of data, and rather uncreative tasks, such as the identification, recording and classifying of information in documents. However, the classification of these tasks as uncreative does not qualify them as easier than the creative ones as they require a high degree of concentration and can lead to (potentially costly) errors.
Furthermore, since the process of manual data extraction can be time- and cost-consuming, the scope of the data or the assessed contractual relationship is often restrained. This can lead to sub-optimal and costly decisions as the basis for an analysis and decision-making is incomplete.
AI software can currently identify and classify relevant documents and can also be trained to find, extract and process relevant documented data. Once the software is trained, it can extract hundreds of data points from documents – such as commercial real estate contracts – within a matter of seconds, working 24/7 and guaranteeing 100% data consistency. Therefore, the identification, classification and extraction of data from documents can be scaled almost infinitely – only being restricted by the processing power of a computer processing unit.
However, any piece of information needs to be considered in the context of an – often dynamic – legal environment and prepared according to a project’s and client’s specific demands. As current AI software predominately focuses on one source type, eg documents, it does not add much value to an accountant’s creative tasks, which need to be approached manually.
AI offers an enormous potential to automate an accountant’s uncreative tasks. However, it is still the (human) accountant’s responsibility to do the creative work – making sense of data and putting data into the context of its framework conditions.
Richard Belgrave, head of Europe, Leverton; Chris Biggs, managing director, Theta Financial Reporting