AI and the Auditor: A New Era for Financial Document Analysis
03.04.2024, by iconicchain
It’s not a secret anymore that AI and Machine Learning (ML) are taking over the world. Providing a huge space for fun and creativity (of course, we are all eagerly waiting for ChatGPT-powered toasters and microwaves) they are also rapidly entering the professional life of people.
Are AI models mostly just fancy toys, or can they indeed be useful for real business and increase the productivity and efficiency of people? At iconicchain we do believe in the latter, so we decided to prove that ML and AI could be helpful even in areas with very high level of privacy and responsibility, such as audit.
An important part of an auditor’s workflow is extracting information from various data sources and comparing it with each other to ensure consistency and correctness. Thousands of invoices, fulfilment notes, bank statements and other types of financial data are being manually processed during the audit process, taking most of the time and attention span of the specialist. With our innovative iconicAudit platform we aim to address this issue with AI and make auditor’s work faster and more efficient by automating routine tasks and letting the specialist stay creative and keep focused on important things.
But why AI? Isn’t it suffering from a lack of privacy and requires incredible hardware to run? While the general answer to these questions is ‘yes’, there are still opportunities to overcome those issues and utilize the intelligence of AI models 100% securely and without top-level hardware. These opportunities are fully utilized in our platform – with iconicAudit you can run AI and ML models locally (without internet access) and with affordable hardware – because parameters (weights) of all models are compressed in a special way and therefore require significantly less space!
But how would these models help? The main goal of our AI assistant is to extract important information from various documents. An auditor specifies fields that he or she is interested in (total sum, issue date, fulfilment period, etc.) and the model helps to quickly obtain values of these fields from imported documents. Our models are trained to extract any type of data and the whole pipeline ensures zero hallucinations, so the extracted information is always based on the provided data. Moreover, we provide an interface to interact with your data in real-time – one can, for example, check if the data coming from two different sources match regardless of the format and language, or to get a short summary of a long contract. Worth mentioning that financial documents are very often represented as images. To “read” data from an image, optical character recognition (OCR) techniques are used. OCR is a process of converting images of typed, handwritten or printed text into machine-encoded text. It is a common method of digitizing printed texts so that they can be electronically edited, searched, stored more compactly, displayed online, and used in machine processes.
So, our pipeline allows:
- Run ML and AI models locally and 100% privately – your data never leaves your IT infrastructure.
- Utilize AI intelligence to extract important data, interact with your documents and summarize them. Works on various languages and any type or format of data – PDFs, texts and images are supported! Zero hallucinations are guaranteed for the extracted data.
- Affordable hardware requirements thanks to reduced model sizes.
- Possibility to adjust models to specific needs by using cost-efficient fine-tuning on your data.
We believe that our approach proves the possibility of including AI in the auditor’s everyday professional life and increasing the performance on monotonous and attention-greedy tasks and our belief is supported by customer feedback.
