Our 3-man team aspires to work in the auditing industry and had pursued internships with several auditing firms to gain relevant experience and industry insights. However, the tedious process of vouching left us daunted. In the age of technology, we firmly believe digitalising and automation is the future to streamline the auditing process to achieve high level of coverage and assurance with minimal effort. It is further elaborated by industry professionals and cabinet minister Indranee Rajah in an ACCA conference: “Technology disruptions will be painful and even bewildering in the short-run. Jobs may be lost or destroyed with advancing technologies. Routine and mundane work that requires lower level of skillset will be replaced”.
Vouching is fundamental to the auditing process, to verify books of accounts by examining supporting documents. However, it is often tedious and time-consuming. The precious time spent during the engagement period could be used to provide analytical insights which is increasingly demanded by clients of the auditing firm. Moving into an era where artificial intelligence and automation is prevalent, the financial landscape will be revolutionised.
What it does
Thus, Vulture is born. We conceptualised Vulture to be an extension to Auditing/Accounting platforms, where output generated by Vulture will be seamlessly integrated and uploaded to the platforms. Machine learning is adopted to convert source documents through scanning of physical copies or documents that are in digital form into a standardised format, both of which are recognisable by Vulture. AI is implemented to intelligently retrieve information of the digital copies and do a 3-way matching between documents. Hence, audit can be forward looking and done perpetually rather than at the end of the year and 100% coverage could ideally be achieved. Discrepancies can be identified early (even on a daily basis) and be sent as queries to clients.
How we built it
We developed our web backend and frontend by using Python and Flask as the web framework complemented with HTML, CSS and JS. This was hosted on PaaS platform PythonAnywhere. Our app prototype is built using Ionic Framework that allows for rapid development and deployment. You may preview our app by downloading Ionic Creator on Play Store or App Store and then keying in our preview code “E3598F” without quotes. We have also tested out Optical Character Recognition (OCR) to allow text recognition and entity recognition as a proof of concept so that Vulture can smartly identify which are the static and dynamic fields and data from both hardcopy and softcopy invoices. This technology can help to orientate photographs taken from invoices which may not be of the orientation as the same plane as the smartphone camera, thereby enabling us to warp the image to an ideal orientation. Additionally, the records are also committed to a general ledger that is made available securely to the auditors and clients as well. This serves as a transparent, time-stamped and consensus-based database of the captured source documents and becomes a distributed ledger transparent for stakeholders to view. Down the road, machine learning will also be further implemented in the actual development stage to create the three-way matching capabilities of various source documents. For example, unsupervised clustering can be used to first segment the various text fields into static and dynamic data, and then the machine can further use deep learning to identify patterns within source documents as well. Further, this artificial intelligence will allow Vulture to smartly classify and match the remaining source documents and also suggest the best person to contact lest the auditor have to make an enquiry on the integrity of the document – a common scenario that plays out during vouching.
From the audit firm’s perspective, the auditor will only have to educate the client on the use of Vulture and resolve discrepancies that are identified by raising to the clients. No physical vouching will be required as all processes and tracing are automated. Instead, a test of IT controls will be required. From the client’s perspective, the client only has to a picture with Vulture Scan and a standardised formatting of the document will be generated. It enables consistency throughout all documents received and ease of detecting errors. Perpetual auditing can hence be adopted to provide insights to the completeness and accuracy of documents kept.
Challenges we ran into
We wanted to build a prototype which would simulate the auditing functions of Vulture and thus we sought to create one which was as close to our envisioned product as possible. That meant spending a considerable amount of time in developing our web backend and frontend, hosting our web app on a web server, developing the app prototype and also carrying out testing on implementing Optical Character Recognition (OCR). Trying to create all the various functions also involved us having extensive discussions with our KPMG mentors who have rendered us a great deal of insight into the auditing scene, as well as further brainstorming among ourselves in trying to craft a prototype as relevant and real as possible along with other crucial aspects to think through such as creation of new controls via the segregation of new roles and responsibilities for the different stakeholders such as the accountants and auditors and the development of a new workflow process for vouching and beyond as well. Lastly, training and testing out the Optical Character Recognition (OCR) technology and cogitating the algorithms for extracting information from physical invoices was also a challenge due to the tight timeline, especially where inconsistent handwriting or badly taken photographs could be difficult to be converted into our standardised version.
What's next for Vulture
We plan to have Vulture to encompass all routine audit functions that are required of an auditor. With a photo taken of a physical asset, Vulture will be able to retrieve information and derive valuation figures with a touch of the screen. Vulture will be able to integrate with external API sources. Similarly, audit of foreign currency transaction will be compared with reliable external sources (Oanda) on Vulture. Data captured with Vulture scan will allow the comparison of exchange rates used in the transaction to the prevailing rate as of the date. Further documents can be processed such as bank statements to allow other functions.