Autodit

Inspiration

According to AEcensus 2016/17 report, audit business units generate the highest revenue compared to other business sectors. However, it remains at a lower level for productivity. The main reason for low productivity is that the audit process contains manual and time-consuming tasks like vouching. Our team wants to use technology to automate the vouching process, so auditors can save time, reduce error and focus on higher-value work.

What it does

We have developed Autodit, an application to automate the auditing process. Auditors have to just simply upload the source documents and Autodit will generate a report with relevant entries (usually manually typed by auditors) and also shows if there are any discrepancies.

How we built it

We have used TAIGER’s Optical Character Recognition (OCR) API to digitalize the uploaded source documents. The key-feature detection algorithm is written in python to capture key features from the digitalized documents. Key-features are processed and report is generated after checking completeness, occurrence, cutoff, accuracy, classification, authorization.

Challenges we ran into

Our team had limited background knowledge in auditing, especially about the vouching process. Our initial prototype was very much far away from automating the vouching process. After having extensive multiple rounds of discussions with our assigned Deloitte mentors and fine-tuning our prototype idea, we were able to simulate the vouching process in our prototype. Another major challenge was the limited number of source documents. We need a set of source documents like purchase order, invoice and delivery order to train the Autodit.

Accomplishments that we're proud of

Despite having limited knowledge in the vouching process and limited resources to train the model. We still manage to develop a working prototype within a short span of time.

What we learned

We learned that a mistake in auditing may cause big penalties, therefore our solutions should not only solve some problems but also has to be highly accurate.

What's next for Autodit

We have planned to include an auto search function in Autodit, where the auditor has to upload all available documents and Autodit will smartly categorize the relevant source documents for each transaction. We also like to test Autodit on a real client-side using actual source documents.

Built With

• machine-learning
• python
• taiger-ocr