Inspiration
Expenses reporting is often a time-consuming and repetitive task for both the finance team and the employee who is making the claim. First, the employee have to remember to keep the receipt [nearly half of the employee surveyed have lost their receipt (Reuters, 2014)] to make the claims before a certain period of time and also fill up the necessary supporting forms. Next, the finance team have to verify the claims and manually key in the data manually.
A recent survey gathered 500 employees with accounts payable duties found that on average, 1 in 5 of the expenses their coworkers submit are in reality non-work related expenses and takes them about 18 minutes to correct any mistakes in the report (Concur, 2017).
By automating and streamlining the whole expense management process and deploying algorithms to detect any abnormal expenses, it will reduce the hassle for both the finance team and their co-worker. Constant and real-time monitoring of the company’s expenses is also made possible to deter any frauds from happening.
What it does
We built Expense Integrated System (EIS) to solve this issue. By using OCR technology, employees can scan their paper receipt and the chatbot will convert it into readable data to be inputted into the system. Next, the data is verify for credibility by the system based on the user profile, the invoice issuing company profile and if there is any abnormality with transaction. A composite score and expense analysis report will then be generated for the account payable officer can quickly identify any abnormal expenses.
How we built it
We used Python Django for web development, and used Telegram Bot platform to create a chatbot. From this, we use a combination of Tesseract, Optical Character Recognition engine, and python library to help identify the paper receipt and generate a composite score for the expenses by looking its various characteristics (user, company, transaction type). Bootstraps is used for front-end development.
Challenges we ran into
It was a real struggle to build and sync the whole end-to-end system and interface in the short time span. At times, we suffer from scope and have deviate from our core requirements.
Accomplishments that we're proud of
Being new to OCR technology and python language, we are glad that we are able to work out the OCR engine and integrate into the telegram interface.
What we learned
Perseverance is key to any success. The debugging process was quite tedious and we have to constantly recheck our logic flow.
What's next for EIS
Currently, the accuracy of the OCR is at around 60%. We would like to increase the accuracy of the OCR by looking into methods such as processing the image. Data visualization is also another area we will like to visit to improve monitoring of the expenses.
Log in or sign up for Devpost to join the conversation.