Speed up a document categorising and auditing process
Implement a machine learning backed optical character recognition solution
A scalable, robust document scanning system saving countless man hours
TraveliT provides travel booking and expense management solutions. Their website and mobile application allows users to make bookings for flights, cars, accommodation etc. and facilitates the logging and tracking of all travel-related expenses.
TraveliT employs roughly 300 employees to handle the auditing of expense claims submitted to them through their platforms. Each document submitted with an expense claim needs to be manually categorized and assessed manually. To keep up with their growing customer base, TraveliT felt the need to supplement their processes with a digital solution and alleviate the mounting pressure on their auditing staff.
Their goal was to incorporate a scanning solution into their auditing process that would automatically categorize and analyse each document submitted through their expense claim platform.
DotModus was chosen to assist TraveliT in achieving this goal.
Users submit photographs of their receipts or invoices through TraveliT application/website as part of their expense claim procedure. Each uploaded photo is stored in Cloud Storage and the data extraction process begins. AutoML analyses each submitted image and determines if it is a photo of a receipt, an invoice or something else. Anything the system can’t identify will be checked manually by one of TraveliT’s staff. AutoML also checks the database for any existing versions of the provided photo. If an existing image is found, it is flagged as a duplicate. Duplicate images usually arise from employees within the same company making claims for the same expense, so they are particularly important to monitor.
The system then commences with the relevant extraction process for each type of document. Each image is run through Vision API which uses Optical Character Recognition (OCR) to convert the text found therein to machine-readable data. Rule based logic is setup to spot relevant words/fields such as “Supplier” or “Total” and extract the values associated with them.
A company can also request that invoices or receipts containing specific words or phrases are flagged. For example, if a supporting document is found to contain wording that does not fit the company’s expense claims policy (e.g. “Proforma Invoice” or “tequila”) then that document can be flagged for investigation by an auditor at a later stage.
Approximately 2,000 claims are processed daily through the Travelit system, with new customers being loaded daily. TraveliT can rest easy knowing that meeting the demands of their growing user base is merely a case of scaling up their infrastructure, and not doubling their workforce. The solution has greatly increased productivity and freed up man hours across the board at TraveliT.