An expense report is a document, either digital or paper, prepared by employees to report expenses incurred in a work context. The creation of expense reports is a common practice in many companies, especially larger ones. This article outlines the expense report preparation process and its key steps, highlighting the challenges of manual processing and the advantages of automating the process.

The expense report process

The preparation of an expense report is necessary whenever employees incur costs that need to be reimbursed by the company. This is a delicate and complex process that typically involves phases such as the liquidation of actual and lump-sum expenses, submission of supporting documents, and more. Ensuring the accuracy and timeliness of reimbursements is crucial, making it essential to have a secure, reliable, and digitized process to avoid delays and errors.

In general, the expense report process allows companies to verify, account for, and reimburse these anticipated costs. It is a critical process for both companies and regulatory bodies, such as the Revenue Agency, as it contributes to the calculation of business income.

Here are some examples of expenses typically covered by reimbursement

• Room and board expenses:cCosts for meals and accommodation during business trips. For lump-sum reimbursements, it's important to check company policies to avoid non-refundable expenses.
• Travel expenses: costs for transportation such as flights, trains, ships, etc. When using a personal vehicle, fuel and mileage are reimbursed based on ACI tables.
• Incidentals or other miscellaneous expenses: costs not directly related to the previous categories, such as parking fees, representation costs, or photocopies of legal documents filed by a lawyer.

Phases of the expense report process

The expense reporting process generally follows these phases:

• Drafting of the Expense Report: in this phase, the employee creates the document to request reimbursement. The drafting is the sole responsibility of the employee and typically requires the following information:
  • Personal and company details.
  • List of expenses: a breakdown of costs incurred, with specific amounts and any reference numbers.
  • Date, time, and location of the expenses.
  • Supporting tax documents: receipts, invoices, or travel documents that prove the actual costs incurred.
  • Additional comments or notes to justify any extraordinary expenses.
• Supervision: this phase involves verifying the accuracy of the information submitted for reimbursement. It can be handled internally or outsourced. The supervisor checks if the provided information is correct, supported by valid tax documentation, and ensures that the expenses comply with company policies.
• Administration: in the final phase, the validated expense report data is transferred to the company’s accounting system. This step registers the expenses in the accounts, after which the employee is reimbursed.

To effectively manage the expense reporting process, it is crucial to efficiently handle the variety of receipts from different purchases involved.

Processing purchase receipts

This section examines the types of documents involved in the expense reporting process and highlights the main challenges that have, until now, hindered the automation of this process in various contexts.

Involved documents

In the expense report management process, the supporting documents are numerous and often quite varied. These are typically semi-structured documents that contain some of the key information needed to complete the expense report. For optimal process efficiency, it is crucial to classify and extract this information automatically. The most commonly involved documents include receipts, invoices, parking tickets, highway tolls, and more.

‍

‍

Main difficulties

Each of the documents mentioned above is a semi-structured document that can vary significantly depending on the vendor, both in terms of the location of key information and the format. For large companies, another challenge in centralizing expense report management is the issue of language. Purchase receipts are often linked to international travel, leading to variations in language, currency, and document format. In some cases, receipts are even handwritten, making both manual and automated processing more difficult. Additionally, receipts are often submitted in various formats, such as degraded scans, PDFs, and mobile phone photos, which are often hard to read.

‍

Main phases

Identification of the class to which the receipt belongs

A key piece of information when completing an expense report is identifying the category of the receipt being attached. Typically, the eligible expense categories include the following:

• Representation expenses (e.g., hairdressing, clothing)
• Transportation (e.g., plane, train, taxi, car rental)
• Overnight stays
• Business meals (breakfast, lunch, dinner)
• Travel tolls
• Parking fees
• Expenses related to conferences, conventions, and events

This information is often not explicitly stated in the document but can be inferred by analyzing the receipt and determining the appropriate category (e.g., verifying that a hotel receipt corresponds to an overnight stay).

Relevant information to extract

Many other relevant details can be directly extracted from the receipt, as they are often included within the document. The information commonly retrieved for accounting, tax, and control purposes includes:

• Seller information: full name, VAT number, country, and address
• Payment details: currency, total amount, taxes paid, and payment method
• Purchase breakdown: itemized lines of purchase
• Date, time, and place of the transaction.

How to process purchase receipts

Manual approach

Extracting data from purchase receipts can be costly, both in terms of time and the potential for errors, when done manually. The processing requires skilled personnel who can consistently identify and extract relevant information from often complex layouts. Some of the challenges and issues related to manual processing include:

1. Cost issues
   While manually extracting data from documents may be feasible for small businesses with limited volumes, it becomes costly as the business grows. Manual data extraction often incurs hidden costs beyond hiring additional employees to handle the workload. Adding more employees can lead to coordination challenges, increasing the likelihood of errors, particularly in identifying and entering data. Data validation is a critical phase that further adds to the processing cost. Without proper verification, error rates can reach up to 4%. The well-known 1-10-100 rule in data entry highlights these costs: verifying data accuracy at the point of entry costs around $1, correcting errors after the fact costs $10, and errors that go unnoticed can cost the company $100 or more.‍
2. Time issues
   Manual data extraction can be time-consuming, particularly in global supply chains, where additional controls and approvals are needed, and coordination between teams in different countries is required. The integration of various processing and verification steps becomes more complex due to the number of people involved and the hierarchical approval levels necessary.‍
3. Human fallibility
   The data entry process is repetitive and tedious, which can lead to low morale. Additionally, purchase receipts lack a standard format. While much of the required information is present in all documents, each vendor uses a different layout, with significant spatial variability. This makes manual data extraction challenging. Linguistic differences between the place of issue and the place of receipt can further complicate understanding the data.
   

Traditional OCR solutions

Processing purchase receipts using traditional OCR techniques combined with template matching or regex is a highly discouraged and costly approach. It requires creating specific sets of rules and templates for each document type. With countless formats and an undefined number of vendors, the process becomes unmanageable. Additionally, global solutions must account for multiple languages, further increasing the number of rules and templates required, which can seem virtually infinite and subject to constant updates as new formats and countries are introduced.

This results in high setup and maintenance costs, often with suboptimal performance. Furthermore, maintaining and configuring such a solution requires trained personnel with technical expertise.

Intelligent Document Processing (IDP)

A modern approach based on Deep Learning techniques is the ideal solution for addressing these challenges. By leveraging the best Computer Vision techniques for image analysis and NLP for natural language understanding, this approach can scale across a wide variety of formats and languages without needing to constantly adapt the solution (e.g., writing new rules or configuring new templates). Instead, the system simply requires a sufficient amount of data for training.

This approach also benefits significantly from human validation, which not only corrects the system’s errors but also enables continuous learning, allowing the algorithm to improve over time and fine-tune itself to the specific process. Unlike traditional solutions, adding new fields for extraction, new document categories for classification, or even supporting a new language doesn’t require coding. It’s often enough to gather new documents and retrain the system, a process that can be managed by someone with knowledge of the business process, without needing technical expertise.

Ultimately, the most effective Intelligent Document Processing (IDP) solutions achieve unprecedented accuracy, surpassing both manual processing and traditional rule-based approaches.

myBiros and benefits

myBiros is an Intelligent Document Processing (IDP) solution that enables the automatic processing of documents. Its core functionalities include information extraction and automatic document classification. myBiros offers these capabilities through a prebuilt set of ready-to-use APIs with pre-trained models for common use cases, and it also allows the retraining of the entire pipeline (both the OCR engine and the document interpretation system) for custom scenarios.

Leveraging advanced deep learning techniques to analyze multimodal features, myBiros can process all types of documents with a single solution. The system uses pre-trained models and data augmentation techniques, making it trainable with a minimal volume of data, thus enabling automation even for processes involving a small number of documents.

This solution incorporates a scoring mechanism to reduce false positives by allowing users to review low-confidence data, minimizing errors. Through Human-in-the-Loop and continuous learning approaches, users can correct system errors while improving its performance to avoid repeating past mistakes. Additionally, the cloud-based architecture ensures high scalability, allowing the processing of highly variable document volumes without the need for upfront resource allocation.

These features make MyBiros highly effective for processing documents typically involved in expense reports, such as receipts, invoices, parking tickets, and highway tolls. If you're interested in learning how myBiros can simplify the processing of any purchase receipt, contact us. We're here to help!