Digitizing Receipts: For Simplified Business Expense Tracking With querifai.ai

With text extraction APIs on querifai.ai, effortlessly digitize your receipts and invoices. Save your time, efforts, and the chances of errors.

This guide illustrates how querifai.ai's document analysis eases expense management by converting financial records, like receipts and invoices, into a digital format. Improve efficiency and accuracy through our user-friendly, no-code AI SaaS platform. Follow our step-by-step guide to digitize receipts within a few clicks.

About querifai.ai Platform

querifai.ai is an AI SaaS platform to explore and evaluate a wide range of readily available AI services specially designed for business people and individuals having no/low expertise in AI. Just provide your data and let our platform handle the rest through multiple service providers at the back end on the cloud like Google, AWS, and Azure.

Challenges in Manual Expense Tracking

Tracking expenses, organizing piles of receipts, and entering data manually, especially in scenarios where business expenditures are frequent and diverse, is tedious. This process is time-consuming and unproductive, increasing the chance of error while entering the data manually.

querifai.ai Helps to Digitize Invoices and Receipts

Optical Character Recognition (OCR in the following), is a technology that converts scanned documents or images into machine-readable and thus editable and searchable format.

The OCR feature of querifai.ai offers a seamless solution for this use-case. By capturing images of receipts, the technology quickly extracts crucial data such as dates, quantities, amounts, and vendor details etc in case of receipts. This information is then efficiently organized into a digital expense report in tabular form, ready for integration into tools like Excel.

querifai.ai's OCR suite is not just a tool for individual expense management on their no-code AI platform, it's a versatile resource for businesses of all sizes. Its capabilities extend to corporate centers, to automate operations and enhance overall efficiency, reduce errors, and streamline the entire expense management process. These capabilities are supported by the easy-access API of querifai. The API provides access to different OCR services (e.g. from Google, Microsoft, …) via one standardized interface.

Step-by-Step Guide to Digitize Invoices and Receipts Using querifai.ai

Let's go through a step-by-step guide demonstrating how to utilize our platform to convert your receipts from physical to digital format.

Create an account on querifai.ai if you haven't done so already:
Go to https://querifai.ai/home
Select Vision from the Use Cases panel.

Select the Document Analysis (Forms) from the Vision use case that extracts key-value pairs and tables from the documents.

To create the dataset containing the receipts, click on Upload a new Dataset and then click here.

To upload your receipts, click the Choose Files button, select the images, and then click the Create button.

You will return to the Select from existing data page. From there, select the newly uploaded data, and make sure to select all the services: Google, Clarifai, and Azure. Finally, click the Evaluate button.

You will see the summary of your use case before proceeding. Click the Proceed button, and that's it!

Analyzing Multiple Cases on the API Results

Evaluation Measures

Once the results are ready, a number of summary statistics are shown, to provide a high-level understanding of the results from each API vendor. Among them, the number of entities that have been recognized, to quickly see whether one service has missed anything, the number of tables, incl. their fill level. The mean confidence to grasp the reliability of individual API results. If mean confidence levels are low, it is advisable to utilize at least one additional vendor for confirmation of critical results, e.g. bank account numbers.

Case 1: A Simple Stationary Shop's Receipt

Let's take a simple receipt, with some info like invoice number, item description, quantity, rate, total items, and total amount.

In a comparative analysis of entity and table recognition, AWS demonstrated the best performance in identifying the most entities, while Azure achieved the highest performance in confidence levels, indicating higher accuracy. All three services—AWS, Azure, and Google, effectively recognized and extracted content from tables in a receipt, with AWS showing particular strength in table content extraction.

To gain deeper insights, click “See Details” below the results. This view shows the key-value pairs on the panel on the right and the extracted tables below.. To highlight the borders on the image for the recognized results by each service, select a vendor or several ones from the “Service provider” drop-down. The identified data is highlighted in different colors. Each color indicates the service provider.

All the services accurately identified and extracted the Item descriptions, quantity, rates, and amount. However, Google also demonstrated the best performance because it identified the total items, price, date, and time fields accurately. However, the date and time are present in an unstructured form in the table.

You can copy the table by clicking on the clipboard icon on the top-right of the table, paste it into the Excel sheet for record-keeping purposes, and visualize the results further.

AWS accurately extracted all the info from the table. However, unlike Google, it did not extract certain fields like total items, total price, date, and time.

Comparison Summary

Each provider shows unique strengths and weaknesses. Google confidently extracts invoice numbers, while AWS accurately identified total amounts, a critical factor for financial documents. Azure demonstrates its performance in phone number recognition, indicating its ability to extract contact details, and total items. However, all services faced challenges in extracting less structured data like user/admin details, and date and time information, with Google showing the lowest confidence in these areas. The key takeaway is that each service has its niche strengths, making them suitable for different types of data extraction tasks.

Case 2: A Garment Store’s Receipt

Another case, where the receipt follows a complex table structure, compared to case 1, where it is challenging to extract the table from the original receipt and organize it properly into a meaningful structure in the output table.

The overview shows again that all services have different strengths. Whether looking at the regions where items were recognized, or at the different accuracy levels. This indicates that the choice between these services depends on the specific needs of the task, whether it's breadth of recognition, accuracy, or completeness of data extraction.

Below is a detailed view of the recognized key-value pairs highlighted in different colors, each indicating another vendor.

Google successfully extracted most receipt elements into the table, including the item code, description, rate, quantity, discount, and amount. However, extraneous entries above the table like the date, time, cashier etc, need to be correctly placed within the table structure by the service providers. This is due to the unstructured nature of the data in the original receipt.

AWS extracts the core item details correctly, including the item code, description, rate, quantity, and amount. It does not include unrelated text in the table, focusing solely on the transaction details. However, the discount field is not correctly recognized.

Google achieves high confidence in extracting the rate and quantity. However, a few pairs are jumpled up.

Azure stands out among all of the services with excellent results. It accurately extracts almost all of the key-value pairs with high confidence.

Key Differences Between the Results of Both Cases

In comparing the digitization of receipts from a simple stationary shop and a garment store, the key difference lies in the complexity of the receipts and the corresponding performance of digitization services.

Case 1 follows a simpler receipt structure and therefore allows for more accurate data extraction across all services.

In contrast, Case 2 has a more complex receipt’s structure compared to Case 1, with additional elements like item codes and discounts. Here In this case Azure demonstrates high performance, effectively handling varied data complexities.

The below images side by side depict the difference between two cases.

Conclusions from Receipt Digitization Through Multiple APIs

The Random Nature of Accuracy in Receipt Digitization

The performance of services like Google, Azure, and AWS can vary. This inconsistency suggests that each model has specific strengths and weaknesses, making the choice of service dependent on the nature of the receipt. This variation highlights the importance of choosing a service based on its suitability to the task at hand, rather than a one-size-fits-all approach.

Enhanced Accuracy with Combined API Results

The result confidence increases when various services provide the same result. For example, if Google and Azure accurately identify the same critical values or table structures, it strongly indicates reliable digitization. This consensus approach can add a layer of validation to the extracted data.

Challenges with Poorly Structured Data

In receipt digitization, a major challenge is dealing with poorly structured or blurred receipts. Receipts with disorganized layouts or faint prints may need to be accurately digitized by AI. Various pre-processing techniques like image enhancement or layout standardization can mitigate these issues, improving data extraction.

Applying the Four-Eye Principle in Receipt Digitization

The "four-eye principle" is beneficial in any context of AI. Validating the results of one service with another or even human verification can combine the precision of AI with human oversight. This approach ensures a higher level of accuracy in extracting data from receipts.

Call Our API in Your Code

For organizations dealing with a large volume of paper receipts, such as accounting departments or financial services, the need for an efficient and accurate digitization solution is necessary. Our service offers a practical solution to this challenge. The Python code below illustrates how the querifai API can be seamlessly integrated into existing systems, enabling the conversion of physical receipts into digital format.

Python Code

Output

Key: Quantity, Value: 1, Confidence: 0.988

Key: TotalPrice, Value: 150, Confidence: 0.987

Key: Description, Value: CANVAS LOCAL 1×1.5, Confidence: 0.982

Key: Price, Value: 200, Confidence: 0.987

Key: Quantity, Value: 1, Confidence: 0.988

Key: TotalPrice, Value: 200, Confidence: 0.987

Key: Description, Value: CANVAS LOCAL 2×1.5, Confidence: 0.982

Key: Price, Value: 300, Confidence: 0.987

Key: Quantity, Value: 2, Confidence: 0.988

Key: TotalPrice, Value: 600, Confidence: 0.987

Key: Description, Value: WRAPPING SHEET 50, Confidence: 0.982

Key: Price, Value: 50, Confidence: 0.987

Key: Quantity, Value: 1, Confidence: 0.987

Key: TotalPrice, Value: 50, Confidence: 0.987

Lastly

Whether you need to digitize a few receipts from a business trip or you are searching for a reliable solution for an accounting center, querifai has a suitable offering. Either access the OCR services via our no-code interface or implement the services in your own environment via our API endpoints.