FR3162890A3 - INTELLIGENT SYSTEM FOR AUTOMATED CLASSIFICATION AND ENUMERATION OF ALPHANUMERIC DATA - Google Patents

Abstract

This specification provides an automated expense enumeration system that leverages optical character recognition (OCR) and machine learning algorithms to facilitate the detection, extraction, and refinement of transaction enumeration from digital receipt images. The system begins with an OCR scan to identify transaction-related text, which is then analyzed using machine learning to accurately classify and segment the data into individual expense entries, with a particular focus on VAT classification and organizing expenses into subcategories. A user interface allows for the verification, adjustment, and confirmation of these entries, incorporating user feedback to continuously improve the accuracy of the machine learning model. Stored in a database, the finalized entries enable improved reporting, accounting practices, and efficient VAT recovery.

Description

INTELLIGENT SYSTEM FOR AUTOMATED CLASSIFICATION AND ENUMERATION OF ALPHANUMERIC DATA DOMAIN

This specification generally relates to the processing of alphanumeric data and more specifically to the automated classification of alphanumeric data.

CONTEXT

The integration of digital technology into financial management processes has progressed significantly, driven by the development of artificial intelligence (AI) and improvements in computer hardware. In particular, the area of expense management has seen transformative potential through these technologies. Optical character recognition (OCR) technology, for example, has evolved to accurately identify, extract, and organize financial data from digital images of receipts and invoices, providing a reliable method for automating expense documentation and reporting.

Despite these advances, the complexity of expenditure statements, particularly those involving multiple items and varying tax rates across different jurisdictions, presents challenges that often require human intervention. While AI and OCR technologies have significantly reduced the manual workload associated with enumerating and classifying expenditures, the nuanced requirements for accurate VAT categorization and the organization of expenditures into specific subcategories still necessitate a certain level of oversight.

SUMMARY

One aspect of the specification provides a method for the automatic intelligent enumeration of expenditures in an electronic expenditure management system, the method including: receiving a digital image of a receipt by an electronic device, in which the receipt includes one or more detailed transactions; initiating an optical character recognition (OCR) process on the received digital image to detect the text indicating said one or more detailed transactions; applying a machine learning algorithm to analyze the detected text to identify the characteristics of the detailed transactions, in which the characteristics include at least the item description, the amount, and the applicable tax rate; segmenting the identified detailed transactions into separate expenditure entries, each expenditure entry corresponding to a distinct detailed transaction, identified on the receipt; the generation of a user interface (UI) display presenting the segmented expense entries, in which the UI display includes options for a user to confirm, modify, or reject the segmentation and characteristics of each expense entry; receiving user input via the UI display regarding the confirmation, modification, or rejection of the segmented expense entries and their characteristics; updating the machine learning algorithm based on the received user input to improve the accuracy of future OCR processes and the identification of characteristics in detailed transactions; and storing the confirmed expense entries in a database associated with the electronic expense management system, in which each stored expense entry is linked to a corresponding expense status and is accessible for report generation, accounting processes, and VAT recovery.

One aspect of the specification provides a method in which the OCR process also includes preprocessing the digital image for noise reduction and contrast enhancement to improve text detection accuracy.

One aspect of the specification provides a process, including the automatic categorization of each of the segmented expenditure entries into predefined expenditure categories based on the identified characteristics and using a classification algorithm in the machine learning algorithm.

One aspect of the specification provides a method in which the generation of the user interface display also includes the display of a graphical representation of the receipt as well as segmented expense entries to facilitate verification and modification by the user.

One aspect of the specification provides a process, including the calculation of a total amount of expenditure from the sum of the amounts of confirmed expenditure entries and its comparison to a total amount shown on the receipt to ensure the completeness of the enumeration process.

One aspect of the specification provides a process in which the machine learning algorithm is further configured to learn user corrections over time, up to the categorization of expense entries, thereby improving the automatic categorization of future receipts based on past user entries.

One aspect of the specification provides a process, including the application of a validation rule that verifies the accuracy of the VAT rate applied to each expenditure entry based on the geographical location of the expenditure and the applicable tax laws.

One aspect of the specification provides a method, in which the machine learning algorithm uses user feedback on the UI screen to train the model specifically to detect detailed lines that are often misrecognized, thereby improving the algorithm's ability to accurately identify and extract detailed transaction data from various receipt formats.

One aspect of the specification provides a method, also including the integration of confirmed expenditure entries with an enterprise resource planning (ERP) system for the automation of financial reporting and accounting processes.

One aspect of the specification provides a process, including the generation of detailed reports that include detailed expenditures for each receipt, with the reports being configurable to include specific data fields based on user or business requirements.

One aspect of the specification provides a computer-implemented method for the automatic enumeration of expenses in an electronic expense management system, the method comprising: (a) the receipt, by the electronic expense management system, of a digital image of a receipt uploaded by a user; (b) the initiation, by the electronic expense management system, of an enumeration detection service configured to check the uploaded digital image for itemized lines using pattern analysis and machine learning techniques; (c) prompting, via a user interface (UI) of the electronic expense management system, the user to confirm whether the receipt contains itemized lines; (d) updating a machine learning model of the electronic expense management system, based on the user's confirmation response, in which the response includes a positive or negative confirmation regarding the presence of itemized lines on the receipt; (e) in response to a positive confirmation from the user, the extraction, by an Optical Character Recognition (OCR) enumeration service of the electronic expenditure management system, of data from the detailed lines on the digital image; (f) the generation, by the UI, of a display of the extracted detailed lines for review by the user and the receipt of user input regarding corrections to the detailed lines; (g) the updating of the machine learning model based on user input to refine the accuracy of the detection of the detailed receipt and the data extraction by OCR; (h) the processing of the expense reimbursement process in the electronic expenditure management system, which includes the detailed receipts, for reimbursement and recovery of VAT; and (i) providing an option in the electronic expenditure management system for auditors to launch the OCR enumeration service using a "detail receipt" feature, further improving the machine learning model with data from auditor interactions; wherein the machine learning model improves the enumeration detection service and the OCR enumeration service by: (i) continuously learning from user and auditor interactions to improve the detection of detailed receipts, and (ii) refining the OCR data extraction process to improve the accuracy of data extraction from detailed receipt lines.

One aspect of the specification provides a server-implemented method for the intelligent automatic enumeration of expenditures in an electronic expense management system, the system configured to: (a) receive a digital image of a receipt, the receipt including one or more detailed transactions, the receipt being downloaded by a user via an electronic device; (b) run an optical character recognition (OCR) module on the server to process the received digital image and detect the text indicating said one or more transactions; (c) apply a machine learning algorithm, run on the server, to analyze the detected text and identify the characteristics of the detailed transactions, the characteristics including at least the item description, the amount, and the applicable tax rate; (d) segment the identified detailed transactions into separate expenditure entries on the server, each expenditure entry corresponding to a unique detailed transaction, identified on the receipt; (e) generate a user interface (UI) that is provided to the electronic device to present the segmented expenditure entries, the UI including options for the user to confirm, modify, or reject the segmentation and identified characteristics of each expenditure entry; (f) receive, at the server level, user input from the UI regarding the confirmation, modification, or rejection of the segmented expenditure entries and their characteristics by the user; (g) update the machine learning algorithm on the server based on the user input to improve the accuracy of subsequent OCR processes and the identification of characteristics in detailed transactions; and (h) store confirmed expenditure entries in a database managed by the server and associated with the electronic expenditure management system, each stored expenditure entry being linked to a corresponding expenditure statement and made accessible for activities such as report generation, accounting processes, and VAT recovery.

BRIEF DESCRIPTION OF THE FIGURES

There FIG. 1 a schematic diagram of a system for the intelligent automatic enumeration of expenses.

There FIG. 2 is an example of a classification server structure of the FIG. 1 .

There FIG. 3 shows a diagram representing an intelligent automatic expense enumeration process.

There FIG. 4 is an example of a receipt.

There FIG. 5 is another example of a receipt, as entered into the system.

There FIG. 6 is an example of a user interface.

There FIG. 7 is another example of a user interface.

There FIG. 8 shows a diagram that illustrates the feedback loop of the machine learning process of the FIG. 3 .

DETAILED DESCRIPTION

There FIG. 1 This shows an intelligent automatic expense enumeration system, usually denoted as 100. System 100 includes a classification server 104. In system 100, the classification server 104 connects to a network 108, such as the Internet. The network 108 interconnects the classification server 104 with: a) a payment processing engine 112; b) a plurality of client devices 116; and d) an administrator workstation 120. As we will see below, the classification server 104 performs a number of processing functions for system 100.

Note that, collectively, customer devices 116-1, 116-2...116-n are referred to generically as device 116, device 116. This nomenclature is used elsewhere in this document.

There FIG. 2 This shows a schematic diagram of a non-limiting example of the internal components of the classification server 104. In this example, the classification server 104 includes at least one input device 204. The input from device 204 is received by a processor 208, which in turn controls an output device 212. The input device 204 can be a traditional keyboard and/or a mouse to provide physical input. Similarly, the output device 212 can be a display. In variations, additional input devices 204 and/or other output devices 212 are considered or can be omitted entirely, depending on the context.

The processor 208 can be implemented as a plurality of processors or several multi-core processors. The processor 208 can be configured to execute different programming instructions in response to input received via one or more input devices 204 to control one or more output devices 212 to generate output on those devices.

To perform its programming functions, the processor 208 is configured to communicate with one or more memory units, including non-volatile memory 216 and volatile memory 220. Non-volatile memory 216 can be based on persistent memory technology, such as electrically erasable programmable read-only memory (EEPROM), flash memory, solid-state drives (SSDs), other types of hard drives, or combinations thereof. Non-volatile memory 216 can also be described as non-transient, computer-readable storage. Furthermore, more than one type of non-volatile memory 216 can be provided.

Volatile 220 memory is based on random access memory (RAM) technology. For example, volatile 220 memory can be based on double data rate (DDR) synchronous dynamic random access memory (SDRAM). Other types of volatile 220 memory are being considered.

The processor 208 also connects to network 108 via network interface 232. Network interface 232 can also be used to connect another computing device that has an input and output device, thus avoiding the need for input device 204 and/or output device 212.

Programming instructions in the form of applications 224 are typically stored persistently in non-volatile memory 216 and used by the processor 208, which reads from and writes to volatile memory 220 during the execution of the applications 224. Various processes discussed herein can be coded as one or more applications 224. One or more arrays or databases 228 are stored in non-volatile memory 216 for use by the applications 224.

The infrastructure of the classification server 104, or a variant thereof, can be used to implement any computing node in the system 100, including the payment processing engine 112. In addition, the classification server 104 and the payment processing engine 112 can also be implemented as virtual machines and/or mirror images to balance the load.

Furthermore, a person skilled in the art will recognize that the essential elements of the processor 208, the input device 204, the output device 212, the non-volatile memory 216, the volatile memory 220, and the network interface 232, as described in relation to the environment of the classification server 104, have analogies in the various form factors of client machines such as those that can be used to implement the client devices 116 and the workstation 120. The client devices 116 and the workstation 120 can be based on any combination of computer workstations, laptops, electronic tablets, mobile phone devices, or the like.

Each device 116 and its user 124 are thus associated with a user identification object 128. A person skilled in the art will recognize that the form of an identification object 128 is not particularly limited, and that it can, in a simple embodiment, simply be an alphanumeric sequence that is entirely unique with respect to other identification objects in the system 100. Identification objects can also be more complex, insofar as they can be combinations of account identification data (e.g., username, name, password, two-factor authentication token, etc.) that uniquely identify a given user 124. The identification objects themselves can also be indexes that point to other identification objects, such as accounts. The essential point is that they are uniquely identifiable within the system 100 in association with what they represent. The user identification object 128 can therefore be used as part of the authentication of an account and/or a session with the classification server 104 and/or the payment processing engine 112.

According to this embodiment, the client devices 116 rely on any suitable client computing platform, operated by the users 124, to submit expense reimbursement requests to be submitted for reimbursement by their employer or other entity. Reimbursement requests include travel expenses such as transportation, accommodation, and meals. Reimbursement requests may also include purchases of small equipment or similar items. The nature of the reimbursement request itself is not particularly limited and is provided for illustrative purposes, although it is not necessarily relevant to the technical aspects of this specification.

As discussed in more detail below, classification server 104 hosts an expense reimbursement application 224-1. The application 224-1 enables an expense reimbursement process which includes a user 124 logging into an account hosted by an engine 104 via identification data associated with their respective identification object 128. The user 124 can then interact with graphical interfaces generated by the engine 104 on the screen of a respective device 116 to receive detailed alphanumeric information articulating the details of the expense reimbursement request and also to upload images of detailed documents such as receipts or other documentation to support the alphanumeric information. In due course, the expense reimbursement request can be approved (or rejected), in whole or in part, by engine 104, either automatically or with human supervision manifested at workstation level 120 by an administrator 132. Payment for the approved parts of the request can be made by the payment processing engine 112 which oversees the transfer of funds to a financial account associated with the respective user.

For the development, the FIG. 3 This diagram shows a process for intelligent automated expense validation, typically designated as 300. Process 300 can be implemented on System 100. Business users can choose to implement Process 300 on System 100 or variations thereof, or with certain blocks omitted, executed in parallel, or in a different order than shown. Therefore, Process 300 can also be modified. However, for explanatory purposes, Process 300 will be described in relation to its execution on System 100, with a specific focus on Processing Method 300, such as, for example, a portion of a 224-1 expense claim stored on the Classification Server 104 and its interactions with other nodes in System 100.

Block 310 includes the receipt of a digital image of a receipt, the receipt including one or more detailed transactions. Receipt 404 at the FIG. 4 shows an example of such a receipt. The 404 receipt can be captured by a camera in a device 116 or otherwise digitized in any manner now envisaged or in the future. The classification server 104 can receive the 404 receipt via the network 108 for further processing.

In particular, receipt 404 is a striking example for this specification as it is a hotel bill which includes a total charge but also a plurality of items including room charges, meals and other services up to restaurant receipts listing the various items purchased.

Block 320 initiates an Optical Character Recognition (OCR) process on the received digital image to detect the text that indicates one or more detailed transactions. OCR technology is used to scan the digital image of the receipt in order to identify and extract textual information related to the transactions listed on the receipt. This step transforms the visual data of the receipt into machine-readable text, enabling further analysis of the detailed transactions.

The information on receipt 404 as a result of block 320 is reproduced below:

Grand Sundance Hotel 7760 Pennington St. Bardstown, KY 40004

Received

Customer Information Name: David Address: 390 West Pearl St. Commack, NY 11725 Contact Number: 1-202-555-0145

Room No.: 16 Nights: 2 Check-in time: 12:30 Check-out time: 11:30 Date Description Taxes Costs Credit 29/03/19 Room accommodation 5% $3,000.00 29/03/19 Restoration 1% $1,000.00 29/03/19 Phone bill 1% $280.00 29/03/19 Linen 1% $150.00 29/03/19 Car rental 1% $450.00

Total tax: 9% Total fees: $4,880.00 Tax rate: $439.20

Grand total: $5,319.20

Customer signature: ___________

Cashier's signature: ___________

Now, referring to the FIG. 5 To continue the discussion regarding an example of process 300, another example of a 504 receipt will be mentioned. (The 504 receipt having already been entered into classification server 104). As shown in the FIG. 5 The OCR process for block 320 is complete, but the information on receipt 504 is reproduced below.

Hotel bill - 0 16/10/2023 Germany - 550.00 EUR

Service fee: 20 EUR

Parking fee: 50 EUR

Internet expenses: EUR 30

Total amount: EUR 650.00

Block 330 incorporates a machine learning algorithm to analyze detected text and identify detailed transaction characteristics, such as item descriptions, amounts, and any applicable tax rates. The use of machine learning algorithms allows the extracted text to discern various attributes for each transaction, including descriptions, costs, and taxes. This algorithmic approach is designed to learn and improve over time, adapting to new data and user feedback to enhance its ability to accurately classify and interpret transaction details.

Block 340 includes the segmentation of detailed transactions identified as separate expense entries, each corresponding to a distinct detailed transaction identified on the receipt. The system segments the extracted transaction details into individual expense entries. This segmentation provides a structured breakdown within a single receipt, facilitating expense management and tracking by categorizing each item separately.

Block 350 includes the generation of a user interface (UI) display (such as on a 116 device) showing segmented expense entries. The UI display includes options for users to confirm, modify, or reject the segmentation and characteristics of each expense entry. The UI is designed to present the segmented expense entries in a clear and navigable manner, allowing users to review, confirm, or adjust the details of each detailed transaction. Options to modify or reject specific entries are provided to ensure that the expense report accurately reflects actual transactions.

There FIG. 6 This shows an example of user interface (UI) 604 for display on device 116 illustrating the execution of blocks 330, 340, and 350 where, in relation to receipt 504: Service charge: EUR 20; Parking charge: EUR 50; and Internet expense: EUR 30 are extracted as separate itemized charges from the room charge of EUR 550. Specifically, "Service charge: EUR 20; Parking charge: EUR 50; and Internet expense: EUR 30" are identified as separate itemized charges subordinate to the room charge itself. These separate itemized charges subordinate to the room charge can be referred to as "sub-items," which are broken down into individual items for proper evaluation, identification, auditing, and/or reimbursement purposes.

To accomplish this, UI 604 includes a section titled "Create Receipt," where users can add digital images or electronic receipt files. This area is bordered by "Attachments," which provides an interactive element labeled "Click or Drag to Add Attachment," allowing users to easily upload their receipt documents. Adjacent to the attachments area (612), there is a categorization header that divides the input fields into two categories: "To be Reimbursed" and "Already Paid by Company," allowing users to classify expenses accordingly. In the "To be Reimbursed" section, users are prompted to complete expense details. This includes "Period" fields, indicating the date of the expense, and "Country," specifying the location where the expense was incurred. Below this is a subsection titled "Receipt Items." This subsection provides a dynamic form where users can enter multiple items associated with a single receipt. It includes dropdown menus for "Receipt Type," where users can select the nature of the expense, and "Tax Rate," where users can specify the tax rate as "standard," "reduced," or "none" for each item. Adjacent to the dropdown menu are fields for users to enter the monetary "Amount" for each item and select the "Currency" from a dropdown menu, ensuring accurate financial documentation in different currencies. For each detailed entry, a line is provided where users can enter a description, such as "Internet Expenses," "Parking Fees," or "Service Fees," with corresponding fields to complete, allowing users to view the amounts clearly and in an organized manner. At the bottom of the "Receipt Items" subsection, a running total, labeled "Total Amount," is displayed, summarizing the costs of all entered items. Below the list area, there is an additional text entry field labeled "Description," allowing users to include a detailed description of the expense, if needed. The final section of UI 604 is titled "Additional Information," containing a text box with a checkbox. This box allows users to enter additional details about the expense, which can be marked for inclusion in the Expense Report for further clarification or to support multi-line entry. A "Save and Close" button is prominently displayed at the bottom of the UI, indicating the action users will take to save the entered data and exit the receipt creation process. UI 604 is designed to facilitate the organized entry of expense data, promoting a streamlined reimbursement process and providing a user-friendly experience for the efficient management of financial records within the electronic expense management system.

The 360 block includes receiving user input via the UI display, where users can confirm, modify, or reject segmented expense entries and their characteristics. Users interact with the UI to finalize the details of the expense entries. This interaction includes confirming the accuracy of the segmented entries, making modifications necessary to match actual transactions, or rejecting incorrect segments. This step can be used to confirm that the expense report is accurate and reflects the user's verification and adjustments.

The 360 block can be implemented using a suitable UI such as UI 604. Another example of a 360 block is UI 704. FIG. 7 which is essentially the same as UI 604 but shows some fields in "edit" mode, including field 708.

Block 370 includes an update to the user-input-based machine learning algorithm to improve the accuracy of future OCR processes and the identification of detailed transaction features. The system incorporates user feedback to refine the machine learning algorithm, making it an expert in identifying and classifying transaction details in future scenarios. This adaptive learning process is designed to progressively improve system performance by reducing errors and increasing the efficiency of processing expense enumerations.

There FIG. 8 shows a diagram 804 which illustrates the machine learning feedback loop 808 envisaged in block 370.

Block 380 includes the storage of confirmed expense entries in a database linked to the electronic expense management system. Each stored expense entry is linked to a corresponding expense report and is accessible for report generation, accounting processes, and VAT recovery. Confirmed expense entries are stored in a structured manner within the system's database, facilitating access and retrieval for various purposes such as generating expense reports, integrating with accounting systems, and processing VAT recovery. This structured storage keeps expense data organized, secure, and readily available for financial management and reporting needs.

As discussed, System 100 specializes in extracting specific sub-items from expense receipts. VAT, in particular, is a crucial type of sub-item. The specification provides a specialized solution for extracting and managing Value Added Tax (VAT) information from detailed receipts—a process of particular importance due to the variable nature of VAT across different jurisdictions. The system recognizes that businesses often encounter a multitude of VAT rates and rules depending on the location and type of services purchased. It is specifically designed to accurately capture and extract this information. Using advanced Optical Character Recognition (OCR) and machine learning algorithms, the system can identify and separate the VAT amounts for each item listed on a receipt. This capability is valuable for businesses operating internationally or in regions with diverse tax regulations, as it enables proper VAT recovery and compliance with local tax laws. Automating VAT extraction simplifies the refund process, ensures accurate financial reporting, and reduces the administrative burden associated with manual VAT calculation, especially when dealing with receipts that aggregate multiple transactions, each potentially subject to different VAT rates.

In general, machine learning, neural networks, or other artificial intelligence techniques can be employed at the 808 loop level, with various manual interventions being successively used to train a machine learning model to determine which types of inputs are sub-items, including the appropriate identification of the tax (e.g., VAT).

In light of the above, it now appears that variants, combinations, and subsets of the previous embodiments are being considered. For example, the classification server 104 can be eliminated, or its function can be distributed across a variant on system 100 and/or executed entirely locally on device 116.

It's also worth noting that process 300 can be implemented or omitted with certain blocks, executed in parallel, or in a different order than shown. For example, process 300 can begin at a modified version of block 350, where the user interacts with a user interface to initiate the receipt delivery process in block 310. Alternatively, the process can begin with the user capturing a digital image of a receipt using the 224-1 expense reimbursement application, followed by navigation to a GUI that visualizes the expenses, or by a pop-up within the 224-1 expense reimbursement application that immediately displays a breakdown of the sub-items extracted from the photographed receipt.

As discussed, one or more applications 224 may incorporate machine learning or artificial intelligence, including but not limited to deep learning-based algorithms and neural networks. These technologies can be trained to enhance the machine learning functions described herein. The machine learning applications 224 can be operated by the processor 208 in a training mode, in which machine learning, deep learning algorithms, and neural networks are trained in accordance with the teachings herein.

Said one or more machine learning algorithms and/or deep learning algorithms and/or neural networks of machine learning applications 224 may include, but are not limited to: a generalized linear regression algorithm; a random forest algorithm; an automatic support vector algorithm; a gradient reinforcement regression algorithm; a decision tree algorithm; a generalized additive model; neural network algorithms; deep learning algorithms; evolutionary programming algorithms; Bayesian inference algorithms; reinforcement learning algorithms, and the like. Preferences for generalized linear regression algorithms, random forest algorithms, automatic support vector algorithms, gradient reinforcement regression algorithms, decision tree algorithms, generalized additive models, neural network algorithms, and deep learning algorithms may be due to factors such as computational efficiency, scalability, data requirements, or suitability for a specific application.

In some implementations, the machine learning algorithm is designed to learn from user corrections over time. As users provide corrections related to the categorization of expense entries, the algorithm can adapt and refine its processes, potentially improving the automatic categorization of future receipts based on accumulated user input.

Furthermore, in some embodiments, the machine learning algorithm leverages user feedback from the UI display to train the model to detect detailed lines that are typically poorly recognized. This training can improve the algorithm's ability to accurately identify and extract data from various receipt formats.

In some embodiments, the OCR process for extracting text from digital images of receipts includes preprocessing steps for noise reduction and contrast enhancement. This preprocessing can improve the accuracy of text detection, enabling more efficient extraction of text data for subsequent analysis by machine learning algorithms.

In another variation, the process can automatically categorize each segmented expenditure entry into predefined spending categories based on identified characteristics. This categorization is facilitated by a machine learning classification algorithm that analyzes the characteristics to assign the appropriate spending categories.

Furthermore, some embodiments include a user interface display that presents a graphical representation of the original receipt along with the segmented expense entries. This feature can help users automatically verify and modify the segmented entries by providing a visual comparison, which can improve user confidence in the accuracy of the segmentation.

Other improvements to the process include calculating the total expenditure amount from confirmed expense entries and comparing this calculated total to the total amount shown on the receipt. This step can improve the completeness and accuracy of the enumeration process, helping to verify that all transactions have been recorded and properly documented.

Another aspect involves a validation rule that verifies the accuracy of the VAT rate applied to each expense entry, taking into account the geographical location of the expense and applicable tax laws. This approach can contribute to compliance and accuracy in the processing of tax-related expenses.

Furthermore, the system can integrate confirmed expense entries into an enterprise resource planning (ERP) system. This integration can facilitate the automation of financial reporting and accounting processes, simplify financial operations, and improve data accuracy within enterprise systems.

Furthermore, the system can generate detailed reports that include a breakdown of expenses for each receipt. These reports can be configured to include specific data fields based on user or company requirements, enabling customized financial reports that meet diverse organizational needs.

Furthermore, this specification provides an embodiment in which the system integrates a risk assessment module specifically designed to evaluate the potential for fraud or error in listed transactions. In this embodiment, the machine learning algorithm can be trained to categorize and list expenses and also assign a risk level to each transaction, based on predefined criteria. These criteria can include unusual spending patterns, discrepancies between entries of similar items, and non-compliance with company spending policies. For example, transactions involving unapproved spending types, such as personal expenses like cigarettes or luxury items typically not covered by standard company policies, are automatically flagged as high-risk.

The system also enhances its risk assessment capabilities by identifying out-of-policy expenses such as luxury accommodations or upgrades during trips that exceed company-approved limits. It can also detect duplicate claims where the same receipt is submitted multiple times, either by an employee or in different reports, indicating potential fraudulent activity. Furthermore, expenses incurred on dates or at locations inconsistent with planned business activities can trigger alerts, suggesting inaccuracies or misuse.

Currency mismatches in transactions—expenses for which reimbursement is requested in currencies not associated with the reported location of the expense—also constitute red flags for potential review. Similarly, non-compliant VAT refund claims, in which VAT recovery is requested for ineligible expenses or incorrect tax rates, are identified and flagged for administrative action.

Each of these risk factors, and others that will appear to business people, can be visually represented in a dashboard of workstation 120 for the administrator 132, or as an icon in the interface of a device 116 for the user 124, allowing for quick and efficient identification and prioritization of transactions that may require further review.

A professional will now appreciate that the lessons learned can improve technological efficiency and the use of IT and communication resources throughout the entire System 100 by automatically identifying items and their types on documents such as receipts during entry, assessments, and audits. This can be optionally coupled with machine learning and generate a risk level for an administrator or user, displayed as a dashboard or icon. In this way, expenses can be allocated appropriately, and anomalies in expense reports can be managed more automatically. This reduces the system resources required to send reports back and forth between the user and administrator, increases the throughput of automatic expense reimbursements, and also provides a graphical interface that facilitates machine learning.

It must be acknowledged that the characteristics and aspects of the various examples provided above can be combined in other examples that also fall within the scope of the present invention. Furthermore, the figures are not to scale and may be exaggerated in size and shape for illustrative purposes.

Claims (12)

An intelligent, automated method for enumerating expenses in an electronic expense management system, the method comprising:
the receipt of a digital image of a receipt by an electronic device, in which the receipt includes one or more detailed transactions;
the initiation of an optical character recognition (OCR) process on the received digital image to detect the text indicating said one or more detailed transactions;
the application of a machine learning algorithm to analyze the detected text in order to identify the characteristics of detailed transactions, in which the characteristics include at least the description of the item, the amount and the applicable tax rate;
the segmentation of detailed transactions identified into separate expense entries, each expense entry corresponding to a distinct detailed transaction, identified on the receipt;
the generation of a user interface (UI) display presenting segmented expense entries, in which the UI display includes options for a user to confirm, modify, or reject the segmentation and characteristics of each expense entry;
receiving user input via the UI display regarding the confirmation, modification, or rejection of expense entries and their characteristics;
the updating of the machine learning algorithm based on received user input to improve the accuracy of future OCR processes and the identification of detailed transaction features; and
the storage of confirmed expenditure entries in a database associated with the electronic expenditure management system, in which each stored expenditure entry is linked to a corresponding expenditure status and is accessible for report generation, accounting processes and VAT recovery. The method according to claim 1, wherein the OCR process further includes the preprocessing of the digital image for noise reduction and contrast enhancement in order to improve the accuracy of text detection. The method according to claim 1 further comprises the automatic categorization of each of the segmented expenditure entries into predefined expenditure categories based on the identified characteristics and using a classification algorithm in the machine learning algorithm. The method according to claim 1, wherein the generation of the user interface display further includes the display of a graphical representation of the receipt as well as segmented expense entries to facilitate verification and modification by the user. The method according to claim 1, further comprising calculating a total amount of expenditure from the sum of the amounts of confirmed expenditure entries and comparing it to a total amount shown on the receipt to ensure the completeness of the enumeration process. The method according to claim 3, wherein the machine learning algorithm is further configured to learn the user's corrections over time, up to the categorization of expense entries, thereby improving the automatic categorization of future receipts, based on the user's past entries. The method according to claim 1 further includes the application of a validation rule which verifies the accuracy of the VAT rate applied to each expenditure entry, based on the geographical location of the expenditure and the applicable tax laws. The method according to claim 1, wherein the machine learning algorithm uses user feedback on the UI screen to train the model specifically to detect detailed lines that are often misrecognized, thereby improving the algorithm's ability to accurately identify and extract detailed transaction data in various receipt formats. The method according to claim 1, further comprising the integration of confirmed expenditure entries into an enterprise resource planning (ERP) system for the automation of financial reporting and accounting processes. The method according to claim 1, further comprising the generation of detailed reports which include detailed expenditures for each receipt, the reports being configurable to include specific data fields, based on user or business requirements. A computer-implemented method for the automatic enumeration of expenditures in an electronic expenditure management system, the method comprising:
(a) the receipt, by the electronic expenditure management system, of a digital image of a receipt downloaded by a user;
(b) the launching, by the electronic expenditure management system, of an enumeration detection service configured to check the uploaded digital image for detailed lines using pattern analysis and machine learning techniques;
(c) the act of inviting, via a user interface (UI) of an electronic expense management system, the user to confirm whether the receipt contains the detailed lines;
(d) updating a machine learning model of the electronic expenditure management system, based on the user's confirmation response, in which the response includes a positive confirmation or a negative confirmation regarding the presence of detailed lines on the receipt;
(e) in response to a positive confirmation from the user, the extraction, by an optical character recognition (OCR) enumeration service of the electronic expenditure management system, of data from the detailed lines on the digital image;
(f) the generation, by the UI, of a display of the extracted detailed lines for review by the user and the receipt of user input regarding corrections to the detailed lines;
(g) updating the user input-based machine learning model to refine the accuracy of detailed receipt detection and data extraction by OCR;
(h) the processing of the expense reimbursement process in the electronic expense management system, which includes detailed receipts, for reimbursement and VAT recovery; and
(i) providing an option in the electronic expense management system for auditors to launch the OCR enumeration service using a "Detail Receipt" feature, further improving the machine learning model with data from auditor interactions; wherein the machine learning model improves the enumeration detection service and the OCR enumeration service by:
(i) continuously learning from auditor and user interactions to improve the detection of itemized receipts; and
(ii) refining the OCR data extraction process to improve the accuracy of data extraction from detailed lines on receipts. A server-based system for the intelligent automatic enumeration of expenses in an electronic expense management system, the system being configured to:
(a) receive a digital image of a receipt, the receipt including one or more detailed transactions, the receipt being downloaded by a user via an electronic device;
(b) run an optical character recognition (OCR) module on the server to process the received digital image and detect the text that indicates said one or more detailed transactions;
(c) apply a machine learning algorithm, run on the server, to analyze the detected text and identify the characteristics of detailed transactions, the characteristics including at least the description of the item, the amount and the applicable tax rate;
(d) segment the identified detailed transactions into separate expense entries on the server, each expense entry corresponding to a unique detailed transaction, distinguished on the receipt;
(e) generate a user interface (UI) which serves the electronic device to present segmented expenditure entries, the UI including options for the user to confirm, modify or reject the segmentation and characteristics of each expenditure entry;
(f) receive, at the server level, user input from the UI regarding the confirmation, modification or rejection of segmented expenditure entries and their characteristics, by the user;
(g) update the machine learning algorithm on the server based on received user input to improve the accuracy of subsequent OCR processes and the identification of detailed transaction features; and
(h) store confirmed expenditure entries in a server-managed database associated with the electronic expenditure management system, each stored expenditure entry being linked to a corresponding expenditure statement and made accessible for activities such as report generation, accounting processes and VAT recovery.