Problem Statement:

The purchase register contains numerous entries, including only those invoices on which Input Tax Credit (ITC) has been claimed. However, it also includes goods and services for which ITC availment and utilization are restricted under Section 17(5) of the CGST Act, yet credit has still been claimed.

The process of reviewing each entry, analyzing its description, and verifying whether ITC is allowable or blocked under Section 17(5) is highly time-consuming. Additionally, the risk of oversight or human error increases the chances of omissions or non-detection. If such ineligible ITC is later identified by the GST authorities, clients often shift the blame onto Chartered Accountants, questioning why they were not informed at the time of filing.

Objective:

The objective of this case study is to leverage ChatGPT's capabilities to automate the identification of ineligible Input Tax Credit (ITC) from a purchase register, ensuring compliance with Section 17(5) of the CGST Act. By utilizing AI-driven analysis, the goal is to streamline the verification process, reduce manual effort, and mitigate the risks of omission or human error. This approach aims to enhance accuracy, improve efficiency in GST compliance, and provide a proactive mechanism to prevent potential disputes or liabilities arising from incorrectly claimed ITC.

AI Solution Overview:

The proposed AI solution simplifies the process of identifying ineligible Input Tax Credit (ITC) by leveraging ChatGPT for automated analysis. The approach involves extracting item descriptions from the purchase register into a text file, which is then uploaded to ChatGPT for processing. Using advanced text analysis techniques, ChatGPT will assess each entry against Section 17(5) of the CGST Act and determine whether the claimed ITC is eligible or blocked.

This system integrates a synergy of cutting-edge technologies, including Natural Language Processing (NLP) for invoice description analysis, Machine Learning for pattern recognition, and Knowledge Graphs to establish relational mapping between tax provisions and transaction details. These technologies work together to ensure that ITC claims are assessed with high accuracy, detecting patterns and connections that might be missed in manual reviews.

The system will provide clear, clause-based reasons for each flagged entry, ensuring transparency and compliance. The results will be structured in a Word file for easy review and documentation. This approach eliminates the need for manual verification, significantly reduces the time required for analysis, and minimizes the risk of human error.

1. Input: The process begins with extracting data from the purchase register generated by ERP software such as SAP, TallyPrime, Busy Accounting Software, or any other accounting system. The key requirement is to extract the description of goods and services from all relevant invoices. These descriptions are then compiled into a plain text file, ensuring compatibility and ease of processing for AI-based analysis. This format allows for seamless ingestion into ChatGPT for further assessment.
2. Processing: Once the text file containing item descriptions is prepared, it is uploaded to ChatGPT along with a structured prompt that instructs the AI to evaluate each entry against sec.17(5) of the CGST Act. ChatGPT utilizes Natural Language Processing (NLP) to analyse descriptions, Machine Learning (ML) to detect patterns in ineligible claims, and Knowledge Graphs to establish contextual relationships between tax provisions and item categories. Based on this structured analysis, the AI systematically identifies potentially ineligible ITC claims and provides justifications referencing specific legal clauses.
3. Output: The AI-generated results are compiled into a structured Word file that categorizes each entry as eligible or ineligible ITC along with a clause-based rationale for ineligibility. This output ensures audit-ready documentation, facilitating compliance verification and reducing the risk of tax disputes. The final report will be reviewed and be discussed with the assessee before filing GST returns or finalizing the audit, enhancing accuracy and efficiency in ITC validation.

Tools and Data Mention:

The primary tool utilized for this process is ChatGPT, which serves as the AI engine for analyzing and identifying ineligible ITC. To facilitate efficient data processing, ChatGPT will internally leverage Python and Pandas for structuring, filtering, and analyzing the extracted data. These libraries enable data manipulation, pattern recognition, and categorization of transactions based on GST compliance rules.

The analysis relies on a purchase register dump extracted from ERP/accounting software such as SAP, TallyPrime, or Busy Accounting Software. The dataset should contain the following key fields:

1. GSTN – Identifies the supplier’s GST registration number.
2. Supplier Name – Helps in recognizing the entity from which the purchase was made.
3. Document No. – Captures the invoice or purchase document reference for audit tracking.
4. Description of Goods/Services – Essential for evaluating whether the item falls under Section 17(5) restrictions.
5. HSN Code – Provides a standardized classification of goods/services to aid in compliance validation.
6. Taxable Amount – Specifies the value of the purchase before tax.
7. GST Amounts (CGST, SGST/UTGST, IGST) – Details the tax component to determine if ITC is claimed correctly.

Example:

1. Articles/ CAs manually review thousands of invoices to check ITC eligibility under Section 17(5) of the CGST Act.
2. Each invoice description must be matched with tax provisions, making the process slow and inefficient.

1. ChatGPT processes purchase register data in minutes, identifying ineligible ITC claims with clause-based justifications.
2. The AI-powered workflow eliminates manual cross-referencing of tax provisions.

1. Due to the volume of transactions, there is a risk of missing ineligible ITC claims.
2. Errors may lead to penalties or disputes with tax authorities.

1. AI uses NLP, Machine Learning, and Knowledge Graphs to detect patterns and ensure precise classification of ineligible claims.
2. Reduces the likelihood of human oversight and misinterpretation of tax laws.

1. The verification process is often unstructured, leading to difficulty in maintaining audit-ready records.
2. Clients may later challenge accountants, questioning why ineligible claims were not flagged earlier.

1. AI automatically generates a structured Word file listing flagged invoices, legal references, and reasoning.
2. This ensures transparency, minimizes disputes, and enhances compliance readiness before GST audits.

Key Metrics:

Time Efficiency: Reduction in time spent from lets say 10 Man Days for checking 13 thousand rows to Very Few Man hours.

Improvement in structured reporting and audit preparedness and Client Satisfaction.

Challenges and Risks:

False positives or false negatives is the 1^st and foremost risk of using AI. This can be mitigated by regular fine-tuning, manual cross-validation, and hybrid rule-based checks. Another risk is that, sissing or inconsistent purchase register data may affect accuracy. But it can be easily mitigated by standardising data extraction, clean descriptions, and using structured formats. AI may misinterpret complex GST provisions. While AI comes with its challenges, regular improvements and thorough validation help ensure it accurately and reliably automates ITC verification, making the process faster and more efficient.

Relevance to Chartered Accountants:

As Chartered Accountants, we are the primary go-to professionals for queries related to GST ITC and audit compliance. By leveraging this AI-driven solution, we can save valuable time, enhance accuracy, and complete our work more efficiently and professionally, ensuring better compliance and client satisfaction.