Accountants and Artificial Intelligence

By: Juan Bautista López, Business Services & Outsourcing Accountant 

 

For many accountants, it is quite common to enter prompts like the following1 into an Artificial Intelligence (AI) tool, such as ChatGPT or Gemini:

  • Hello ChatGPT, please act as an expert in IFRS and tell me the accounting treatment of a foreign exchange swap contract when the company…
  • Hello Gemini, act as a tax expert familiar with the Peruvian Income Tax Law and indicate the tax treatment of interest incurred from financing by a company domiciled in a tax haven…

And the result of such prompts is often a confusing, biased, and totally incorrect response that does not align with IFRS. In addition, when asked about tax matters, the AI may even cite reports from SUNAT2 or rulings from the Peruvian Tax Court that either do not exist or, if they do, are not related to the query. In this context, there is now a digital concept known as an “AI hallucination”.

The scenario described above reflects that, to date, there has been limited progress among accountants in the correct use of AI and reveals one of the most common mistakes they make. Of course, accountants who attempt to use AI do so with the objective of improving their efficiency.

This raises a key question: What is the correct way to use AI in our accounting and tax practice?

The answer is simple: by training it properly. To achieve this, we must follow the following basic guidelines:


1. Build a Knowledge Dataset:

A dataset is a structured and consistent source of knowledge used to train AI, enabling it to identify patterns and provide more accurate responses.

In the accounting/tax field, a dataset may include:

  • The complete set of IFRS.
  • Interpretations and pronouncements issued by the IASB.
  • International guidance published by leading audit firms.
  • The Peruvian Tax Code
  • General tax laws
  • Case law and related doctrines

 

2. Design Your Prompts:

You do not need to be an expert in prompt engineering to know how to enter a prompt into an AI tool. However, an effective prompt should be structured with the following components:

  • Role: Define the profile the AI must ssume to respond.
    Example:
    “Act as an accountant specialized in the application of IFRS and guidance issued by leading audit firms”.
     
  • Knowledge Dataset: Specify which previously provided sources the AI must use to support its analysis.
    Example:
    “Base your analysis on international guidance issued by leading global audit firms”.
     
  • Transaction Dataset: Explain the context of the information and describe the contents of any files provided.
    Example:
    An Excel file containing a journal with transaction data, including columns for dates, accounts, descriptions, and amounts.
     
  • Specific Task: Define the analysis you want the AI to perform with the data provided.
    Example:
    “Identify duplicate journal entries”, “Reconcile revenue accounts against monthly income statements”, Validate the foreign currency translation effect for transactions in U.S. dollars and euros”.
     
  • Restrictions and Rules: Establish clear limits to prevent inaccurate results.
    Example:
    “Only consider amounts greater than USD 1,000”, “Use only the amounts recorded in the general ledger; do not hallucinate numbers or figures”.
     
  • Output Format: Indicate how you want the results to be presented.
    Example:
    “Provide a summary table of all transactions showing discrepancies between the general ledger and the cash and banks book”.

 

3. Information Confidentiality

A critically important aspect is the protection of business information. Accountants handle highly sensitive data; therefore, any information shared with an AI tool must be anonymized and/or masked.

Moreover, it is advisable to use AI tools that offer protection guarantees; otherwise, the confidentiality of company information, such as contracts, accounting books, financial statements, tax calculations, or detailed account movements may be compromised.


It is also worth pointing out that there are AI tools that use official datasets and act as legal/tax specialists, such as VLEX/Vincent, Lexius Peru, and Lex.pe. These platforms offer solutions that ensure data confidentiality and significantly reduce hallucinated responses.

The information provided above are initial guidelines for accountants to start using this powerful tool effectively. Today, it is no longer sufficient to simply know how to use data processing software or advanced spreadsheet functions. Our profession is evolving rapidly and there is a growing need for professionals that are experts in the use of advanced digital automation and AI technologies.

AI is here to stay, and we can no longer delay its adoption. We must train ourselves in order to train our AI systems. While AI will replace those accountants who do not know how to use it, those who learn how to integrate it effectively will optimize their time and will improve their professional productivity.

In our accounting and tax services, we use specialized, state-of-the-art AI tools that enable agile data processing and the dynamic generation of financial reports tailored to your business needs.

 
1 Instructions or queries entered into an artificial intelligence (AI) system.
2 Peruvian tax authority responsible for the collection and audit of corporate income taxes.

Contact Us