The role of artificial intelligence in the tax administration of bankruptcy proceedings of insolvent organizations

Krichevsiki Evgenii Nikitich Postgraduate student; Department of Taxes and Tax Administration; Financial University under the Government of the Russian Federation 109147, Russia, Moscow, Marxistskaya str., 9, sq. 121 evgenyikrichevsky@mail.ru Other publications by this author

The relevance of research. In modern conditions of the development of information systems and resources of the Federal Tax Service of Russia, as well as the gradual introduction of electronic document management into the life cycles of enterprises, one of the acute problems is the need to maintain, analyze and identify the risks of a huge array of information about taxpayers. The main advantage of digitalization and the introduction of AI in the tax administration of insolvent organizations is the speed of debtor data analysis, automation of control processes, obtaining information about the debtor in real time, as well as reducing the labor costs of tax authorities and accelerated interdepartmental interaction. As a result, the tax authorities are reducing their efforts to identify the possibility of unfair asset withdrawal, questionable transactions, etc.

The novelty of this study lies in the adaptation of tools and the implementation of AI mechanics in the tax administration of insolvent organizations (bankrupts).

The purpose of the presented research is to explore the possibility of using AI in the tax administration of bankruptcy procedures for insolvent organizations, to demonstrate examples of successful international experience and to suggest ways to adapt it to the needs of the Russian system.

The object of the study is the economic relations formed during the bankruptcy procedure of insolvent organizations.

The basis of the research methodology is formed on the basis of general scientific and special research methods, including methods of comparative analysis, a method of summarizing results when formulating conclusions and presenting priority areas, a method of system analysis and expert assessment.

The results of the study. The tax administration of insolvent organizations (bankrupts) in the Russian Federation (hereinafter referred to as the Russian Federation) has a number of difficulties, primarily related to: an increase in taxpayers in bankruptcy proceedings (as of 31.12.2022 – 301.3 thousand, as of 31.12.2023 – 365.8 thousand), significant time and financial costs for procedures, lack of automation and paperwork document management. In this article, the definitions of "insolvency" and "bankruptcy" are synonymous, in accordance with Article 2 of Federal Law No. 127-FZ dated October 26, 2002 "On Insolvency (Bankruptcy)".

Due to the completion of the moratorium on bankruptcy proceedings on 01.10.2022 on applications from creditors and unilateral sanctions against the Russian Federation, the number of organizations involved in bankruptcy proceedings is increasing, however, most of the analyzed indicators show positive dynamics due to the development of information systems and resources of the Federal Tax Service of Russia aimed at balancing the interests of all participants in bankruptcy proceedings, as well as financial recovery of enterprises (especially systemically important ones).

According to the Unified Federal Register of Legally Significant Information on the facts of the Activities of Legal Entities (hereinafter referred to as the Fedresource), individual entrepreneurs and other economic entities, the degree of satisfaction of creditors' claims for completed cases has been at an extremely low level for many years, despite positive trends in the theory and practice of bankruptcies. (Table 1).

Table 1 - Main results of bankruptcy proceedings in completed debtor bankruptcy cases

An analysis of the results of the activities of the Federal Tax Service of Russia in bankruptcy cases representing the interests of the Russian Federation as a creditor shows that the main difficulty in ensuring the interests of the Russian Federation in modern bankruptcy cases is that the interests of participants in bankruptcy cases suffer from low efficiency in the implementation of bankruptcy procedures (understood as the degree of satisfaction of creditors' claims), despite active improvement bankruptcy legislation, a change in the vector of tax policy on tax debt collection towards conciliation procedures and client-centricity, as well as the development of information systems and resources.

At the same time, the number of corporate bankruptcy cases initiated by the Federal Tax Service of Russia is increasing every year (January-June 2023 – 12%, January-June 2024 – 26.4%). This trend once again proves the need for AI and algorithms aimed at improving the efficiency of tax administration of insolvent organizations (bankrupts).

Modern economic scientists regularly discuss the issue of introducing AI into tax administration and bankruptcy institutions. Thus, according to A.N. Ryakhovskaya, the undoubted advantage of digitalization in the bankruptcy institute is the ability to control, timely and conveniently obtain information on almost any issue and various documents, including from government agencies.

In addition, modern scientists such as O.I. Lyutova and N.I. Rudneva are of the opinion that it is necessary to integrate AI into the Automated Information System Tax-3 and a number of other automated systems in the arsenal of the Federal Tax Service of Russia, and also support the integration of technologies in order to optimize and automate the implementation of other functions assigned to tax authorities.

M.V. Polinskaya, M.A. Chailyan and A.A. Yeghizaryan, studying the role of AI in tax administration, came to the conclusion that tax software using AI is an innovative direction that can improve and systematize the work of the entire field of tax administration as a whole. The researchers note that AI can only act as an assistant in solving tasks at various levels and requires human control. We should agree with the opinion of the authors, because with the development of AI, information systems and resources, the number of civil servants, including those who administer bankruptcy procedures, is decreasing. However, control over the functioning of the technical processes being developed remains solely for humans. S.E. Kozyreva also notes the need for strict AI control, as well as the introduction of legal liability in order to offset negative consequences.

E.V. Kuzmina notes that the introduction of chatbots that are able not only to provide advice to individuals, but also to calculate taxes and insurance premiums. In addition, taxpayers can use it to make an appointment with any institution of the tax authorities. Due to the demand and simplicity of chatbot programming, it is possible to implement mechanisms in tax administration, in terms of obtaining interdepartmental information about the debtor.

In their study, V.I. Malyar and E.A. Aksenova analyzed the development of information systems and resources in bankruptcy proceedings using the example of the Federal Resource. The authors note the urgent problem of the risk of loss or distortion of digital information, attempts at unauthorized access to it, as well as the emergence of additional responsibility to participants, requiring transparency and completeness of data. We should agree with this opinion, because in the era of the development of information systems and resources of the Federal Tax Service of Russia and the Institute of Bankruptcy in general, there is an urgent need to encrypt and protect information about the debtor, as well as filtering information obtained from open sources.

At the same time, the research of N.V. Apatova and V.B. Popov, regarding the use of neural networks in predicting bankruptcy procedures of insolvent organizations, is of particular interest. The authors have built a neural network designed to solve forecasting problems. Neural network characteristics such as non-linearity and a good generalization ability make it possible to successfully learn to solve complex problems and produce correct results for new initial data on the state of enterprises.

At the same time, it seems possible to use neural network mechanisms for tax administration purposes, in terms of "preventing" bankruptcy, as well as identifying distortions in the debtor's financial statements.

A.S. Mestnikova, analyzing the possibility of applying and adapting machine learning for the purposes of predicting bankruptcy of legal entities, identifies 4 of the 95 most significant attributes that affect bankruptcy forecasting.: the ratio of debt to equity, the ratio of current liabilities to current assets, the debt ratio and the ratio of current liabilities to assets. These attributes can be used in developing the machine learning process and creating an accurate and effective model.

Of particular interest was the research of Arinichev I.V., Matveeva L.G. and Arinicheva I.V., regarding the application of the logistic regression method, decision trees and neural networks for data analysis, revealing high accuracy in predicting bankruptcy. Scientists have concluded that it is possible to use these methods for early detection of bankruptcy risks and preventive measures, which significantly increases the effectiveness of financial risk management in companies. It is possible to expand this study and apply the logistic regression method to predict the bankruptcy of backbone enterprises.

Foreign scientists have developed a large number of different mathematical and economic models. It should be noted that L. Cultrera's bankruptcy forecasting model "Logit" was developed on the basis of Belgian enterprises. The Logit model is based on control variables such as size and operating time. T. Korol presented a model for EU companies that compares methods of fuzzy sets, neural networks and decision trees. The assessment included an analysis of the decline in efficiency in the 10 years prior to bankruptcy. At the same time, the most well-known and frequently used methods for predicting bankruptcy are W. Beaver's one-factor analysis and discriminant analysis, reflected in R. Tafler's research.

D. Bugachi and A. Alhavaldeh, explore the application of methods such as Support Vector Machines, random forests, and neural networks to process and analyze data specific to financial institutions.

The study compares the effectiveness of different models in predicting bankruptcy, emphasizing the importance of choosing the appropriate model based on the specifics of the data. The authors emphasize the importance of adaptive approaches when working with diverse financial datasets in order to take into account the unique features and challenges associated with bankruptcy prediction.

Kim Hyunjun, Hong Cho, and Dudzin Ryu explore the use of machine learning techniques to analyze temporary financial data in bankruptcy forecasting tasks. The authors focus on recurrent neural networks (RNNs), including their advanced variants such as LSTM and GRU. The authors analyze the specifics of using these models to process financial time series, including indicators of liquidity, changes in debt structure, profit and cash flow. Special attention is paid to how consistent data can improve the accuracy of forecasts, since time dynamics is often an important indicator of a company's financial stability.

Hoang Hyep Nguyen, Jean-Laurent Viviani and Sami Ben Jaber conducted a study on the application of machine learning methods to predict corporate bankruptcy. Particular attention is paid to the use of methods for interpreting results based on additive explanatory methods, such as the Shapley Values. This approach helps to make the models more transparent and understandable, which is important for practical application in the financial sector. The authors combine powerful machine learning algorithms (for example, gradient boosting, random forests) with interpretation methods to improve understanding of the factors influencing the results of forecasts.

Regarding the study of the possibility of using AI in the bankruptcy procedure of backbone enterprises, the study of S. Letkovsky, S. Jenkova and P. Vasanikova should be noted. Scientists have developed an author's model for predicting bankruptcy in the chemical industry, applicable in Slovakia. The results showed that the use of AI-based methods does not reduce the accuracy of forecasting. On the contrary, these methods can improve the accuracy of forecasting, especially in the long term.

An analysis of modern scientific research allows us to conclude that the mechanisms for implementing AI in the bankruptcy procedure of insolvent organizations need to be adapted for tax administration purposes. In terms of analyzing foreign experience, we can conclude that it is popular in terms of developing models (algorithms) for preventing bankruptcy. However, there is a gap in the application of these models and methods for tax administration purposes. Also, a review of foreign studies has shown that most models are focused on predicting the insolvency procedure.

The tax administration of insolvent organizations includes the analysis of a large amount of data: accounting and tax statements, accounting documents, transactions, court decisions, etc. Traditional methods of data processing require significant time, which increases the average duration of bankruptcy proceedings to 3 years. Thus, the main challenges in the Russian practice of tax administration of insolvent organizations are:

1. The volume and complexity of the analyzed data received by the tax authorities. As part of on-site and on-site tax audits, inspectors have difficulty analyzing information reflected in taxpayers' tax and accounting records, and reconciling information reflected in tax returns and primary accounting documents.

2. Lack of qualified personnel. Due to the formation of a two-tier system in many regions of the Russian Federation, which involves the transfer of functions of territorial tax authorities to the relevant Departments of the Federal Tax Service of Russia for the constituent entities of the Russian Federation, there is a reduction in jobs and an outflow of civil servants (in 2023, 2024). Thus, there are fewer and fewer specialists in the field of bankruptcy tax administration, and a large amount of time and resources is spent on training new ones and assigning appropriate qualifications.

3. Manual data processing. Most of the data received by the Federal Tax Service of Russia is sent electronically. However, not all information systems and resources of the Federal Tax Service of Russia are able to translate received documents in the "screenshot" format into a digital document, due to the lack of a single form, format, as well as the procedure for sending and presenting such information. In this regard, inspectors are forced to manually verify and identify discrepancies in the information received.

The potential of AI in the tax administration of insolvent organizations (bankrupts) lies in the fact that AI is able to automate complex analysis processes, including:

- Classification of data. AI systems can categorize data (balance sheet, transactions, reporting) and simplify their processing.

- Anomaly detection, identification of distortions and risks: Machine learning algorithms detect time differences, distortions in the net asset value and, accordingly, by the amount of the expected value of the competitive mass.

- Making forecasts. Predictive models based on AI help determine the outcome of bankruptcy proceedings for insolvent organizations.

International experience in the use of AI in the work of tax and judicial authorities. In order to qualitatively assess the need to use AI in the tax administration of insolvent bankrupt organizations, it is necessary to consider examples of AI applications in the USA, Great Britain and China, including the functionality of platforms, their areas of application and results.

The US experience: "Lex Machina" and "CaseMine". Functionality of the Lex Machina platform:

- analysis of court cases to identify patterns of behavior of debtor companies;

- automatic generation of reports for courts, creditors and participants in the process;

- predicting the outcome of cases based on historical data and similar processes;

- assessment of defense or attack strategies in bankruptcy cases.

Functionality of the CaseMine platform:

- Analyzing case law to create an optimal case management strategy;

- identification of key court decisions that may affect the outcome of the current case;

- integration with jurisdictional databases for faster search of necessary information.

For the purposes of tax administration of bankruptcy procedures of insolvent organizations:

- automatic analysis of debtors' assets and liabilities;

- preparation of recommendations for liquidation commissions;

- forecasting the probability of successful completion of the case in favor of creditors.

According to the American Bankruptcy Institute, the use of Lex Machina has reduced the number of errors in submitted documents by 25%. UK experience, FraudNet platform functionality:

- scanning of transactions, banking data and corporate reports;

- building graphs of relationships between companies, beneficiaries and assets;

- identification of potentially fraudulent transactions based on historical data and patterns of behavior;

- Generate reports for court proceedings and investigations.

For the purposes of tax administration of bankruptcy procedures of insolvent organizations:

- search for hidden assets, such as property registered with third parties;

- investigation of complex fraudulent schemes using offshore companies;

- checking the credit history of debtor companies to prevent fictitious bankruptcies.

According to the UK Insolvency Service, in 85% of cases, debtors' assets can be returned to creditors, including hidden or unaccounted for funds. The detection of fraudulent schemes at an early stage led to a 40% reduction in damage to creditors. The FraudNet platform has allowed to increase the number of disclosed assets by 35% in 2023. The time required to analyze suspicious transactions has been reduced from a few weeks to several days.

The Chinese experience. The functionality of the social credit system.

- combining information from banks, tax authorities, state registers and courts;

- analysis of transactions, debt obligations and payment discipline of companies;

- credit rating assessment by assigning points based on the transparency and reliability of the company;

- creation of registers of trustworthy and unreliable companies;

- prevention of fictitious bankruptcies through monitoring of operations;

- restrictions on the activities of low-rated companies, including a ban on new loans;

- companies with high ratings receive benefits and access to government contracts.

For the purposes of tax administration of bankruptcy procedures of insolvent organizations:

- reducing the number of insolvent companies by 25% due to early detection of risks;

- increasing trust between businesses and lenders;

- reduction of time for bankruptcy procedures due to data transparency.

According to the State Committee for Development and Reform of the People's Republic of China, in 2022, the social credit system helped repay more than 2 trillion yuan of debts. Companies with low ratings faced a 30% reduction in transaction volume.

International experience shows that the use of AI significantly speeds up and simplifies bankruptcy procedures for legal entities. Platforms such as Lex Machina, CaseMine, FraudNet, and the Chinese social Credit system demonstrate high potential for data analysis, predicting outcomes, and fraud prevention. Their use makes it possible to significantly reduce time and financial costs, increase transparency of processes and protect the interests of creditors. The analysis of best practices can become the basis for the development and implementation of similar solutions in other countries, including the Russian Federation.

The mechanics of AI implementation in the tax administration of insolvent organizations in Russia.

As part of the solution to the above problems regarding the introduction of AI in the tax administration of insolvent organizations (bankrupts), several stages of implementation are proposed.:

1. Development of an algorithm (model) for carrying out a reformation of the balance sheet indicators of bankrupt organizations in order to eliminate the distorting effect of continued application of the assumption of continuity and deferred taxes on the value of net assets (hereinafter referred to as the net asset value) and, accordingly, on the expected value of the bankruptcy estate.

Table 2 - Algorithm for reforming the balance sheet indicators of bankrupt organizations in order to eliminate the distorting effect of deferred taxes on the value of net assets

At the same time, the algorithm assumes to reveal the influence of balance sheet indicators of deferred tax assets (hereinafter referred to as IT) and deferred tax liabilities (hereinafter referred to as IT) on the calculated indicators of the NSA in bankruptcy proceedings, consisting in the fact that the indicators of IT and IT (in conditions of inapplicability of the assumption of continuity) - distort the indicators of the NSA and, accordingly, the calculated the size of the competitive mass. At the same time, the fundamental principles of taxation enshrined in Article 3 of the Tax Code of the Russian Federation (hereinafter referred to as the Code) are violated.

2. The introduction of machine learning technology into the SKUAD integrated debt management system for processing bank statements requested by tax authorities in accordance with Articles 93.1 and paragraph 2 of Article 86 of the Code. At the moment, no single algorithm has been created that allows pinpointing discrepancies in the debtor's bank transactions, as well as highlighting the risk levels of certain operations, using the example of a "traffic light" with red, yellow and green risk levels. The integration of the mechanics of processing bank statements will expand the digital portrait of the debtor and increase the efficiency of tax administration.

3. Detection of anomalies in debtors' bank transactions.

Using machine learning algorithms such as Decision Trees and Random Forest to search for abnormal transactions. The use of these algorithms in the work of tax authorities will make it possible to identify transactions made with affiliated companies before the bankruptcy procedure.

4. Automation of interaction with arbitration managers.

AI can be used to create chatbots that process management requests, analyze data, and provide information in real time.

Conclusion. The introduction of AI into the work of the Federal Tax Service of Russia, in terms of tax administration of insolvent organizations, is a major vector towards improving the effectiveness of bankruptcy procedures for insolvent organizations. At the same time, the development of the bankruptcy institute with the gradual introduction of AI contributes to: the development of predictive analytics, the construction of models for predicting the outcome of cases, the identification of distortions in the debtor's reporting and balance sheet, reducing the cost of administering bankruptcy procedures, reducing the time spent on data processing.

International experience demonstrates the high demand and effectiveness of such technologies. The adaptation of the analyzed tools to Russian practice will open up new opportunities for digitalization, increase the transparency of tax administration, and also contribute to the development of international interdepartmental electronic interaction.