Page 28 - CMA Journal (Jul-Aug 2026)

F oc us S ec t ion
AI-Powered Fraud Detection in Public Procurement:
Transforming Audit and Tax Compliance in the Digital Economy
Introduction Economic Crime Survey
2024 found that
Public procurement contains governments and procurement fraud
state-owned enterprises attaining goods, services, and ranked among the top
works, revealing substantial public funding that
produces prospects of e ciency and ethical three (along with
procurement practices (Sampaio et al., 2026). Yet, fraud revenue fraud and
in the buying and sourcing procedure is a major issue, transfer pricing fraud)
and high amounts of money are lost in the process, and most disruptive
public con dence eroded (Alves Batista, 2024). It is economic frauds in the
projected that the global economy loses more than 5% past two years,
of its GDP due to corruption, and the leakage of over impacting 59% of all the
one trillion dollars in bribes annually (Sampaio et al., respondents. Likewise,
2026). Procurement fraud is often among the top three only 26% of Mahwish Mahmood
most disruptive economic crimes the world over, after organizations leverage Inspector Inland Revenue
cybercrime and corruption (PwC, 2024). data analytics to identify Federal Board of Revenue (FBR)
unusual bidding
As procurement systems have engraved their path into
the virtual world, transactional information has grown behavior, which is signi cant as there is still plenty of
in important and substantial volumes, providing potential to make use of the technology in areas of
opportunities as well as di culties for fraud detection. detecting potential fraud techniques (PwC, 2024).
Over time, the traditional fraud detection means based Limitation of Traditional Approaches
on rules are becoming inadequate to deal with the
volume and complication of procurement data (OECD, Conventional methods of fraud deterrence and
2024). Arti cial Intelligence (AI), including machine detection are profoundly reliant on manual audits,
learning (ML), o ers a transformative mechanism– and rules-based systems and random sampling methods.
methods to deduct, envisage and lessen fraud and The complicated and extensive nature of the
corruption threat in a scale not possible before with procurement processes makes it challenging for these
traditional means (OECD, 2024). methods to deduct highly developed fraud designs
(Salazar et al., 2024). A high number of false-positives is
It looks at how technologies powered by AI can be generated by rules-based systems, which are
used for Fraud Detection in Public Procurement mainly non- exible to new types of fraud and human auditors
through Audit and Tax Compliance Applications. The
analysis re ects three main areas – (1) Anomaly struggle in processing large volumes of data (Sampaio
et al., 2026). Furthermore, outdated approaches are
detection with AI for detecting procurement fraud, (2)
AI applications for monitoring tax compliance, and (3) reactive and not practical, meaning fraud is deducted
AI changes to the audit process with automation. after a signi cant amount of money has been lost, and
Furthermore, the ethical and practical insinuations of not in real-time.
using AI in this eld are observed, with a precise focus AI-Powered Fraud Detection
on the challenges of depending on AI tools so broadly. Technologies
The Landscape of Procurement Fraud Machine Learning for Anomaly Detection
and Tax Evasion Machine learning algorithms are most superior in
The Scale of the Problem analyzing complex information, and provide excellent
Procurement fraud can be considered as any type of capability to detect anomalies that might easily be
illegal act that is committed with the intention of overlooked by human auditors. AI algorithms can
securing economic or other advantages in the examine massive amounts of data in just a few minutes
procurement process of goods/services or contracts when handling with public procurement data, tax
(Taxmann, 2024). Some general illustrations are invoice records, or government expenses, which supports
fraud, shell company schemes, supplier collusion, deduction anomalies that could show bribery, fraud, or
con ict of interest and bid rigging. The PwC Global embezzlement (Atta-ur-Rahman, 2025).
26 ICMA’s Chartered Management Accountant, Jul-Aug 2026
Machine learning permits these mechanisms to further help detect corrupt activities in public
advance and get better over time, as they are exposed procurement, such as repeated bidder groups, unusual
supplier activity, etc., which can aid in doing due
to increasingly more data to ag suspicious activity.
Supervised learning datasets consist of transactions diligence including risk assessments. The use of this
methodology was applied to information from Chilean
which are labelled as fraudulent and legitimate, and
classi cation models are developed using these public procurement, where it showed its possibilities
datasets. Machine learning (ML) algorithms have of improving transparency and risking the reduction of
proven successfully e ective in sleuthing and corruption.
prioritizing investigations of reported purchases in AI Use-Cases in Audit and Tax
enterprise data, as this helps advance the e ectiveness
Compliance
of an organization’s control for the bene t of the
organization. Rather, unsupervised learning AI-Enhanced Audit Processes
techniques, like clustering and anomaly detection, AI is substantially changing audit processes in the
detect patterns that deviate from those the training public sector. Audits of bene ciary schemes using the
data has seen, and can therefore be utilized in spotting supremacy of AI/ML have exposed many more cases of This comprises a detailed concern about algorithmic compliance systems restructure and increase
new fraud patterns, without the requirement of frauds, evading the misuse of huge nancial resources. bias, such as the Dutch tax agency's penalty for tens of e ciency and e ectiveness. But ethical challenges
labelled training information. AI-driven forensic audits can reveal fraud trends which thousands of families, mostly from lower-income and regarding transparency, fairness and accountability
VigIA is a tool developed to monitor public can only be deducted by fraud machine learning. ethnic groups, on using a self-learning algorithm to need to be improved, which can be done by
procurement using machine learning models in order deduct bene ts fraud (Anonymous, 2024). developing systems like PRO-Trust. Therefore, to
to assist detection of ine ciencies as cost overrun and In the domain of auditing, intelligent invoice The PRO-Trust Framework appreciate the advantages and reduce the dangers of
delivery delay, and risk indicators to highlight veri cation systems prove to be a practical use of AI. AI, procurement professionals and auditors will need
irregularities in the public procurement process Enhanced Apriori algorithm has very high veri cation The PRO-Trust pipeline delivers an ethics-focused to invest in developing their technical expertise and
(Salazar et al., 2024). In this data-driven method, the e ciency, and its application, particularly in nancial framework to building reliable AI-based Fraud the framework of ethical governance.
robots for normal, abnormal taxation, fraud supplier Detection solutions for the sector of Public
oversight agencies can focus their scarce resources on References
highest-risk contracting activities. Risk Guard is also invoice, can be used in practice. These systems enable Procurement. PRO-Trust covers ve recurring ethical • Alves Batista, D. (2024). Enhancing transparency and accountability in public
questions system opacity, user transparency, explain
one of the leaders in machine learning and transparent you to scan the procurement invoices against the procurement: Exploring blockchain technology to mitigate records fraud. Records
decision-support tools that detect unusual procurement contracts, the delivery record, and tax ability instruments and, role of the operator, and Management Journal, 34(2-3), 151–170. https://doi.org/10.1108/RMJ-
10-2023-0054
procurement patterns and risks, which include led documents on real-time basis. strategic consideration of ethical principles with • Anonymous. (2024). Applied Machine Learning to Anomaly Detection in Enterprise
supervised and unsupervised machine learning, AI for Tax Compliance Monitoring measures that can be taken during the advancement Purchase Processes. Harvard ADS. https://ui.adsabs.harvard.edu
/abs/2024arXiv240106234A
network science, and statistical/probabilistic modeling AI is intensifying tax administrations' capabilities to of the AI-based technology, deduction of the • Anonymous. (2024). Arti cial intelligence for detecting and preventing
(Anonymous, 2024). responsible actors, and application of technological procurement fraud. OUCI. https://ouci.dntb.gov.ua/en/works/7Xxg0G9Q/
reduce fraud and corruption by re ning risk strategies to operationalize ethical principles. In • Anonymous. (2024). Challenges in Addressing the Ethical Aspects of Arti cial
Natural Language Processing for Fraud Detection management processes, accompanying taxpayer conclusion, PRO-Trust puts its e orts into the making Intelligence to Detect Fraud in Public Procurement Processes. OUCI.
https://ouci.dntb.gov.ua/en/works/7Xxg0G9Q/
Using Natural Language Processing (NLP) practices, compliance, and decreasing taxpayer burdens (OECD, of more reliable, transparent and accountable AI-based • Anonymous. (2024). Detection of Structured Fraud Supported by Shell Companies
anyone can get automated insight into unstructured 2024). Tax administrations can evade non-compliance fraud detection systems in public administration. on Goods and Services Trading Operations. Mendeley. https://www.mendeley.com/
by nding ways to mediate at earlier stages of taxpayer • Anonymous. (2025). Data-Driven Transparency: Machine Learning and Social
procurement documents, such as tender documents, phases. AI-driven network analysis, blockchain Challenges and Future Directions Network Analysis for Corruption Detection in Public Procurement. ScienceDirect.
https://www.sciencedirect.com/science/article/pii/S1877050925029722
assessments, and contracts. The study directly aimed at veri cation, and computer vision have already proven Though there is great promise, there are also • Anonymous. (2025). Intelligent Invoice Dynamic Veri cation for Accounting-Type
evaluating the potential of automated corruption risk operative for customs fraud detection, including under important issues to reduce when it comes to AI and its Financial Robots Based on an Improved Apriori Algorithm. Semantic Scholar.
https://www.semanticscholar.org/
indicator detection and identi cation in procurement valuations, smuggling, and false goods classi cation enactment in the world of procurement fraud • Atta-ur-Rahman. (2025). AI against corruption. The News. https://www.thenews.
texts has been completed by analyzing state-of-the-art detection. Data quality and availability remain to be com.pk/print/1315923-ai-against-corruption
technologies for automated corruption risk indicator (Anonymous, 2024). Similarly, AI solutions in VAT fraud • European Commission. (2024). Member States' measures to combat tax fraud and
protect EU nancial interests. EUR-Lex. https://eur-lex.europa.eu/
detection and making a comparative analysis of these have e ectively drive to the investigation of carousel vigorous issues and AI practices demand large, • OECD. (2024). Arti cial Intelligence: Anti-Corruption and Integrity Outlook 2024.
high-quality datasets for training and validation
technologies, and suggesting the application of the fraud, fake invoice systems, and handlings of refunds, OECD Publishing. https://www.oecd.org/
Contextual Transformer Models for scalable, and therefore allowed tax authorities to recover purposes. Capacity limitations in the organization are a • PwC. (2024). Global Economic Crime Survey 2024. PwC. https://www.pwc.com/
hurdle to implementation, including lack of technical
transparent and operational anti-corruption billions of dollars in loss in tax revenues. knowledge and/or resistance to change (OECD, 2024). • Salazar, A., Pérez, J. F., & Gallego, J. (2024). VigIA: Prioritizing public procurement
oversight with machine learning models and risk indices. Data & Policy, 6.
monitoring systems (Salazar et al., 2024). Ethical Challenges and PRO-Trust Though there is a massive investment in AI solutions https://www.semanticscholar.org/paper/VigIA%3A-prioritizing-public-procurement-o
versight-Salazar-P%C3%A9rez/de20e997a6d549da1655e838bd1f795a25fcb526
Social Network Analysis Applied to Collusion Framework for procurement fraud detection, the market is still in • Sampaio, I. G. B., de Oliveira Andrade, E., de Souza Barbosa Fontes, S., Bernardini, F.
C., & Viterbo, J. (2026). Trustworthy AI in public administration: The PRO-Trust
its infancy, having reached USD 1.98 billion with a
Detection Ethics of AI Fraud Prevention CAGR of 23.6% in 2024 (Anonymous, 2025). pipeline for ethical fraud detection in public procurement. Government
Information
Quarterly.
https://www.sciencedirect.com/science/article/pii/
Traditional tools of detection fail to work as well with
S0740624X26000547
supplier collusion and con icts of interest. With the As AI is employed in fraud detection, there are Conclusion • Taxmann. (2024). Combatting 'Procurement Fraud' – Classi cation | Key Risks |
help of Machine Learning and Social Network Analysis, substantial ethical challenges at stake, including Mitigation Strategies. Taxmann. https://www.taxmann.com/
it is feasible to identify fraudulent supplier respect for human privacy, prevention of harm, AI technologies transform fraud detection, audit
relationships. Machine Learning together with Social equality, and explain ability (Sampaio et al., 2026). procedures and tax compliance in public procurement.
Network Analysis can be a powerful tool to detect System opacity, lack of user transparency, automated The use of machine learning algorithms and natural
explain ability tools con nes, lack of operational error
suspicious supplier relationships (Anonymous, 2025). and lack of strategic control are recurring ethical language processing (NLP), along with social network
analysis, can support in assessing fraudulent patterns
CASE Studies: Anti-Fraud S using AI. The models can problems. on a large scale, and AI-driven audit tools and tax