💻 computer science

Artificial Intelligence-Enabled Accounting Information Systems and Fraud Detection in Nigeria's Financial Services Sector: The Moderating Role of Natural Language Processing

This study demonstrates that Artificial Intelligence-enabled Accounting Information Systems significantly enhance fraud detection and auditing effectiveness in Nigeria's financial services sector, with Natural Language Processing acting as a positive moderator that improves semantic interpretation and analytical explainability.

Original authors: Timothy Oluwapelumi Adeyemi, Abigail Omotola Ojogbede

Published 2026-07-03

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Timothy Oluwapelumi Adeyemi, Abigail Omotola Ojogbede

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine Nigeria's financial system as a massive, bustling marketplace where money moves faster than a heartbeat. In the past, the "guards" watching this marketplace (the auditors) used to check the receipts after the day was over. They looked for obvious mistakes or known tricks using a simple checklist. But today, the "thieves" have upgraded their tools. They aren't just stealing cash; they are using complex digital disguises, fake identities, and speed to slip through the cracks before anyone notices.

This paper is like a report card on a new set of high-tech security guards that Nigerian banks and financial companies are trying out. Here is the breakdown of what the study found, using simple analogies:

1. The Problem: The Old Flashlight vs. The New Storm

The study explains that the old way of checking for fraud is like trying to find a needle in a haystack using a dim flashlight at night. You can only see what's right in front of you, and you have to wait until the mess is made to start looking.

The Reality: Modern fraud is like a storm that happens in seconds. It moves too fast for the old "flashlight" (traditional auditing) to catch it.
The Solution: The paper looks at AI-enabled Accounting Systems. Think of this as replacing the dim flashlight with a satellite camera that watches the whole marketplace 24/7, in high definition, and can spot a tiny movement from a mile away.

2. The Main Hero: The "Smart Satellite" (AI)

The researchers asked 186 financial experts in Nigeria (bankers, auditors, and tech workers) if these new "Smart Satellites" (AI systems) actually help catch bad guys.

What they found: Yes, they work. The study says that when banks use these AI tools, they get much better at:
- Prevention: Stopping the crime before it happens (like a smart lock that knows a thief is trying to pick it).
- Detection: Spotting the crime as it happens.
- Investigation: Figuring out how it happened quickly.
The Big Winner: The study found that Prevention is the most powerful part. It's better to have a guard who stops the thief at the door than one who just chases them after the money is gone.

3. The Secret Weapon: The "Translator" (NLP)

Here is where it gets interesting. The AI "Smart Satellite" is great at looking at numbers (like checking if $100 was transferred). But fraudsters are also good at lying with words. They write fake emails, fake reports, and confusing messages to hide their tracks.

The Gap: A computer is usually bad at reading human lies or sarcasm in a text message.
The Fix: The paper introduces Natural Language Processing (NLP). Think of NLP as a super-smart translator that sits next to the AI.
- While the AI looks at the numbers, the NLP reads the stories (emails, complaints, reports).
- It can tell if a sentence sounds "shady" or if a report is trying to hide the truth, even if the numbers look okay.
The Result: The study found that when you add this "Translator" (NLP) to the "Smart Satellite" (AI), the whole system gets even better. It's like having a security guard who can not only see the thief but also understand the thief's secret code.

4. The Verdict

The paper concludes that in Nigeria's fast-moving digital world:

AI Systems are essential for catching fraud because they are faster and smarter than human-only checks.
NLP is the "secret sauce" that makes the AI even smarter by helping it understand human language and hidden meanings.
Together, they make the financial system safer, more transparent, and harder for criminals to trick.

In short: The paper says that to catch modern digital thieves in Nigeria, you can't just use a rulebook anymore. You need a high-tech AI system that watches the numbers and a smart translator that reads the words, working together to stop fraud before it steals the money.

Technical Summary: Artificial Intelligence-Enabled Accounting Information Systems and Fraud Detection in Nigeria's Financial Services Sector

1. Problem Statement

The rapid digitalization of financial systems in emerging economies, particularly Nigeria, has enhanced operational efficiency and financial inclusion but simultaneously increased exposure to sophisticated cyber-enabled fraud. Contemporary fraud schemes—ranging from synthetic identity fraud and phishing to coordinated digital deception networks—often bypass conventional auditing mechanisms. Traditional systems, which rely on retrospective verification, rule-based monitoring, and periodic sampling, struggle to address the velocity, complexity, and adaptive nature of modern financial crime.

Furthermore, a significant gap exists in the literature regarding the application of these technologies within emerging African financial ecosystems. Existing empirical evidence is heavily concentrated in technologically advanced Western and East Asian economies, leaving the operational realities of fragmented infrastructures and evolving regulatory conditions in countries like Nigeria underexplored. Specifically, there is limited understanding of how Natural Language Processing (NLP) moderates the relationship between AI-enabled Accounting Information Systems (AIS) and auditing effectiveness, particularly regarding the interpretation of unstructured textual data and the explainability of AI-driven decisions.

2. Methodology

Research Design and Philosophy:
The study employs a positivist research philosophy and a quantitative cross-sectional survey design. It utilizes a hypothetico-deductive approach to empirically test theory-driven hypotheses regarding the relationships between AI-enabled AIS, NLP, and fraud detection effectiveness.

Data Collection:

Population: Professionals in auditing, accounting, compliance, fraud management, and financial risk governance within Nigeria's regulated financial sector (commercial banks, insurance firms, and FinTech organizations).
Sampling: A stratified purposive sampling technique was used to ensure representation across sectors. From a sampling frame of 310 professionals (sourced from ICAN and CIBN directories), 186 valid responses were retained (74.8% response rate).
Instrument: A structured questionnaire with 47 items measuring eight constructs: Data Gathering, Data Analysis, Risk Assessment, Detection, Prevention, Investigation, Auditing and Fraud Detection Effectiveness, and NLP Capability. Items were measured on a five-point Likert scale.
Validity and Reliability: Content validity was established via expert review. Construct validity was assessed using Exploratory Factor Analysis (EFA), removing items with loadings below 0.60. Internal consistency was confirmed with Cronbach's alpha values exceeding 0.70 for all constructs. Common method bias was assessed via Harman's single-factor test (34.6% variance).

Analytical Strategy:
Data were analyzed using IBM SPSS Statistics 27 and AMOS 27. The analysis followed four stages:

Descriptive statistics to summarize respondent profiles and variable distributions.
Diagnostic testing for normality, multicollinearity (VIF < 10), and residual assumptions.
Multiple linear regression to examine the influence of AI-enabled AIS dimensions on auditing effectiveness.
Hierarchical moderated regression to test the moderating role of NLP, following the procedures of Baron and Kenny and Cohen et al., with mean-centered variables to reduce multicollinearity.

3. Key Contributions

Contextual Empiricism: The study provides empirical evidence from Nigeria, an emerging economy characterized by rapid digitalization and infrastructural fragmentation, thereby extending the applicability of intelligent auditing research beyond developed Western and East Asian contexts.
Theoretical Integration: It anchors the investigation in the Fraud Diamond Theory (incorporating capability as a fourth dimension of fraud) and the Technology Acceptance Model (TAM) (focusing on perceived usefulness and ease of use). This dual framework explains both the structural conditions facilitating fraud and the institutional dynamics driving technology adoption.
Moderating Role of NLP: The research specifically isolates and tests the moderating effect of NLP, addressing a gap in literature regarding how semantic interpretation capabilities enhance the effectiveness of AI-assisted fraud detection in environments where unstructured textual data (e.g., audit reports, compliance narratives) is critical.

4. Results

Descriptive Findings:
Respondents generally held positive perceptions of AI-enabled AIS and NLP capabilities. "Data Analysis" (Mean = 3.96) and "Detection" (Mean = 3.93) were rated highest, while "Data Gathering" and "Investigation" scored lower, reflecting potential infrastructural and integration challenges.

Hypothesis Testing (H1):
Multiple regression analysis revealed that AI-enabled AIS dimensions collectively explain approximately 62.6% of the variance in auditing and fraud detection effectiveness ( $R^2 = 0.626$ , $p < 0.001$ ).

Prevention emerged as the strongest predictor ( $\beta = 0.384, p < 0.001$ ), indicating that proactive, predictive monitoring is the most critical component of effective fraud governance.
Detection ( $\beta = 0.173, p < 0.001$ ), Data Gathering ( $\beta = 0.149, p = 0.001$ ), and Data Analysis ( $\beta = 0.117, p = 0.015$ ) also showed statistically significant positive influences.
Risk Assessment showed a positive but marginally insignificant relationship ( $p = 0.058$ ), suggesting its effectiveness may depend on complementary institutional conditions.

Moderation Analysis (H2):
Hierarchical moderated regression confirmed that NLP significantly moderates the relationship between AI-enabled AIS and auditing effectiveness.

The inclusion of NLP improved the model's explanatory power ( $\Delta R^2 = 0.035, p = 0.002$ ).
The interaction term (AI $\times$ NLP) was statistically significant ( $\Delta R^2 = 0.013, p = 0.041$ ).
The results indicate that NLP acts as a complementary capability, enhancing the effectiveness of AI systems by extending analysis from structured numerical data to unstructured textual environments (e.g., audit narratives, disclosure statements), thereby improving semantic interpretation and investigative precision.

5. Significance and Claims

The paper claims that AI-enabled AIS and NLP are increasingly vital for strengthening fraud governance, regulatory accountability, and institutional trust within emerging digital financial environments.

Practical Significance: The study argues that financial institutions must transition from retrospective, rule-based auditing to continuous, intelligence-driven monitoring. It highlights that preventive capabilities are the most effective mechanism for mitigating fraud losses. Furthermore, it posits that NLP is essential for addressing the "black-box" nature of AI by translating algorithmic outputs into comprehensible audit narratives, thereby enhancing explainability and regulatory defensibility.
Theoretical Significance: The findings extend the Fraud Diamond Theory by demonstrating how AI reduces the "opportunity" dimension of fraud through continuous surveillance and automated compliance. Simultaneously, the results reinforce the TAM by suggesting that institutions adopt these technologies primarily due to their perceived usefulness in improving governance performance and operational efficiency.
Limitations: The authors modestly acknowledge that the cross-sectional design limits causal inference and that reliance on perceptual survey data may introduce bias. They suggest future research should employ longitudinal or mixed-method approaches to examine long-term organizational implications.

In conclusion, the study asserts that while AI-enabled AIS provides the foundational architecture for modern fraud detection, the integration of NLP is a critical moderating factor that enhances analytical scope, interpretability, and overall governance effectiveness in complex, data-intensive financial ecosystems like Nigeria's.