AI-driven financial fraud detection in Pakistan’s banking sector: bridging strategic intent and operational implementation

Classified as probably not a substantial claim. May combine two claims.

The relationship between perceived usefulness/ease of use and acceptance is significantly conditioned by technological readiness and perceived AI usability, and ethical and integrity considerations moderate these relationships.

“Recent empirical studies suggest that the relationship between perceived usefulness/ease of use and acceptance is significantly conditioned by technological readiness and perceived AI usability, and that ethical and integrity considerations moderate these relationships (Al-Maaitah et al. 2025).”

• cites Al-Maaitah et al. 2025 sourcePartly supported

  The theoretical framework (section 1.1; page unavailable) discusses perceived usefulness and ease of use as factors in technology adoption. But the study tests user readiness, ease of use, and AI adoption tools as predictors of sustainability accounting education—not whether readiness and AI usability condition a usefulness/ease-of-use-to-acceptance relationship. Hypothesis testing (section 3.5 and Table 4; page unavailable) finds significant academic-integrity moderation for readiness and AI adoption tools, but not for ease of use (p = 0.277). (LLM, conf=0.90)

  Experimental — accuracy not yet measured

May combine two claims.

AI-based techniques such as machine learning models for anomalous behavior detection and biometric analyses can quickly analyze large volumes of transactional data and detect and flag potentially fraudulent activity in real time.

“Techniques based on AI, such as machine learning models for anomalous behavior detection and biometric analyses, can provide a way to quickly analyze large volumes of transactional data and detect and flag potentially fraudulent activity in real time (Bhatla et al. 2003; Jang-Jaccard and Nepal 2014; Homer 2020).”

• cites Bhatla et al. 2003Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites Jang-Jaccard and Nepal 2014 sourcePartly supported

  Section 3.4 (page unavailable) describes self-learning anomaly detection that examines network traffic, often in real time, and reports abnormal behavior. It does not establish that biometric analyses or these models rapidly process large volumes of transactional data to flag potentially fraudulent transactions; its discussion of biometrics concerns authentication. (LLM, conf=0.88)

  Experimental — accuracy not yet measured
• cites Homer 2020 sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Ownership structure and state involvement can moderate the relationship between board characteristics and ESG performance, influencing how governance mechanisms shape responsible technology use in emerging and Asia-Pacific markets.

“Recent studies further demonstrate that ownership structure and state involvement can moderate the relationship between board characteristics and ESG performance, influencing how governance mechanisms shape responsible technology use in emerging and Asia-Pacific markets (Al-Tahat et al. 2025).”

• cites Al-Tahat et al. 2025 sourcePartly supported

  The Introduction states that state ownership can moderate the relationship between board characteristics and ESG performance in Asia-Pacific markets. The paper does not connect that finding to responsible technology use, so that extension of the citing sentence is unsupported. (LLM, conf=0.88)

  Experimental — accuracy not yet measured

Probably not a substantial claim: may not pass S3 (load-bearing).

Governance, accountability, and strategic alignment are central to realizing long-term value from AI adoption.

“Bibliometric evidence on digital technologies and corporate sustainability further indicates that governance, accountability, and strategic alignment are central to realizing long-term value from AI adoption (Alshdaifat et al. 2024).”

• cites Alshdaifat et al. 2024 sourcePartly supported

  The Introduction links AI to sustainability and social and environmental governance. Section 3.7 treats accountability as important to sustainable blockchain practices, while the Conclusions recommend strategic partnerships and policy frameworks for digital technologies. The paper does not establish that governance, accountability, and strategic alignment together are central to long-term value from AI adoption; in particular, its accountability finding concerns blockchain, not AI. (LLM, conf=0.83)

  Experimental — accuracy not yet measured

Classified as probably not a substantial claim. May combine two claims.

Managers cannot rely solely on AI investment to achieve strategic benefits; they need complementary investments in human resources, governance structures, and operational processes to convert technical capabilities into strategic performance.

“The most important implication for managers is that they cannot rely solely on AI investment to achieve strategic benefits; they need to make complementary investments in human resources, governance structures, and operational processes to convert technical capabilities into strategic performance (Mikalef et al. 2020).”

• cites Mikalef et al. 2020 sourcePartly supported

  Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

  Experimental — accuracy not yet measured

May combine two claims.

Inadequate training directly impacts the "perceived ease of use" factor of the Technology Acceptance Model (TAM), which impacts employee's confidence in utilizing AI output and applying it to make decisions in their day-to-day work activities.

“Inadequate training directly impacts the “perceived ease of use” factor of the Technology Acceptance Model (TAM), which impacts employee’s confidence in utilizing AI output and applying it to make decisions in their day-to-day work activities (Taherdoost 2021).”

• cites Taherdoost 2021 sourceCould not verify

  Split into 2 atoms — 0 confirmed, 0 partially supported, 0 disputed, 0 misattributed, 2 unverifiable. Expand the claim to see per-atom verdicts.

  Experimental — accuracy not yet measured

Probably not a substantial claim: may not pass S3 (load-bearing).

Recent bibliometric analyses have found increasing overlap between AI, corporate social responsibility (ESG), and governance accountability.

“A number of recent bibliometric analysis have found that there is increasing overlap between AI, corporate social responsibility (ESG) and governance accountability (Alhasnawi et al. 2024; Alshdaifat et al. 2024).”

• cites Alhasnawi et al. 2024 sourceNot checked — no full text of the source
• cites Alshdaifat et al. 2024 sourcePartly supported

  The Introduction links AI with sustainability and social and environmental governance, while §3.2 reports growing publication activity concerning AI and sustainability (page numbers unavailable in our copy). The paper does not establish an increasing overlap specifically among AI, corporate social responsibility/ESG, and governance accountability. (LLM, conf=0.82)

  Experimental — accuracy not yet measured

May combine two claims.

In Pakistani banks, AI-enabled fraud detection systems are theoretically capable of rapidly identifying anomalies, but their preventive capabilities are limited by organizational preparedness, system integration, and workforce capacity.

“Although theoretically capable of rapidly identifying anomalies, interviews suggest that these systems’ preventive capabilities remain limited by factors such as organizational preparedness, system integration, and workforce capacity.”

May combine two claims.

Advanced AI techniques such as machine learning, anomaly detection, deep learning, and natural language processing are technically available in Pakistani banks, but their operational utilization remains constrained.

“Though advanced techniques such as machine learning, anomaly detection, deep learning, and natural language processing are technically available, their operational utilization remains constrained.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

India's Banks have employed Behavioural Biometric methods for fraud prevention.

“In comparison, India’s Banks have employed Behavioural Biometric methods for fraud prevention”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

In the West, there has been an increase in Hybrid Models being used, which incorporate both Behavioural Biometric and Anomaly Detection Methods.

“whereas, as can be seen from the West, there has been an increase in Hybrid Models being used, which incorporate both Behavioural Biometric and Anomaly Detection Methods.”

May combine two claims.

Both bank management and staff are concerned about how to deal with government regulators who have rules and guidelines for using AI technology but there are no specific rules or guidelines available yet.

“Both bank management and staff are concerned about how to deal with government regulators who have rules and guidelines for using AI technology but there are no specific rules or guidelines available yet.”

May combine two claims.

Operational staff at branches have limited opportunity to engage with, receive sufficient training in, or make decisions using AI-based fraud detection systems.

“while the operational staff at branches have limited opportunity to engage, receive sufficient training or to make decisions using AI-based fraud detection systems”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

Concerns from regulatory bodies over data privacy and the lack of a defined regulatory framework also expressed concerns about potential liability resulting from false positives.

“Concerns from regulatory bodies over data privacy and the lack of a defined regulatory framework also expressed concerns about potential liability resulting from false positives.”

Classified as probably not a substantial claim.

IT officials stated that without additional training for their employees at the branch level, they could not effectively use all the technological features available within the AI system.

“IT officials stated that without additional training for their employees at the branch level, they could not effectively use all the technological features available within the AI system.”

May combine two claims.

Branch Managers and Compliance Officers have repeatedly indicated that AI Systems are located centrally within Head Offices, thus excluding Branches from being involved with them.

“Branch Managers and Compliance Officers have repeatedly indicated that AI Systems are located centrally within Head Offices, thus excluding Branches from being involved with them.”

May combine two claims.

Although AI is used to detect fraudulent transactions through anomaly monitoring and classification, the use of AI to prevent fraud is still in its infancy.

“Although AI is being used to detect fraudulent transactions through anomaly monitoring and classification, the study shows that the use of AI to prevent fraud is still in its infancy.”

May combine two claims.

Bank managers are willing to use and develop AI technology, but they rely on it mostly for reporting and compliance purposes rather than as a tool to enhance their day-to-day operations.

“While bank managers are willing to use and develop AI technology, they rely on it mostly for reporting and compliance purposes rather than as a tool to enhance their day-to-day operations.”

May combine two claims.

Interviewees reported that they had little or no knowledge of technology and were inadequately trained on the technical capabilities of AI.

“Interviewees reported that they had little or no knowledge of technology and were inadequately trained on the technical capabilities of AI.”

May combine two claims.

AI-driven technologies such as anomaly detection and NLP for transaction data have potential, but interviewees noted staff training, data quality issues, and compliance concerns.

“AI-driven technologies, such as anomaly detection and NLP for transaction data, have potential; however, interviewees noted staff training, data quality issues, and compliance concerns.”

May combine two claims.

AI-based fraud detection systems in Pakistani banks are primarily managed at the head-office level, with limited branch-level autonomy.

“AI-based fraud detection systems are primarily managed at the head-office level, with limited branch-level autonomy.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

Branch managers know nothing about AI tools, while upper management handles fraud detection.

“Branch managers know nothing about AI tools, but upper management handles fraud detection.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Uncertainty around accountability, explainability, and data privacy reduces trust in AI systems.

“Uncertainty around accountability, explainability, and data privacy reduces trust in AI systems.”

May combine two claims.

In Pakistan, banks have started implementing AI-based solutions such as BPC's SmartVista platform to continuously monitor all transaction types across various channels, and they also use link analysis to detect complex fraud networks.

“For instance, in Pakistan, banks have started implementing AI-based solutions (e.g., BPC’s SmartVista platform) to continuously monitor all transaction types across various channels, while also using link analysis to detect complex fraud networks (Bhatla et al. 2003).”

• cites Bhatla et al. 2003Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S4 (based).

If the underlying organizational capabilities such as training programs and data management practices do not exist, the effectiveness of AI-based fraud detection systems will be severely limited.

“In short, if the underlying organizational capabilities do not exist (i.e., training programs and data management practices), the effectiveness of AI-based fraud detection systems will be severely limited.”

Classified as probably not a substantial claim.

Characteristics of boards of directors, oversight mechanisms, and other elements of governance are significant predictors of digital preparedness, cybersecurity disclosure, and the effective governance of advanced technologies such as AI.

“Governance research demonstrates that characteristics of boards of directors, oversight mechanisms, and other elements of governance are significant predictors of digital preparedness, cybersecurity disclosure, and the effective governance of advanced technologies such as AI (Alodat and Mansour 2024).”

• cites Alodat and Mansour 2024Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The State Bank of Pakistan is developing policies to facilitate the responsible and transparent use of AI in financial services as part of its larger agenda to encourage innovation in financial technology.

“The SBP is developing policies to facilitate the responsible and transparent use of AI in financial services, and is doing so as part of its larger agenda to encourage innovation in financial technology (FinTech).”

Probably not a substantial claim: may not pass S3 (load-bearing).

Data collection continued until thematic saturation was reached.

“Data collection continued until thematic saturation had been reached.”

Classified as probably not a substantial claim.

A systematic literature search was undertaken using multiple databases, including Google Scholar, ScienceDirect, SpringerLink, Scopus, Web of Science, and State Bank of Pakistan reports.

“A systematic literature search was undertaken using several databases, including: Academic and Industry Databases; Google Scholar; ScienceDirect; SpringerLink; Scopus; Web of Science; as well as Reports from Regulators/Professional Bodies (i.e., the State Bank of Pakistan).”

Probably not a substantial claim: may not pass S3 (load-bearing).

Banks in neighboring India have leveraged AI technologies to enhance their fraud prevention efforts.

“banks in neighboring India have also leveraged AI technologies to enhance their fraud prevention efforts (see Malali and Gopalakrishnan 2020).”

• cites Malali and Gopalakrishnan 2020Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Classified as probably not a substantial claim.

Data from literature reviews and qualitative interviews were triangulated to enhance analytical rigour, validity, and contextualization.

“Data from both literature reviews and qualitative interviews were deliberately triangulated so as to compare data from the qualitative interviews to literature reviews to enhance analytical rigour, validity, and contextualization.”

Probably not a substantial claim: may not pass S3 (load-bearing).

According to SBP surveys, approximately half of Pakistan's regulated financial entities have either developed or are in the process of developing AI tools for fraud detection, customer service, credit assessment, and other purposes.

“Reports that according to surveys conducted by the SBP, approximately half of Pakistan’s regulated financial entities have either developed or are in the process of developing AI tools for fraud detection, customer service, credit assessment, and other purposes.”

• cites SBPNot matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

In Pakistan's banking sector, AI-based fraud management tools are typically deployed centrally and do not have a direct link to fraud detection at the branch level.

“Findings in Pakistan’s banking sector further reinforce these results; while Pakistan’s banks utilize AI-based fraud management tools, these tools are typically deployed in a central manner and do not have a direct link to fraud detection at the branch level.”

Probably not a substantial claim: may not pass S4 (based).

Moderating factors-regulatory oversight, ethical issues such as data privacy and algorithmic bias, organizationally-related issues such as infrastructure and systems integration, and employee knowledge/skill levels-have a significant impact on how successful AI-based fraud prevention systems will be.

“These moderating factors have a significant impact on how successful AI-based fraud prevention systems will be.”

Classified as probably not a substantial claim. May combine two claims.

Studies were included only if they used AI or machine learning for fraud detection in banking or financial services, were review, regulatory, or evidence-based studies, and were written in English.

“Studies were included if they: (1) used an artificial intelligence or machine learning-based approach to detect fraud and were applied in the banking or financial services sector; (2) were review studies, regulatory/industry reports, or employed an evidence-based methodology; (3) were written in English.”

May combine two claims.

Pakistan's banking system offers a specific context for AI fraud detection with both strong drivers for innovation and significant operational constraints.

“Pakistan’s banking system offers a very specific context for fraud detection using AI, where there are both a number of strong drivers for innovation and significant operational constraints.”

• AssumptionRests on an assumption: Pakistan has a young and tech-savvy population and a very high mobile phone penetration rate that creates a large opportunity for digital banking to grow and for AI to help.
• AssumptionRests on an assumption: There are recently reported figures showing a 12.2 percent year-over-year increase in AI adoption within Pakistan’s banking sector that is primarily focused on improving the risk management and fraud prevention capabilities of banks.
• AssumptionRests on an assumption: Banks in Pakistan are experimenting with AI in a variety of areas, including mobile app security and back-end risk analysis.
• AssumptionRests on an assumption: BPC’s SmartVista platform has been used by multiple banks in Pakistan (including Muslim Commercial and Askar Banks Limited) to enable them to monitor transactions in real time and prevent fraud.

Classified as probably not a substantial claim.

According to TAM, the adoption of AI-based fraud detection systems is influenced, at least in part, by bankers' perceptions of the utility and usability of the technology.

“According to TAM, the adoption of AI-based fraud detection systems will be influenced, at least in part, by the perceptions of bankers of the utility and usability of the technology.”

• cites TAMNot matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

New regulations are intended to require banks to define a clear risk appetite for AI systems and address compliance, security, and the environmental impact of AI.

“The new regulations are intended to require banks to define a clear risk appetite for AI systems and address compliance, security, and the environmental impact of AI.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

These are rare examples of how Pakistani-based Banks use AI for preventative measures because they lack Branch Level Expertise.

“however these are rare examples of how Pakistani-based Banks use A.I. for Preventative Measures because they lack Branch Level Expertise.”

Probably not a substantial claim: may not pass S4 (based).

Perceived ease of use of AI systems affects employees' perceived ability to interpret AI output and integrate it into normal operational processes.

“The perceived ease of use of AI systems affects how employees perceive their ability to interpret AI output and integrate it into their normal operational processes.”

• AssumptionRests on an assumption: TAM and RBV are valid theoretical frameworks for explaining AI adoption and effectiveness in fraud detection.

May combine two claims.

Many AI-based projects in the financial services industry in developing economies fail to perform adequately due to lack of workforce skills, inadequate digital platforms, and inability to integrate strategic intent and operational execution.

“Developing economy evidence, as presented by, shows that many AI-based projects in the financial services industry fail to perform adequately, resulting in a lack of workforce skills to implement these technologies, inadequate digital platforms to support their deployment, and an inability to effectively integrate strategic intent and operational execution.”

Probably not a substantial claim: may not pass S4 (based).

Few studies have investigated the intersection of technological factors, such as algorithm efficacy and user acceptance of AI tools, with organizational factors, such as availability of skilled employees and manager support, related to the deployment of AI for fraud detection.

“Notably, few studies have investigated the intersection of technological factors (i.e., the efficacy of algorithms and user acceptance of AI tools) with organizational factors (i.e., availability of skilled employees, manager support, etc.) related to the deployment of AI for fraud detection.”

Classified as probably not a substantial claim.

From a TAM perspective, the degree to which employees perceive AI-based fraud detection as useful and the ease of using AI-based fraud detection systems directly relate to the usage and acceptance of AI-based fraud detection by employees.

“From a TAM perspective, the degree to which employees perceive AI-based fraud detection as useful and the ease of using AI-based fraud detection systems directly relate to the usage and acceptance of AI-based fraud detection by employees.”

• cites TAMNot matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

Significant deficiencies in IT infrastructure, data quality, and digital skills exist among Pakistani banks.

“Reported that significant deficiencies in IT infrastructure, data quality, and digital skills exist among Pakistani banks.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Quality assessments were cross-checked among reviewers, and disagreements were resolved through discussion until consensus was reached.

“To ensure that there was consistency in the review process and to minimize subjectivity bias, quality assessments were cross-checked among reviewers, and any disagreements were resolved through discussion until a consensus was reached.”

Classified as probably not a substantial claim.

Only studies assessed as high or medium quality were used for the final synthesis.

“Only studies that were assessed as either high or medium quality were used for the final synthesis.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

Perceived usefulness of AI systems impacts branch managers' and front-line staff's beliefs that the system improves fraud detection, reduces manual workloads, and supports timely decision making, thus promoting active employee involvement with AI-generated alerts.

“The perceived usefulness of AI systems impacts how much branch managers and front-line staff believe that the system improves fraud detection, reduces manual workloads, and supports timely decision making, thus promoting the active involvement of employees with AI-generated alerts.”

• AssumptionRests on an assumption: TAM and RBV are valid theoretical frameworks for explaining AI adoption and effectiveness in fraud detection.

May combine two claims.

Institutional pressures affect both whether a bank adopts AI systems and how it designs, governs, and implements AI-based fraud detection systems.

“These institutional pressures affect not only whether a bank will adopt AI systems but also how they will design, govern, and implement AI-based fraud detection systems.”

• AssumptionRests on an assumption: The State Bank of Pakistan’s emerging AI and digital risk governance guidelines emphasize transparency, accountability, data privacy, and ethical use of automated systems.

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

Fraud prevention is connected to other bank-wide outcomes including improved financial results, better compliance, and increased customer trust.

“In addition to addressing fraud prevention, the framework illustrates how fraud prevention is connected to other bank-wide outcomes. Examples of these outcomes include: improved financial results, better compliance, and increased customer trust.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

If employees find AI systems too difficult to understand or too opaque, employee engagement with the systems will decrease, resulting in underutilization despite significant strategic investments.

“If employees find AI systems too difficult to understand or too opaque, then employee engagement with the systems will decrease, resulting in the underutilization of the systems despite significant strategic investments.”

• AssumptionRests on an assumption: TAM and RBV are valid theoretical frameworks for explaining AI adoption and effectiveness in fraud detection.

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

AI techniques-machine learning, deep learning, natural language processing, and hybrid approaches-feed into two separate but interrelated outcomes: fraud detection, comprising post-transaction monitoring and anomaly identification, and fraud prevention, comprising real-time blocking, biometric authentication, and predictive protection.

“outcomes, namely fraud detection (post-transaction monitoring and anomaly identification), and fraud prevention (real-time blocking, biometric authentication, and predictive protection).”

Classified as probably not a substantial claim.

The quality of all included studies was appraised using a modified Critical Appraisal Skills Programme (CASP) checklist.

“The quality of all included studies was appraised by a modified version of the Critical Appraisal Skills Programme (CASP) checklist.”

Probably not a substantial claim: may not pass S4 (based).

Perceived usefulness of AI systems increases a manager's willingness to invest in an AI system.

“Perceived usefulness increases a manager’s willingness to make an investment in an AI system”

• AssumptionRests on an assumption: TAM and RBV are valid theoretical frameworks for explaining AI adoption and effectiveness in fraud detection.

Probably not a substantial claim: may not pass S4 (based).

The adoption and utilization of AI-based fraud detection systems depend on employee attitudes toward perceived usefulness and perceived ease of use.

“From a TAM viewpoint, the adoption and utilization of AI-based fraud detection systems are dependent on employee attitudes toward the perceived usefulness and perceived ease of use of those systems.”

• AssumptionRests on an assumption: TAM and RBV are valid theoretical frameworks for explaining AI adoption and effectiveness in fraud detection.

Classified as probably not a substantial claim.

The hypotheses are treated as theory-informed analytical expectations that guide the analysis and interpretation of the research findings.

“Therefore, the hypotheses are viewed as theory-informed analytical expectations that guide the analysis and interpretation of the research findings.”

May combine two claims.

Together, AI techniques form a layered fraud-prevention architecture; alerts generated by machine learning models might trigger higher-level reviews using graph network analysis or prompt additional biometric verification.

“Together, these techniques form a layered fraud-prevention architecture; alerts generated by machine learning models might trigger higher-level reviews using graph network analysis or prompt additional biometric verification.”

• AssumptionRests on an assumption: There is no one fraud detection model/technique that will prevent all types of fraud.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The researcher analyzed discrepancies between the literature and interview findings rather than treating them as anomalous, increasing the interpretive validity of the findings.

“When discrepancies arose between the expected theoretical and empirical patterns found in the literature and the actual findings from the interviews, the researcher chose to analyze these discrepancies rather than treat them as anomalous, thereby increasing the overall interpretive validity of the findings from the study.”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

If governance-by-design principles are not followed, technologically sophisticated fraud detection systems may face limited adoption, mistrust, and suboptimal utilization.

“If such “governance by design” principles are not followed, technologically sophisticated fraud detection systems may face limited adoption, mistrust, and suboptimal utilization.”

May combine two claims.

The SBP's draft guidelines on responsible AI use emphasize applications in fraud detection and risk management and aim to ensure that banks utilize AI ethically and effectively.

“The SBP’s draft guidelines on responsible AI use specifically emphasize applications in fraud detection and risk management and aim to ensure that banks utilize AI ethically and effectively.”

• cites SBPNot matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

The study does not use statistical analysis to test causal relationships between variables; it uses TAM and RBV as lenses for qualitative analysis and to interpret themes from the literature review and interviews.

“This study does not utilize statistical analysis to test causal relationships between variables, but instead utilizes TAM and RBV as lenses for qualitative analysis and to provide a structured method for interpreting themes identified through the literature review and interview methods.”

• cites TAMNot matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites RBVNot matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

The study increased rigor by comparing the thematic structure of the interview findings to the findings of the Systematic Literature Review (SLR).

“In addition to conducting an axial coding method, the researcher attempted to increase the overall rigor of this study by comparing the thematic structure of the study’s findings to the findings of the Systematic Literature Review (SLR).”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

Pakistani bank managers exhibit moderate confidence in AI but express concerns about cultural and organizational preparedness, including resistance to change and lack of employee training.

“Likewise, reported that while Pakistani bank managers exhibit moderate confidence in AI, they express concerns about cultural and organizational preparedness (i.e., resistance to change or lack of employee training).”

May combine two claims.

A lack of training negatively influences the ease of use perception (perceived ease of use), and reinforces the interdependence of acceptance at the individual level and capabilities at the firm level, as indicated in the research framework.

“Additionally, a lack of training negatively influences the ease of use perception (perceived ease of use), and reinforces the interdependence of acceptance at the individual level and capabilities at the firm level, as indicated in the research framework.”

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

Bridging the gap between AI fraud detection and fraud prevention requires more than a technology upgrade; it also requires changes in governance, organizational preparedness, and employee training.

“Therefore, bridging this gap will require more than just an upgrade in technology, but also changes in governance, organizational preparedness, and the training of employees.”

May combine two claims.

Limited exposure to an organization's AI system limits the perceived usefulness and ease of use by the front line staff which limits operational usage.

“In terms of TAM, limited exposure to an organization’s AI system limits the perceived usefulness and ease of use by the front line staff which limits operational usage.”

May combine two claims.

Indian banks use AI to implement both behavioral biometric monitoring and real-time fraud prevention at the customer interface level and are able to do so because of better digital infrastructure and employee training.

“1) Indian banks use AI to implement both behavioral biometric monitoring and real-time fraud prevention at the customer interface level and are able to do so because of better digital infrastructure and employee training (Malali and Gopalakrishnan 2020).”

• cites Malali and Gopalakrishnan 2020Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

A limitation of the study is its focus on selected banks in Pakistan.

“Limitations of this study are: (a) focus on selected banks in Pakistan”

May combine two claims.

The Technology Acceptance Model explains operational staff resistance to AI-based fraud detection: their limited exposure to such systems leads them to perceive the systems as neither useful nor easy to use.

“the Technology Acceptance Model (TAM) provides insight into why the operational staff resist the adoption of AI-based fraud detection, based on the fact that their limited exposure to AI-based fraud detection systems leads them to perceive such systems as being neither useful nor easy to use.”

• AssumptionRests on an assumption: The Resource-Based View and the Technology Acceptance Model are applicable frameworks for interpreting AI adoption and resistance in Pakistani banks.

Probably not a substantial claim: may not pass S3 (load-bearing).

There has been concern over explainability, transparency and clarity within regulatory frameworks.

“As discussed by Rezaee (2004), Peterson et al. (2005) in the literature there has been concern over explainability, transparency and clarity within regulatory frameworks.”

• cites Rezaee (2004) sourceNot checked — no full text of the source
• cites Peterson et al. (2005)Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S2 (checkable), S4 (based). May combine two claims.

Pakistani banks can better serve their customers by reducing the time spent on fraud management with AI, improving the security of their systems, and building customer trust.

“Pakistani Banks can better serve their customers by reducing the amount of time spent on fraud management with AI, improving the security of their systems and building customer trust.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

The absence of governance frameworks, compliance mechanisms, and ethical oversight is an example of an intangible organizational resource that enables organizations to deploy AI responsibly and at scale.

“The absence of governance frameworks, compliance mechanisms, and ethical oversight is an example of an intangible organizational resource that enables organizations to deploy AI responsibly and at scale.”

May combine two claims.

The barriers align with international governance standards for AI, such as the OECD AI Principles, which classify AI systems used in financial risk management as high-risk and require greater explainability, robustness, and accountability.

“These barriers align well with the international governance standards for AI, such as the OECD AI Principles and the, which classify AI systems used in financial risk management as high-risk and therefore require greater levels of explainability, robustness and accountability.”

• cites OECD AI PrinciplesNot matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

The State Bank of Pakistan's Draft AI Guidelines (2025) require responsible AI governance, model transparency, data protection and accountability within financial institutions.

“the State Bank of Pakistan’s Draft AI Guidelines (2025) address the aforementioned issues by requiring responsible AI governance, model transparency, data protection and accountability within financial institutions (State Bank of Pakistan 2024, 25).”

• cites State Bank of Pakistan 2024, 25Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S4 (based).

AI systems can provide technical support to an organization for detecting anomalies (thus supporting H2).

“Although AI systems can provide technical support to an organization for detecting anomalies (thus supporting H2)”

Classified as probably not a substantial claim.

Data privacy, explainability, and regulatory accountability concerns are very close to H3 as it is stated that institutional and cultural elements can limit or enhance the use of AI (H3).

“The data privacy, explainability, and regulatory accountability concerns are also very close to the H3 as it is stated that institutional and cultural elements can limit or enhance the use of AI (H3).”

May combine two claims.

Limited investment by banks in the education and skill building of their employees represents a limitation to strategic human capital, thereby hindering a bank's ability to develop sustainable organizational value through the conversion of its AI technological investments.

“from a Resource-Based View (RBV) perspective, limited investment by banks in the education and skill building of their employees represents a limitation to strategic human capital, thereby hindering a bank’s ability to develop sustainable organizational value through the conversion of its AI technological investments (Lee and Bruvold 2019).”

• cites Lee and Bruvold 2019 sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

These governance principles include transparency, accountability, fairness, data protection, and human oversight in high-risk AI applications such as financial services.

“These are based on several principles, including transparency, accountability, fairness, data protection and human oversight in high-risk AI applications, such as those in financial services.”

• cites OECD AI PrinciplesNot matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites European Union’s AI Act (2024)Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

The need to implement formal governance structures for AI in high-risk activities has been codified by numerous countries through legislation such as the OECD AI Principles and the EU AI Act (2024).

“As a result, the need to implement formal governance structures regarding the use of AI in high-risk activities has been codified by numerous countries globally through legislation such as the OECD AI Principles and the European Union’s AI Act (2024).”

• cites OECD AI PrinciplesNot matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites European Union’s AI Act (2024)Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S2 (checkable).

AI-based fraud detection should be seen as more than a technology implementation challenge: it is a strategic and organizational transformation challenge concerning governance, human capital development, and regulatory alignment.

“AI-based fraud detection should be seen as more than a technology implementation challenge, but rather as a strategic and organizational transformation challenge concerning governance, human capital development, and regulatory alignment.”

May combine two claims.

Prior research on AI in banking has largely focused on algorithm development and theoretical performance, with evidence that advanced analytics can improve fraud detection accuracy.

“Prior research on AI in banking has largely focused on algorithm development and theoretical performance, with evidence that advanced analytics can improve fraud detection accuracy (Hasham et al. 2019; Kaya et al. 2019).”

• cites Hasham et al. 2019Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites Kaya et al. 2019Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

Practical adoption of AI in banking is uneven and is often challenged by data and resource limitations.

“However, practical adoption is uneven and often challenged by data and resource limitations.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

If AI is not strategically integrated into a banking organization's strategy, it will be difficult for the organization's employees to utilize it and it will likely remain an unused technology system.

“If AI is not strategically integrated into the banking organizations strategy it will be difficult for the AI to be utilized by the banking organizations employees and will likely remain an unused technology system.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Emerging-market banks often encounter barriers such as limited technical infrastructure, regulatory constraints, and workforce skill gaps that can slow down AI implementation in fraud prevention.

“Emerging-market banks often encounter barriers such as limited technical infrastructure, regulatory constraints, and workforce skill gaps that can slow down AI implementation in fraud prevention.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

Regulatory frameworks demand explainability, auditability, and accountability, which opaque AI models do not easily meet.

“Compliance and Governance: Regulatory frameworks demand explainability, auditability, and accountability, which opaque AI models do not easily meet”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

Many banks lack the IT infrastructure to integrate AI tools with legacy systems, creating bottlenecks in real-time fraud monitoring.

“System Integration Issues: Many banks lack the IT infrastructure to integrate AI tools with legacy systems, creating bottlenecks in real-time fraud monitoring.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Informed consent was obtained from all participants before the interviews.

“Informed consent was obtained from all participants before the interviews.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Deep learning and ensemble methods often act as black boxes, making it difficult for managers to understand how decisions are generated.

“Interpretability and Transparency: Deep learning and ensemble methods often act as “black boxes,” making it difficult for managers to understand how decisions are generated.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Incomplete, imbalanced, or noisy datasets limit the effectiveness of machine learning models, especially in environments where fraudulent cases are under-reported.

“Data Quality and Availability: Incomplete, imbalanced, or noisy datasets limit the effectiveness of machine learning models, especially in environments where fraudulent cases are under-reported.”

May combine two claims.

Branch managers and frontline staff often lack training in AI systems, which prevents effective use and limits proactive fraud prevention.

“Workforce Skills Gap: Branch managers and frontline staff often lack training in AI systems, which prevents effective use and limits proactive fraud prevention.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Ethical approval for the study was obtained from the relevant institutional ethics committee (approval number IRB/2025/123, dated 30 March 2025).

“Ethical approval was obtained from the relevant institutional ethics committee, approval number IRB/2025/123, dated 30 March 2025.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

Even high-performing algorithms produce false positives and false negatives, which can erode user confidence and increase operational costs.

“Model Reliability and Accuracy: Even high-performing algorithms produce false positives and false negatives, which can erode user confidence and increase operational costs.”