Journal Home

OPEN ACCESS

e-Tax Filing and Assessment in the Age of Artificial Intelligence (AI): A Case of Bangladesh

Prof. Dr Kazi Abdul Mannan Department of Business Administration Faculty of Business Shanto-Mariam University of Creative Technology Dhaka, Bangladesh ORCID: https://orcid.org/0000-0002-7123-132X

Dr Khandaker Mursheda Farhana Associate Professor Faculty of Humanities & Social Sciences Department of Sociology & Anthropology Shanto-Mariam University of Creative Technology Bangladesh ORCID: https://orcid.org/0009-0009-1526-6147

G. M. Omar Faruque Chowdhury CEO GM Chowdhury & Co Dhaka, Bangladesh

Corresponding author: Prof. Dr Kazi Abdul Mannan: drkaziabdulmannan@gmail.com

Cap. mark. financ. rev.  2026, 6(2); https://doi.org/10.64907/xkmf.v6.i2.cmfr.3

Submission received: 1 April 2026 / Revised: 19 May 2026 / Accepted: 1 June 2026 / Published: 11 June 2026

Download PDF (727 KB)

Abstract

This study examines the transformative impact of artificial intelligence (AI) on Bangladesh’s tax administration, with a focus on the implementation and user acceptance of AI-enabled e-tax filing and assessment systems. Using the Unified Theory of Acceptance and Use of Technology (UTAUT) as the guiding framework, the research integrates quantitative data from 400 taxpayers and qualitative insights from key informant interviews with tax officials and policymakers. The findings reveal that performance expectancy, effort expectancy, and facilitating conditions have a significant influence on behavioural intention to adopt AI-based e-Tax systems. Correlation and regression analyses confirm the predictive power of these constructs, while qualitative data highlight institutional readiness challenges, ethical concerns, and gaps in digital literacy. The study uncovers both optimism and caution in AI adoption. While users appreciate improved efficiency, reduced errors, and faster processing, concerns persist regarding data security, algorithmic transparency, and unequal access across demographics. Institutional analysis suggests that, although the National Board of Revenue shows interest in AI integration, bureaucratic inertia and a lack of regulatory frameworks hinder the comprehensive implementation of AI. This research contributes to the growing literature on digital governance and AI in public administration by contextualising technological innovation within the socio-political and infrastructural realities of a developing country. It provides policymakers with actionable recommendations, including enhancing digital literacy, establishing ethical guidelines, and promoting inclusive design strategies to ensure equitable access and adoption.

Keywords: e-Tax, Artificial Intelligence, UTAUT, Bangladesh, Digital Governance, Tax Administration, Public Sector Innovation

1. Introduction

The rapid evolution of information and communication technology (ICT) has revolutionised the way governments manage public services, particularly in the realm of revenue administration. Among the most significant developments is the digital transformation of tax filing and assessment systems, commonly referred to as e-Tax. These systems streamline tax collection, improve compliance, and enhance the transparency of fiscal governance. In recent years, artificial intelligence (AI) has become a crucial component of digital tax ecosystems, enabling predictive analytics, real-time risk assessment, automated processing, and intelligent customer service. As countries strive to enhance tax administration, the integration of AI into e-tax platforms has become both a technological and policy imperative (OECD, 2021).

In developing nations like Bangladesh, where public revenue collection systems have historically been plagued by inefficiencies, corruption, and limited outreach, e-Tax platforms represent a pivotal step toward modernisation. The government of Bangladesh has committed to a vision of “Digital Bangladesh,” which emphasises the transformation of governance through ICT (World Bank, 2020). The National Board of Revenue (NBR) has played a central role in this transition, introducing electronic filing, online payment gateways, digital taxpayer identification numbers, and AI-supported chatbots for user assistance. Despite these advancements, the adoption and functionality of AI in e-Tax systems remain underexplored in academic literature.

Bangladesh presents a unique case for examining the intersection of AI and tax administration due to its socio-economic context, demographic diversity, and digital development trajectory. With a population exceeding 165 million and a growing number of internet users, the country offers both opportunities and challenges for digitised taxation. While urban areas have demonstrated a higher uptake of e-Tax services, rural regions still face infrastructural deficits, digital illiteracy, and limited access to online resources. Moreover, concerns surrounding data privacy, algorithmic transparency, and institutional capacity have become increasingly salient in public discourse (Rahman, 2019).

The role of AI in taxation extends beyond basic digitisation. AI-powered algorithms can process vast datasets to identify anomalies, detect fraudulent activities, and flag high-risk taxpayers for audit, thereby augmenting the decision-making capabilities of tax authorities (Alm & Soled, 2020). Moreover, natural language processing (NLP) enables AI chatbots to assist taxpayers in navigating complex filing procedures. Machine learning models, when properly trained on localised datasets, can automate repetitive administrative tasks and reduce human error in tax assessment. These capabilities, when effectively leveraged, can significantly enhance tax compliance and increase public trust in revenue authorities.

However, the deployment of AI in tax systems also introduces a set of complex challenges. Concerns over algorithmic bias, misuse of personal data, lack of explainability in decision-making processes, and dependency on imported technologies have been identified as potential risks (UN, 2021). These issues are particularly pronounced in developing countries where regulatory frameworks for AI are nascent or non-existent. Ensuring fairness, transparency, and accountability in AI applications requires not only technical solutions but also robust policy measures, ethical guidelines, and public engagement.

This research is timely and significant for several reasons. First, it contributes to the limited body of scholarly work on AI in the context of public finance in Bangladesh. Second, it addresses practical concerns related to digital inclusion, system transparency, and technological capacity that are relevant for policymakers and international development agencies. Third, it aligns with global efforts to promote efficient, inclusive, and equitable tax systems as part of the United Nations’ Sustainable Development Goals (SDGs), particularly Goal 16, which aims to promote just, peaceful, and inclusive societies.

1.1. Objectives of the Study

The primary objective of this study is to examine the role and effectiveness of Artificial Intelligence (AI) in enhancing the e-Tax filing and assessment system in Bangladesh. Specifically, the study aims to assess taxpayers’ perceptions, readiness, and acceptance of AI-enabled platforms from the perspective of the Unified Theory of Acceptance and Use of Technology (UTAUT). It also aims to explore the socio-demographic factors influencing user behaviour toward digital tax solutions. Additionally, the research examines the institutional and infrastructural preparedness of the National Board of Revenue (NBR) and its related stakeholders for implementing AI-driven systems.

Furthermore, the study aims to identify the benefits, challenges, and ethical considerations associated with the adoption of AI in public tax administration. By integrating both quantitative and qualitative data, the study aims to provide comprehensive insights into how AI can be effectively and equitably leveraged to improve tax compliance, efficiency, and transparency in Bangladesh. The ultimate goal is to inform policy recommendations that support inclusive, user-centred, and ethically sound digital taxation strategies in developing countries.

2. Theoretical Framework

To investigate the integration of artificial intelligence (AI) into Bangladesh’s e-tax filing and assessment infrastructure, a robust theoretical framework is essential. This section outlines and synthesises relevant theoretical perspectives that underpin the study, including theories of digital governance, technological acceptance, and algorithmic accountability. These theories help contextualise the implementation of AI in public tax administration and guide the analysis of both the opportunities and challenges associated with this transformation.

2.1 Digital Governance Theory

Digital governance theory explores how information and communication technologies (ICTs) reshape the functioning and structure of government institutions. According to Bannister and Connolly (2012), digital governance not only enhances service delivery but also transforms administrative processes, accountability mechanisms, and citizen engagement. In the context of Bangladesh’s e-Tax system, digital governance theory helps explain how AI technologies are integrated to enhance institutional capacity, reduce corruption, and improve transparency. The National Board of Revenue (NBR)’s initiatives, such as AI-powered chatbots and real-time fraud detection, illustrate digital governance principles in practice.

Bangladesh’s alignment with its “Digital Bangladesh” vision demonstrates an institutional commitment to leveraging technology for improved governance outcomes (World Bank, 2020). By applying digital governance theory, we can assess how AI transforms tax administration beyond automation, shaping organisational culture, public trust, and state-citizen relations.

2.2 Technology Acceptance Model (TAM)

The Technology Acceptance Model (TAM), developed by Davis (1989), is a foundational model for understanding users’ acceptance and utilisation of technology. TAM proposes that perceived usefulness and perceived ease of use significantly influence users’ attitudes toward technology adoption. This model is particularly relevant in assessing the receptivity of taxpayers and tax officials in Bangladesh toward AI-driven e-Tax systems.

As surveys in this study indicate, while 72% of taxpayers found the e-Tax system user-friendly, awareness of AI components remained low. This discrepancy suggests that, while overall system usability may be acceptable, the perceived value of specific AI features is not fully recognised. TAM can help interpret these findings by evaluating user experiences, attitudes, and behavioural intentions toward AI-enhanced digital tax platforms (Venkatesh & Davis, 2000).

2.3 Socio-Technical Systems Theory

Socio-technical systems theory views technological systems as inherently intertwined with social contexts. This perspective recognises that technological effectiveness is dependent on alignment between the technical tools and the social structures in which they operate (Trist & Bamforth, 1951). In the case of AI-based tax systems in Bangladesh, socio-technical theory highlights the importance of user literacy, institutional readiness, and infrastructural adequacy.

For instance, although AI can facilitate automated risk profiling and document processing, its success depends on the quality of training datasets, user competence, and institutional capacity. Disparities between urban and rural adoption of e-Tax systems further illustrate the socio-technical divide. Applying this theory enables a more holistic understanding of why AI systems may succeed in some settings while failing in others.

2.4 Algorithmic Governance and Accountability

As AI algorithms increasingly play a role in governance, concerns have emerged about their transparency, fairness, and ethical implications. Theories of algorithmic governance focus on how algorithmic decision-making reshapes traditional mechanisms of control, oversight, and accountability in public administration (Aneesh, 2006; Pasquale, 2015).

In tax systems, algorithmic tools determine audit targets, assess compliance risk, and assist users through AI chatbots. However, these systems often operate as “black boxes”-complex, opaque, and unexplainable to end-users. This opacity can lead to issues of trust, especially in contexts where algorithmic decisions may have material consequences (Eubanks, 2018).

In Bangladesh, algorithmic governance theory helps interrogate the implications of using imported AI technologies and unregulated algorithmic models in tax administration. It also highlights the need to develop localised, explainable AI systems and establish ethical guidelines for their deployment. Incorporating algorithmic governance into the theoretical framework provides a lens through which we can examine the balance between technological efficiency and democratic accountability.

2.5 Public Value Theory

Public value theory, as articulated by Moore (1995), posits that public institutions should create value not only through service efficiency but also by enhancing democratic values, public trust, and social equity. This theory is particularly relevant in assessing the broader impacts of AI in public tax administration.

While AI may improve operational efficiency, it must also serve the broader public interest. In Bangladesh, where public scepticism toward state institutions can affect compliance behaviour, AI-driven tax systems must be perceived as fair, inclusive, and transparent. Public value theory thus supports the notion that technological adoption should align with normative democratic goals, not just technical performance metrics.

The theoretical framework of this study is grounded in five interconnected perspectives: digital governance, technology acceptance, socio-technical systems, algorithmic governance, and public value theory. Together, these frameworks provide a comprehensive foundation for examining the multifaceted nature of AI implementation in Bangladesh’s e-Tax system. They enable us to analyse the technological, institutional, behavioural, and ethical dimensions of AI adoption, offering valuable insights into the conditions under which AI can enhance public sector performance.

3. Literature Review

The intersection of artificial intelligence (AI) and tax administration has gained increasing academic and policy interest over the past decade, especially in developing economies transitioning from manual to digital systems. This literature review examines scholarly perspectives on AI-driven tax systems, e-Tax infrastructure in developing countries, and the socio-technical and ethical dimensions of deploying AI in public governance. It aims to contextualise the development and application of AI in Bangladesh’s e-Tax filing and assessment system within global and regional academic discourse.

3.1 Evolution of e-Tax Systems

The transformation from traditional tax systems to electronic tax (e-tax) platforms has been a crucial component of digital governance globally. E-Tax systems are designed to improve tax compliance, transparency, and administrative efficiency. Countries such as Estonia and Singapore have demonstrated the potential of comprehensive digital tax infrastructures (OECD, 2021). These systems enable electronic filing, automated calculations, real-time verification, and prompt refunds. The World Bank (2020) notes that e-Tax initiatives in developing countries have led to increased tax revenue and decreased corruption through automation and digital traceability.

In South Asia, countries such as India and Sri Lanka have initiated digitised tax services, though varying degrees of success are recorded due to infrastructural, socio-economic, and political barriers (Kumar & Prasad, 2019). Bangladesh introduced e-TINs (Taxpayer Identification Numbers), electronic return submissions, and e-payment systems, aiming to streamline revenue collection and reduce administrative burdens (Ahmed & McGillivray, 2015; Mannan et al., 2021). However, challenges such as digital literacy, network coverage, and cybersecurity persist.

3.2 Artificial Intelligence in Public Administration

Artificial intelligence in public administration encompasses a wide range of applications, including data analytics, automated decision-making, natural language processing, and machine learning. In tax systems, AI is applied to identify tax evasion patterns, automate risk assessment, and provide interactive taxpayer assistance through AI-powered chatbots (Alm & Soled, 2020). For example, the U.S. Internal Revenue Service (IRS) employs AI tools for fraud detection and predictive audits.

AI applications in developing countries are more recent but growing rapidly. Several African and Southeast Asian governments have piloted the use of AI in digital ID verification and welfare distribution systems (UNDP, 2022). In tax administration, the effectiveness of AI depends on the availability of data, regulatory frameworks, and institutional capacity. According to Mikhaylov et al. (2018), the lack of robust data ecosystems and AI governance models remains a significant impediment to progress.

3.3 AI and e-Tax Systems in Developing Countries

Developing countries face unique challenges and opportunities in integrating AI into tax systems. On the one hand, AI can help overcome resource constraints by automating labour-intensive tasks and enhancing fraud detection capabilities. On the other hand, issues such as algorithmic opacity, public mistrust, and limited technical expertise can hinder the adoption of these technologies (Eubanks, 2018).

In Bangladesh, AI-based tax tools remain in a nascent phase. The National Board of Revenue (NBR) has piloted chatbots and machine learning algorithms to streamline taxpayer services and detect inconsistencies in filed returns (Rahman, 2019). However, academic studies specifically focusing on the impact of these technologies in Bangladesh remain limited.

3.4 Ethical and Legal Considerations

AI deployment in public finance must adhere to principles of accountability, transparency, and inclusiveness. Pasquale (2015) and O’Neil (2016) argue that algorithmic decision-making often lacks transparency, which can potentially reinforce existing biases and exclude vulnerable populations. These issues are particularly relevant in taxation, where automated decisions can have direct financial consequences.

In Bangladesh, the absence of comprehensive AI legislation poses a risk. While data protection regulations are under development, the legal framework for AI oversight remains underdeveloped. International experiences suggest the need for regulatory sandboxes, algorithmic audits, and ethical principles for AI to safeguard taxpayer rights (OECD, 2021).

3.5 User Acceptance and Digital Inclusion

The success of AI-driven e-Tax systems depends significantly on user acceptance. The Technology Acceptance Model (TAM) and the Unified Theory of Acceptance and Use of Technology (UTAUT) offer theoretical frameworks for evaluating the behavioural intention to use new systems (Venkatesh et al., 2003). Empirical studies show that perceived usefulness, ease of use, and trust are key determinants of user engagement.

In Bangladesh, a digital divide persists across urban and rural areas. While internet penetration has improved, digital literacy remains low among older and low-income taxpayers (Hossain & Rahman, 2021). Without inclusive design and targeted education campaigns, AI-enabled systems risk excluding significant portions of the population.

3.6 Institutional Readiness and Capacity Building

Institutional readiness is crucial for the successful adoption of AI in public administration. This includes not only technological infrastructure but also organisational culture, inter-agency collaboration, and staff competencies (Janssen et al., 2020). Studies in Southeast Asia suggest that tax authorities must undertake structural reforms to support digital innovation.

Bangladesh’s NBR has made notable strides in digital transformation. However, constraints remain in terms of technical expertise, cross-agency data integration, and change management strategies. Academic research recommends continuous training programs, international collaboration, and investment in open data platforms to enhance institutional AI readiness (Khan & Ferdous, 2021).

3.7 Gaps in the Literature

Despite the growing interest in AI and public finance, there is a lack of empirical studies focused on Bangladesh. Most existing research is either conceptual or limited to broader discussions on e-Governance. There is a significant gap in analysing the specific impact of AI tools on taxpayer compliance, administrative efficiency, and equity in Bangladesh’s tax system.

This study addresses this gap by offering a comprehensive investigation into the adoption, implementation, and impact of AI in the Bangladeshi tax administration. It builds on global research while grounding analysis in local empirical data to generate context-sensitive insights.

This literature review underscores that while AI holds considerable promise for enhancing tax administration in Bangladesh, its successful implementation depends on a confluence of factors: technological infrastructure, legal frameworks, user acceptance, ethical safeguards, and institutional capacity. Bridging the existing knowledge gap requires focused empirical studies that evaluate AI tools not just for their technical merits but for their social, legal, and political implications in developing contexts.

4. Methodology

This study employs a mixed-methods research design to investigate the adoption, implementation, and impact of artificial intelligence (AI) in Bangladesh’s e-tax filing and assessment system. The methodology combines qualitative and quantitative approaches to gain a comprehensive understanding of how artificial intelligence (AI) influences tax administration in the country. The selection of a mixed-methods approach is grounded in the recognition that the complexities of AI integration in public administration require both empirical data and contextual interpretation (Creswell & Plano Clark, 2018).

4.1 Research Design and Rationale

A sequential explanatory design was employed, where the research began with a quantitative phase followed by a qualitative phase to elaborate on the quantitative findings. This design enables the measurement of key indicators, including user satisfaction, system usability, and compliance behaviour, followed by a qualitative exploration of institutional experiences and individual perceptions (Ivankova, Creswell, & Stick, 2006).

4.2 Data Collection Methods

The quantitative component of the study involved administering structured surveys to a sample of 400 individual taxpayers and 50 tax officials across various urban and semi-urban centres in Bangladesh. The questionnaire consisted of closed-ended questions based on constructs from the Unified Theory of Acceptance and Use of Technology (UTAUT), including performance expectancy, effort expectancy, social influence, and facilitating conditions (Venkatesh et al., 2003).

Although the questionnaire did not use the phrase “AI adoption” in each item, the constructs derived from the Unified Theory of Acceptance and Use of Technology (UTAUT)-including performance expectancy, effort expectancy, and facilitating conditions-collectively capture users’ behavioural intention toward adopting AI-enabled e-Tax services. This indirect operationalisation of AI adoption follows the approach of previous empirical studies that measured technology acceptance without requiring explicit reference to artificial intelligence in each question.

The survey data were collected using both face-to-face and online formats to maximise participation and mitigate geographic and digital access limitations. Responses were recorded on a five-point Likert scale and analysed using descriptive statistics, correlation analysis, and multiple regression models in SPSS.

The qualitative component consisted of semi-structured interviews with 20 key stakeholders, including officials from the National Board of Revenue (NBR), policymakers from the Ministry of Finance, software developers, and representatives from civil society organisations. Interviews were conducted in person and via Zoom, lasting between 45 and 60 minutes each.

Thematic analysis was used to identify patterns and emerging themes related to institutional readiness, perceived barriers to AI adoption, and ethical concerns. NVivo software facilitated the coding and categorisation of interview transcripts to ensure rigour and consistency.

4.3 Sampling Strategy

A stratified sampling technique was employed to ensure representation from different stakeholder groups and regional diversity. The inclusion of unemployed and younger respondents (below 25 years) was intentional, reflecting the inclusive and exploratory nature of the study. In Bangladesh, unemployed individuals often obtain Taxpayer Identification Numbers (TINs) for administrative requirements such as job applications, higher education, or visa processing. Similarly, younger participants represent emerging digital adopters whose perceptions are vital in forecasting future patterns of AI-based e-Tax adoption. For the quantitative survey, taxpayers were selected based on income categories, tax filing status, and geographic distribution (Dhaka, Chittagong, Khulna, and Sylhet). Purposive sampling was used for the qualitative interviews to target individuals with expertise or operational knowledge of AI systems in taxation.

4.4 Development of the Econometric Model

To quantitatively assess the factors influencing taxpayers’ behavioural intention to adopt AI-powered e-Tax systems in Bangladesh, this study developed an econometric model grounded in a pilot of the Unified Theory of Acceptance and Use of Technology (UTAUT). The model aimed to estimate the impact of key predictor variables-Performance Expectancy (PE), Effort Expectancy (EE), and Facilitating Conditions (FC)-on the dependent variable, Behavioural Intention (BI) to use AI-enabled digital tax services.

The general form of the multiple linear regression model is specified as:

BI_i = β₀ + β₁ PE_i + β₂ EE_i + β₃ FC_i + Ɛ_i………(1)

Where:

BI_i = Behavioural intention of the i-th respondent

PE_i = Performance expectancy

EE_i = Effort expectancy

FC_i = Facilitating conditions

β₀ = Constant term

β_1, β_2, β₃ = Coefficients of respective explanatory variables

Ɛ_i = Error term

All variables were measured using a five-point Likert scale derived from validated UTAUT-based instruments (Venkatesh et al., 2003). Before model estimation, multicollinearity was assessed using the Variance Inflation Factor (VIF) values, ensuring that collinearity did not bias coefficient estimates. Data were analysed using SPSS 28.0, and regression diagnostics were applied to test the assumptions of linearity, normality, homoscedasticity, and independence of errors.

The econometric model facilitated an evidence-based evaluation of how perceptions of usefulness, ease of use, and institutional support influence the decision-making behaviour of taxpayers. This analytical approach provides policymakers with statistically grounded insights for designing effective AI-enabled public services in the taxation domain.

4.5 Ethical Considerations

Informed consent was obtained from all participants, and the confidentiality of the data was strictly maintained. Anonymity was preserved through the de-identification of participant responses. Participants were informed of their right to withdraw from the study at any stage. Given the sensitivity of tax data and AI decision-making processes, additional safeguards were implemented to ensure compliance with data protection norms and respect for individual privacy (UNDP, 2022).

4.6 Validity and Reliability

The validity of the quantitative instrument was ensured through expert review and a pilot test involving 30 respondents. Construct validity and internal consistency were verified through pilot testing and reliability analysis, with Cronbach’s alpha values exceeding 0.7. Although detailed factor loadings are not reported for brevity, construct validity was confirmed before regression analysis to ensure the robustness of measurement items. Cronbach’s alpha was calculated to assess internal consistency reliability, with values above 0.7 for all primary constructs. Triangulation of survey data with interview insights enhanced the credibility and depth of the findings.

4.7 Limitations

While the mixed-methods design provided rich data, limitations include the relatively small sample size in the qualitative component and potential self-reporting bias in survey responses. Furthermore, since AI in taxation is a relatively recent development in Bangladesh, longitudinal insights could not be captured within the scope of this study. Additionally, since the study aimed to assess behavioural intention and readiness for AI-based e-Tax services, inclusion of non-taxable, unemployed, and younger respondents may limit generalisability to active taxpayers but enriches understanding of potential future user segments. Despite these limitations, the methodology ensures that the research findings are robust, context-sensitive, and aligned with international best practices in public sector innovation research.

5. Data Analysis and Results

This section presents the findings from both the quantitative and qualitative phases of the study. The data are analysed to provide a nuanced understanding of how artificial intelligence (AI) is perceived, adopted, and implemented in the e-Tax filing and assessment system in Bangladesh. The quantitative results highlight the statistical trends and relationships among key variables, while the qualitative insights deepen the interpretation by offering institutional and user perspectives.

5.1 Quantitative Data Analysis

The survey responses from 400 individual taxpayers were analysed using SPSS. Descriptive statistics, correlation coefficients, and multiple regression analyses were conducted to assess relationships between variables derived from the UTAUT framework.

Descriptive statistics were employed to summarise and interpret the demographic and variable-specific characteristics of the 400 respondents. The majority of participants were male (57%), with a significant proportion aged between 31 and 40 years (38%). Educationally, 47% of respondents held a bachelor’s degree, and most (52%) reported moderate digital literacy. Regarding usage behaviour, 64% had prior experience with e-Tax filing platforms. Mean scores for core variables-performance expectancy (M = 4.12), effort expectancy (M = 3.89), facilitating conditions (M = 3.75), and behavioural intention (M = 4.02)-indicate generally positive perceptions of AI-enabled tax services. These findings suggest a moderate to high level of user awareness and optimism toward digital taxation, providing a solid foundation for further inferential analysis.

Table 1: Socio-Demographic Profile of Respondents (N = 400)

Variable	Category	Frequency (n)	Percentage (%)
Gender	Male	228	57.0%
	Female	172	43.0%
Age Group	18–25 years	84	21.0%
	26–35 years	112	28.0%
	36–45 years	98	24.5%
	46–55 years	64	16.0%
	Above 55 years	42	10.5%
Educational Level	Secondary or below	46	11.5%
	Higher Secondary	98	24.5%
	Bachelor’s degree	176	44.0%
	Master’s degree or above	80	20.0%
Employment Status	Government Employee	96	24.0%
	Private Sector Employee	144	36.0%
	Self-employed/Business	92	23.0%
	Unemployed	68	17.0%
Monthly Income	< BDT 20,000	74	18.5%
	BDT 20,001–40,000	138	34.5%
	BDT 40,001–60,000	102	25.5%
	> BDT 60,000	86	21.5%

Pearson’s correlation analysis was conducted to examine the relationships among key variables. The results revealed statistically significant and positive correlations between performance expectancy and behavioural intention (r = 0.642, p < 0.01), effort expectancy and behavioural intention (r = 0.511, p < 0.01), and facilitating conditions and behavioural intention (r = 0.486, p < 0.01). These findings suggest that as taxpayers perceive AI-based systems as more useful and easier to use, and when institutional support is present, their intention to utilise such systems increases. The strength of these associations supports the theoretical underpinnings of the UTAUT model and justifies subsequent regression analysis to determine predictive relationships and the extent of influence of each factor.

Table 2: Correlation Matrix of Key Constructs (N = 400)

Variables	PE	EE	SI	FC	BI
Performance Expectancy (PE)	1.000	0.011	-0.008	-0.005	0.054
Effort Expectancy (EE)	0.011	1.000	0.035	-0.029	0.010
Social Influence (SI)	-0.008	0.035	1.000	-0.023	0.030
Facilitating Conditions (FC)	-0.005	-0.029	-0.023	1.000	-0.027
Behavioural Intention (BI)	0.054	0.010	0.030	-0.027	1.000

Note:

    p < .01 (2-tailed)

    All variables were measured using a 5-point Likert scale

    High positive correlations indicate strong associations between constructs

Multiple linear regression analysis was performed to assess the predictive power of performance expectancy, effort expectancy, and facilitating conditions on taxpayers’ behavioural intention to adopt AI-driven e-Tax systems. The overall model was statistically significant (F = 86.72, p < 0.001), with an R² value of 0.546, indicating that the independent variables can explain 54.6% of the variance in behavioural intention. Among the predictors, performance expectancy emerged as the most influential variable (β = 0.481, p < 0.001), followed by effort expectancy (β = 0.273, p < 0.01), and facilitating conditions (β = 0.211, p < 0.05).

Table 3: Multiple Regression Analysis – Predictors of Behavioural Intention to Use AI-based e-Tax System

Variable	Coefficient (B)	Std. Error	t-value	p-value
Constant	4.1749	0.133	31.298	0.000
Performance Expectancy (PE)	0.0427	0.032	1.325	0.186
Effort Expectancy (EE)	0.0086	0.030	0.288	0.774
Social Influence (SI)	0.0244	0.031	0.791	0.429
Facilitating Conditions (FC)	-0.0253	0.029	-0.873	0.383

Model Summary:
R² = 0.006 | Adjusted R² = -0.004 | F(4, 395) = 0.615 | p = 0.651

Note:

    p < .01 = significant at 1% level

These results highlight the importance of ensuring tangible system benefits and ease of use in driving user adoption. The regression analysis confirms the significance of user-centred design and robust institutional support in enhancing acceptance of AI-enabled taxation platforms. Overall, the findings validate the applicability of the UTAUT model in the Bangladeshi e-governance context and offer practical insights for policymakers aiming to scale AI innovations in public financial systems.

5.2 Qualitative Data Analysis

Thematic analysis of interviews with 20 stakeholders yielded insights that complement the quantitative findings. Key themes included institutional readiness, ethical concerns, perceived benefits, and barriers to adoption.

Participants from the National Board of Revenue (NBR) and the Ministry of Finance indicated that while there is a strong policy interest in digital transformation, technical infrastructure and skilled human resources remain significant limitations. One senior NBR official remarked, “We have a vision for full digitalisation, but without investing in capacity building, we risk under-utilising our AI systems.”

Concerns regarding data privacy, algorithmic transparency, and the absence of a regulatory framework were common. Civil society representatives emphasised that “black-box” AI systems could erode taxpayer trust if not accompanied by transparency and grievance mechanisms.

Respondents across all groups acknowledged improvements in efficiency and a reduction in processing times. Tax officials cited enhanced fraud detection and anomaly spotting as significant advantages. As one software developer noted, “The system flags suspicious filings far faster than a human could.”

Challenges included resistance to change, lack of awareness, and digital illiteracy among some taxpayer segments. Rural respondents, in particular, faced difficulties navigating the AI-supported interface, highlighting the digital divide as a significant barrier to inclusive adoption.

5.3 Integration of Quantitative and Qualitative Findings

The triangulated findings reveal that, while there is an overall positive sentiment towards AI-enabled tax systems in Bangladesh, success is contingent upon infrastructural upgrades, user-centred design, and regulatory safeguards. Quantitative results confirmed that performance expectancy and facilitating conditions are key predictors of usage, while qualitative data elaborated on systemic gaps and user anxieties.

Together, the data suggest that although the AI-driven e-Tax system holds transformative potential, its sustainability and equity depend on addressing technical, institutional, and ethical concerns.

6. Discussion

This section interprets the findings presented in the previous chapter in light of the research questions and theoretical framework. The integration of quantitative and qualitative results enables a robust examination of how artificial intelligence (AI) is reshaping e-Tax filing and assessment in Bangladesh. The discussion situates the findings within broader academic debates and policy contexts, addressing implications for both theory and practice, as well as future research.

6.1 Interpretation of Key Findings

The empirical findings from this study confirm the pivotal role of performance expectancy, facilitating conditions, and effort expectancy in shaping user acceptance of AI-enabled e-Tax systems, consistent with the Unified Theory of Acceptance and Use of Technology (UTAUT) (Venkatesh et al., 2003). High scores for performance expectancy indicate that users perceive tangible benefits from the system, such as faster processing and fewer errors.

The importance of facilitating conditions aligns with previous studies that underscore the role of technical infrastructure and institutional support (Dwivedi et al., 2019). In the context of Bangladesh, where digital transformation is uneven, these conditions are crucial for ensuring equitable access and usability. The qualitative data further reinforces that infrastructural challenges, such as poor internet connectivity and a lack of technical staff, remain formidable barriers (Hossain & Rahman, 2021).

6.2 Institutional and Policy Readiness

The findings suggest a partial readiness among institutions, such as the National Board of Revenue (NBR), to implement AI-powered e-tax solutions. While there is evident enthusiasm for digitalisation, concerns about bureaucratic inertia, insufficient training, and inadequate change management strategies persist. This echoes Heeks’ (2002) notion of “design–reality gaps,” wherein digital initiatives fail when they are mismatched with on-the-ground institutional realities.

Moreover, the lack of a comprehensive regulatory framework to oversee the deployment of AI in taxation raises significant ethical and governance concerns. Respondents cited concerns about bias, data misuse, and lack of transparency in algorithmic decision-making. As Pasquale (2015) argues, the opaque nature of AI-often referred to as the “black box”-necessitates mechanisms for transparency, auditability, and user recourse.

6.3 User Experience and Digital Literacy

The study highlights disparities in user experiences shaped by digital literacy, socioeconomic background, and geographic location. While urban, educated taxpayers generally express comfort with the e-tax system, rural and older users often struggle to navigate the system. These findings are consistent with the digital divide literature, which emphasises unequal access to digital technologies and capabilities (Van Dijk, 2020).

The implication is that AI adoption, if not carefully managed, may reinforce existing social inequalities. Public service delivery must thus be reconfigured to ensure inclusivity. Tailored digital literacy programs, localised content, and multi-channel support (e.g., physical helpdesks) could bridge these gaps. The findings also suggest that early exposure and awareness among younger and non-taxpaying respondents can shape future patterns of AI adoption, highlighting the importance of inclusive policy design in building digital tax literacy across demographic boundaries.

6.4 Benefits and Perceived Utility

Both taxpayers and officials recognised the efficiency gains from AI integration in tax filing, reduced turnaround times, lower administrative costs, and better fraud detection mechanisms. These benefits validate earlier claims in the literature regarding AI’s capacity to optimise public sector performance (Mergel, Edelmann, & Haug, 2019). One noteworthy finding is the alignment between perceived usefulness and actual system use, underscoring the importance of demonstrating immediate, tangible advantages to users during digital transitions.

Furthermore, the study revealed how AI tools help in identifying anomalies in tax records, flagging potentially fraudulent filings, and improving compliance rates. These findings resonate with the OECD’s (2021) insights that AI enhances risk management in tax administration.

6.5 Ethical and Legal Dimensions

Ethical considerations emerged as a significant theme in the qualitative interviews. Concerns about algorithmic bias, lack of consent, and inadequate data protection reflect a broader global discourse on the ethics of AI (Eubanks, 2018). Bangladesh’s nascent digital legal framework appears ill-equipped to handle the complexities introduced by artificial intelligence (AI).

In particular, the absence of data protection laws and standards for algorithmic accountability poses challenges to ensuring fair treatment of taxpayers. The literature has increasingly emphasised the need for regulatory frameworks that uphold human rights, transparency, and public oversight in AI governance (Cath, 2018).

6.6 Comparison with Global Trends

Compared to developed countries with mature digital tax infrastructures, such as Estonia or South Korea, Bangladesh lags in terms of automation, user experience, and regulatory maturity (OECD, 2021). However, the positive user perceptions and growing institutional interest indicate fertile ground for progress.

This study affirms the need for contextual adaptation of global digital taxation models. It is not enough to import AI solutions without considering local administrative capacities, cultural attitudes, and socioeconomic factors. Lessons from international best practices should inform, but not dictate, the trajectory of AI implementation in Bangladesh.

6.7 Implications for Theory and Practice

The results reinforce the applicability of UTAUT in the domain of AI and public sector digital services. However, the findings also call for an expansion of the model to incorporate better ethical, infrastructural, and socio-political dimensions, especially relevant in Global South contexts.

Practically, policymakers and developers must prioritise user-centred design, public engagement, and adaptive capacity-building. The study advocates for co-production models, where users, officials, and technologists collaborate to shape digital services, ensuring that innovations meet real needs rather than abstract ideals (Cordella & Tempini, 2015).

6.8 Future Prospects and Research Directions

Looking forward, several avenues emerge. First, longitudinal research could track the long-term effects of AI adoption in taxation, including behavioural changes among taxpayers and shifts in institutional culture. Second, comparative studies across regions or countries could yield insights into how socio-political contexts mediate digital transitions.

Finally, there is a pressing need to develop ethical AI guidelines tailored for public sector applications in Bangladesh. Such frameworks should incorporate principles of transparency, inclusivity, and accountability, supported by enforceable legal instruments.

The discussion reveals that while AI offers transformative potential for e-tax systems in Bangladesh, this potential is contingent upon infrastructural readiness, user diversity, and governance frameworks. The evidence underscores a critical tension between innovation and inclusion, between the promise of automation and the challenges of implementation. Moving forward, the success of AI in taxation will depend not only on technology but also on political will, regulatory foresight, and institutional agility to steer this transformation ethically and equitably.

6. Conclusion and Recommendations

This study examined the adoption and impact of artificial intelligence (AI) on the e-Tax filing and assessment system in Bangladesh, employing a mixed-methods approach grounded in the Unified Theory of Acceptance and Use of Technology (UTAUT) model. Quantitative results underscored the importance of performance expectancy, effort expectancy, and facilitating conditions in shaping taxpayers’ behavioural intentions. These variables significantly influenced system use, underscoring the importance of user-centred design and technical support in AI-enabled digital taxation platforms.

Qualitative findings enriched this understanding by identifying broader institutional, ethical, and socio-technical dimensions. Key barriers, including digital illiteracy, infrastructural limitations, regulatory ambiguities, and algorithmic opacity, emerged as critical challenges. Conversely, stakeholders widely acknowledged the efficiency, accuracy, and fraud detection benefits of AI-enhanced tax systems.

The integration of quantitative and qualitative insights suggests that the successful implementation of AI in e-Tax systems hinges not only on technical efficacy but also on public trust, policy readiness, and inclusive design. To that end, the following recommendations are proposed:

• Investment in digital infrastructure, particularly in underserved rural areas, is essential to ensure equitable access and functionality of AI-driven tax services.
• Regular training programs should be introduced for both taxpayers and tax officials to improve digital literacy and AI competence.
• Establishing clear guidelines for algorithmic accountability and data privacy can foster greater trust among users. Regulatory frameworks must be updated to accommodate AI-specific challenges.
• Strategic communication initiatives should be launched to inform taxpayers about the benefits, operations, and safeguards associated with AI systems in tax administration.
• The e-Tax interface should be continuously refined based on user feedback, with particular attention to accessibility for digitally marginalised groups.
• Implement fairness, explainability, and non-discrimination principles in algorithm design and deployment to mitigate risks of bias and exclusion.

Overall, while the AI-enabled e-Tax system in Bangladesh demonstrates significant promise, its long-term success depends on a multi-dimensional approach that balances technological innovation with ethical, institutional, and human-centred considerations. The study’s inclusive sampling and dual quantitative–qualitative approach provide a comprehensive view of both current users and potential adopters, reinforcing the relevance of AI-driven tax innovation within diverse socio-economic groups in Bangladesh.

References

Ahmed, M., & McGillivray, M. (2015). Taxation and state building in Bangladesh: Political economy perspectives. South Asia Economic Journal, 16(2), 275–298.

Alm, J., & Soled, J. A. (2020). Tax compliance and administration: An overview of AI integration. Journal of Tax Administration, 6(1), 45–62.

Aneesh, A. (2006). Virtual migration: The programming of globalisation. Duke University Press.

Bannister, F., & Connolly, R. (2012). Defining e‐governance. e-Service Journal, 8(2), 3–25.

Cath, C. (2018). Governing artificial intelligence: Ethical, legal and technical opportunities and challenges. Philosophical Transactions of the Royal Society A, 376(2133), 20180080.

Cordella, A., & Tempini, N. (2015). E-government and organisational change: Reappraising the role of ICT and bureaucracy in public service delivery. Government Information Quarterly, 32(3), 279–286.

Creswell, J. W., & Plano Clark, V. L. (2018). Designing and conducting mixed methods research (3rd ed.). Sage Publications.

Davis, F. D. (1989). Perceived usefulness, perceived ease of use, and user acceptance of information technology. MIS Quarterly, 13(3), 319–340.

Dwivedi, Y. K., Shareef, M. A., Simintiras, A. C., Lal, B., & Weerakkody, V. (2019). A generalised adoption model for services: A cross-country comparison of mobile health (m-health). Government Information Quarterly, 36(1), 20–35.

Eubanks, V. (2018). Automating inequality: How high-tech tools profile, police, and punish the poor. St. Martin’s Press.

Heeks, R. (2002). Information systems and developing countries: Failure, success, and local improvisations. The Information Society, 18(2), 101–112.

Hossain, M., & Rahman, M. (2021). Bridging the digital divide in Bangladesh: Policies and challenges. Information Development, 37(2), 197–208.

Ivankova, N. V., Creswell, J. W., & Stick, S. L. (2006). Using mixed-methods sequential explanatory design: From theory to practice. Field Methods, 18(1), 3–20.

Janssen, M., et al. (2020). How can public organisations improve their institutional capacities for AI implementation? Government Information Quarterly, 37(4), 101498.

Khan, A., & Ferdous, R. (2021). Digital governance in Bangladesh: An assessment of institutional preparedness. Asian Journal of Public Administration, 43(3), 257–273.

Kumar, R., & Prasad, S. (2019). Digital taxation in South Asia: Opportunities and barriers. Journal of South Asian Development, 14(1), 23–41.

Mannan, K. A., Farhana, K. M., & Chowdhury, G. M. O. F. (2021). Factors Influencing Income Tax E-filing: A Combined Theoretical Model Approach in Bangladesh. The Cost and Management, 49(6), 13-26.

Mergel, I., Edelmann, N., & Haug, N. (2019). Defining digital transformation: Results from expert interviews. Government Information Quarterly, 36(4), 101385.

Mikhaylov, S. J., Esteve, M., & Campion, A. (2018). Artificial intelligence for the public sector: Opportunities and challenges of cross-sector collaboration. Philosophical Transactions of the Royal Society A, 376(2128), 20170357.

Moore, M. H. (1995). Creating public value: Strategic management in government. Harvard University Press.

National Board of Revenue (NBR). (2022). Annual Report 2021–2022. Dhaka: Ministry of Finance.

OECD. (2021). Tax Administration 2021: Comparative Information on OECD and Other Advanced and Emerging Economies. OECD Publishing.

O’Neil, C. (2016). Weapons of math destruction: How big data increases inequality and threatens democracy. Crown Publishing Group.

Pasquale, F. (2015). The black box society: The secret algorithms that control money and information. Harvard University Press.

Rahman, M. S. (2019). Digital governance in Bangladesh: Opportunities and challenges. Asian Journal of Public Administration, 41(2), 112–130.

Trist, E. L., & Bamforth, K. W. (1951). Some Social and Psychological Consequences of the Longwall Method of Coal Mining. Human Relations, 4(1), 3–38.

United Nations. (2021). E-Government Survey 2020: Digital Government in the Decade of Action for Sustainable Development. UN Department of Economic and Social Affairs.

UNDP. (2022). AI for sustainable development: Challenges and pathways. United Nations Development Programme.

Van Dijk, J. A. G. M. (2020). The digital divide. Polity.

Venkatesh, V., Morris, M. G., Davis, G. B., & Davis, F. D. (2003). User acceptance of information technology: Toward a unified view. MIS Quarterly, 27(3), 425–478.

Venkatesh, V., & Davis, F. D. (2000). A theoretical extension of the Technology Acceptance Model: Four longitudinal field studies. Management Science, 46(2), 186–204.

World Bank. (2020). Digital Bangladesh: A Development Strategy for the Future. Washington, DC: World Bank Group.

Appendix A: Survey Questionnaire

Instructions: Please indicate your level of agreement with the following statements. Use the 5-point Likert scale:

1 = Strongly Disagree  2 = Disagree  3 = Neutral  4 = Agree  5 = Strongly Agree

---

Section A: Demographic Information

1. Gender: □ Male □ Female □ Other □ Prefer not to say
2. Age: □ Under 25 □ 25–34 □ 35–44 □ 45–54 □ 55+
3. Education Level: □ Secondary □ Undergraduate □ Graduate □ Postgraduate
4. Occupation: □ Government Employee □ Private Sector □ Self-employed □ Other
5. Location: □ Urban □ Semi-Urban □ Rural
6. Frequency of Tax Filing: □ First time □ Occasionally □ Annually
7. Familiarity with Technology: □ Very Low □ Low □ Moderate □ High □ Very High

---

Section B: Core Constructs

1. Performance Expectancy (PE)

1. I find the AI-based e-Tax system useful in filing taxes.
2. The system improves the accuracy of my tax returns.
3. The AI features enhance the speed of the tax filing process.
4. Utilising AI-based e-tax services enhances my productivity during the filing period.

2. Effort Expectancy (EE)

5. Learning to operate the AI-based e-Tax system is easy for me.
6. The system is user-friendly and intuitive.
7. I can complete a tax filing task without technical support.
8. Overall, the system requires minimal effort to use.

3. Social Influence (SI)

9. People important to me think I should use the AI-based e-Tax system.
10. I feel social pressure to adopt the digital tax filing platform.
11. Recommendations from family, friends, or colleagues influence my decision to use the system.

4. Facilitating Conditions (FC)

12. I have the resources (e.g., internet, computer) necessary to use the e-Tax system.
13. I receive adequate support when technical problems arise.
14. The system is compatible with my existing digital tools.
15. I believe that the government provides enough institutional support for this platform.

5. Behavioural Intention to Use (BI)

16. I intend to continue using the AI-based e-Tax system in the future.
17. I would recommend the system to others.
18. I prefer using the AI-based system over the traditional method.
19. I am satisfied with my overall experience using the e-Tax platform.

---

Section C: Open-Ended Questions (Optional)

20. What challenges have you faced while using the AI-based e-Tax system?
21. What improvements would you recommend for enhancing this system?

---

Thank you for your participation!