Transparency by Design: A Narrative Synthesis of AI Disclosure, Explainability, and Trust in Consumer-Facing FinTech

Abstract

Artificial intelligence is increasingly embedded in consumer-facing FinTech, but trust in AI-enabled finance depends not only on performance, but also on whether users can understand and appropriately evaluate algorithmic outputs. This review synthesizes research on AI disclosure, explainability, and related transparency cues in consumer-facing FinTech, with particular attention to whether these cues support trust calibration rather than merely increasing trust or adoption. Searches in Scopus and Web of Science identified nine formally included studies and six adjacent contextual studies. The available evidence base is concentrated in robo-advisory and adjacent AI-enabled investment advising, with only limited evidence on automated credit decisions and crowdfunding recommendation platforms. The most studied cues are explanation/explainable AI and broader advisory or platform transparency, whereas disclosure, responsibility attribution, user control, and information-quality cues remain underexamined. Across the formal corpus, transparency cues are generally associated with more positive trust-related outcomes, especially trust and adoption-oriented responses. However, only a small subset of studies addresses trust calibration through outcomes such as reliance, fairness, accountability, and contestability. Overall, the current literature supports transparency more strongly as an acceptance mechanism than as a basis for appropriately bounded trust.

1. Introduction

Artificial intelligence is becoming increasingly embedded in consumer-facing FinTech, including robo-advisors, AI-enabled investment guidance, automated credit decisions, and recommendation systems that shape financial participation and allocation [1,2,3,4,5]. Yet the current empirical evidence on transparency and trust in these settings remains relatively small and is concentrated mainly in robo-advisory and adjacent advisory contexts rather than being broadly distributed across consumer-facing FinTech domains. In these settings, users are not simply interacting with digital interfaces; they are being asked to rely on systems that influence financially consequential judgments. Trust is therefore central [6,7,8]. In finance, the relevant question is not only whether AI improves efficiency or personalization, but whether users are given a sufficient basis to understand, evaluate, and appropriately rely on AI-mediated outputs. Throughout this review, trust calibration refers to cases in which transparency supports more appropriate reliance, contestability, and understanding of system limits, whereas mere acceptance refers to increased trust or willingness to use without demonstrated improvement in judgment.

Disclosure, explainability, transparency, and related cues are often proposed as mechanisms for building such trust [3,6,9,10,11]. These cues are expected to reduce opacity, improve understanding, and make automated systems appear more trustworthy. Yet transparency remains conceptually diffuse. It may refer to explicit AI disclosure, explanations of recommendations, or broader openness about processes, costs, criteria, or system functioning. These cues are frequently grouped together, even though they may activate different mechanisms and produce different trust-related consequences [5,7,12,13,14].

This ambiguity matters because greater trust is not necessarily better trust [1,6,15,16]. A transparency cue may increase comfort, perceived usefulness, or willingness to adopt without improving users’ ability to judge when reliance is warranted, when skepticism is appropriate, or when a decision should be challenged [1,17,18]. In financial settings, where under-trust may inhibit beneficial uptake but over-trust may encourage overreliance, the more demanding question is whether transparency supports trust calibration: trust that is appropriately aligned with the system’s logic, limits, and decision stakes.

The literature on transparency and trust in consumer-facing FinTech remains fragmented [19,20]. Relevant studies span robo-advisory, AI financial advice, automated credit decisions, crowdfunding recommendation platforms, and adjacent consumer-finance settings, but they differ in cue type, outcome choice, and methodological design [7,19,20,21]. The available evidence is dominated by robo-advisory and adjacent AI-based investment advising, with only isolated studies addressing automated credit decisions and crowdfunding recommendation platforms. They are also unevenly distributed across contexts, with much stronger concentration in robo-advisory than in credit, platform-based finance, chatbots, or mobile financial services [22,23,24,25]. At the same time, the topic falls between two adjacent but insufficiently integrated studies: technical work on explainability, fairness, and algorithmic accountability in finance, and broader consumer-oriented FinTech adoption research. The former often emphasizes model properties, regulatory safeguards, or system design without examining trust-related consumer outcomes, whereas the latter often examines trust and uptake without making transparency-related cues the focal mechanism [26,27,28,29]. Consequently, it is still uncertain which transparency cues have been examined, what their impact is on trust-related outcomes, and whether the evidence advocates for informed trust rather than simple acceptance. To address these questions, the review adopts a focused review approach using a targeted search of Scopus and Web of Science, explicit eligibility criteria centered on consumer-facing AI financial services, transparency-related constructs, and trust-related outcomes, and a structured narrative synthesis [30,31,32]. It distinguishes between formally included studies, which form the core synthesis, and adjacent contextual studies, which inform interpretation without being merged into the main evidence base [17,33,34,35,36].

The review makes four contributions. It provides a structured synthesis of a fragmented literature, differentiates among transparency cue types, maps the evidence across outcomes, contexts, and conditioning factors, and—most importantly—introduces trust calibration versus mere acceptance as the central interpretive lens. It also helps bridge the gap between technical explainability/fairness discussions and broader FinTech adoption research by bringing transparency-related cues and trust-related consumer outcomes into a common analytical frame. In doing so, it shifts attention from whether transparency produces more positive user responses to whether it helps consumers trust AI-enabled financial systems in a more informed, bounded, and contestable way.

2. Background Literature

2.1. Consumer-Facing FinTech and AI Systems

This review focuses on consumer-facing FinTech, or digital financial services where users directly interact with recommendations, rankings, or decisions mediated by AI or algorithms. The primary application domains in the formal corpus include robo-advisory, AI financial advice/advisory systems, automated credit decision-making, and crowdfunding recommendation platforms [21,33,34,37]. Robo-advisory is the most common context, and it usually means automated portfolio guidance or help with making investment decisions [17,19,37].

Similar to robo-advisory, AI financial advice can also refer to more recent systems, such as LLM-based personalized advisory systems [34]. In this review, LLM-based advisory is treated as a distinct AI financial advice/advisory system context rather than being subsumed under robo-advisory because it reflects a more generative and interaction-oriented advisory architecture than the conventional automated portfolio-guidance systems typically classified as robo-advisors. Automated credit interfaces vary in how AI influences access, approval, or denial in high-stakes scenarios, thereby enhancing fairness and contestability [33]. Crowdfunding recommendation systems broaden the discourse to platform-centric financial participation, wherein trust is influenced by both recommendation algorithms and the governance and engagement dynamics of the platform [21]. Chatbots, mobile banking, and digital wallets are conceptually pertinent yet predominantly lacking in the formal evidence base.

2.2. Transparency Cues: Disclosure, Explainability, Interpretability, and Control

The review classifies a number of design elements under the heading of transparency cues since they are all meant to lessen opacity in financial transactions mediated by AI. They are not, however, analytically identical. According to Mendel et al. [35], AI disclosure makes the presence of AI apparent without necessarily elucidating reasoning. This indicates that AI is involved in generating or supporting a recommendation or decision. Explainable AI offers criteria, justifications, or explanations for how an output was produced [33,34]. Interpretability or comprehensibility is about whether users can really understand that logic. It is related to explanation, but it is not the same thing [38]. Advisory or platform transparency encompasses openness regarding processes, costs, service logic, or platform operations [19,21,37].

User control concerns the extent to which users can inspect, influence, or override automated decisions [37]. Information quality/clarity of criteria measures how efficiently decision-relevant information is presented [17]. Responsibility attribution is about who is seen as responsible when AI is involved. This is especially important in hybrid advisory settings [35]. These cues can therefore be grouped as transparency-related, while still differing in mechanism and likely consequence.

2.3. Trust-Related Outcomes

The review treats several outcomes as trust-related, but not interchangeable. The most direct is trust, or positive confidence in the system or recommendation [19,34]. Credibility concerns whether an output appears believable or convincing, while perceived risk captures uncertainty or anticipated harm. Fairness perception is especially important in credit settings, where trust may depend on whether decisions appear procedurally justified [33]. Reliance is more behaviorally diagnostic because it reflects whether users actually depend on AI outputs rather than merely evaluate them favorably [35]. Adoption intention and continuance intention indicate willingness to begin or continue use, but these are closer to uptake than to calibrated judgment [17,37]. Psychological comfort and engagement are also relevant where studies link transparency to reassurance or platform participation, but they remain downstream or adjacent outcomes rather than direct substitutes for trust or reliance [21]. The key point is that these outcomes are related but conceptually distinct.

2.4. Trust Calibration Versus Mere Acceptance

The review’s central conceptual distinction is between trust calibration and mere acceptance. Trust calibration refers to cases in which transparency helps users judge AI more appropriately by improving understanding, clarifying limits, supporting selective reliance, or enabling challenges when needed [33,35]. It concerns trust that is better aligned with the system’s logic, uncertainty, and stakes. Mere acceptance, by contrast, refers to cases in which transparency mainly increases comfort, trust, or willingness to use without evidence of better judgment [17,37]. This distinction matters because in consumer-facing finance, trust is not automatically beneficial: under-trust may inhibit useful uptake, but over-trust may encourage uncritical reliance. For that reason, outcomes such as reliance, fairness, contestability, and demonstrated understanding are more diagnostic of calibration than trust ratings or adoption intention alone. Practically, this means that current transparency designs and the evidence supporting them appear to function more as mechanisms of adoption and reassurance than as robust supports for contestability, accountability, and appropriately bounded reliance. This distinction provides the conceptual basis for the review’s broader argument that transparency in consumer-facing FinTech has so far been validated more strongly as an acceptance mechanism than as a calibration mechanism. Against this conceptual background, the review addresses five research questions:

RQ1. 

What types of AI disclosure and explainability cues have been examined in consumer-facing FinTech contexts?

RQ2. 

How do AI disclosure and explainability cues influence trust-related consumer outcomes in consumer-facing FinTech?

RQ3. 

In which consumer-facing FinTech application contexts have these relationships been studied?

RQ4. 

What methodological, user-related, and contextual factors condition the relationship between transparency cues and trust-related outcomes in consumer-facing FinTech?

RQ5. 

To what extent do AI disclosure and explainability cues support trust calibration, rather than merely increasing acceptance, in consumer-facing FinTech?

3. Research Method

3.1. Search Strategy

This review employed a targeted search to identify studies at the intersection of consumer-oriented FinTech, AI-driven or algorithmically mediated financial services, transparency-related design features, and trust-related consumer outcomes. The search strategy was developed iteratively to balance breadth and precision. Initial pilot searches were either too broad, capturing general AI-and-finance work, or too narrow, risking omission of relevant studies. The final query was selected because it best represented the focal literature while keeping the screening set manageable. Accordingly, the review was designed as a focused review with structured search procedures intended to map a fragmented literature and clarify conceptual patterns, rather than to support meta-analysis or strong causal inference.

Title-, abstract-, and keyword-level searches were performed in Scopus and Web of Science. The searches were last run on 24 December 2025. The final search string was:

(“robo-advisor” OR “robo advisory” OR chatbot OR algorithmic OR automated OR “decision-support system*” OR “automated financial advisory”) AND (transparency OR “algorithmic transparency” OR explainab* OR disclosure OR interpretab* OR comprehensib* OR “user control” OR “AI disclosure”) AND (trust OR “perceived risk” OR fairness OR credibility OR reliance OR “adoption intention”) AND (“digital wallet*” OR “mobile banking” OR “investment platform*” OR “personal finance” OR fintech OR “credit scoring” OR lending).

The search was intentionally broad in subject coverage because the topic spans FinTech, information systems, consumer behavior, human–computer interaction, digital trust, and AI governance. Restricting subject categories at the database level risked excluding relevant work indexed under business, computer science, social sciences, or interdisciplinary journals. The review was not prospectively registered.

The Scopus search identified 97 records. After screening for document type, language, and accessibility, 30 records remained; specifically, 62 records were removed on document-type grounds, 1 non-English record was removed, and 4 inaccessible records were excluded. The Web of Science search identified 35 records, of which 31 remained after filtering; 3 records were removed on document-type grounds, and 1 non-English record was excluded. The two filtered database sets therefore yielded 61 records before deduplication. After the removal of 20 duplicates, 41 unique records remained for screening.

3.2. Eligibility Criteria

Studies qualified if they satisfied three criteria: they investigated an AI-enabled or algorithmically mediated consumer-facing financial service; they incorporated a construct, feature, manipulation, or evaluative dimension related to disclosure, transparency, or explainability; and they reported a minimum of one trust-related consumer outcome, including trust, perceived risk, fairness, credibility, reliance, or adoption intention. Screening was conducted by one reviewer. No automated tools were utilized during the screening process.

These criteria were applied with deliberately tight scope boundaries (

Table 1. Inclusion/exclusion criteria.

Decision Rule	Include When	Exclude When
Service context	The study examines a consumer-facing financial service, such as robo-advisors, AI-enabled financial advice, FinTech chatbots, digital wallets, mobile banking, crowdfunding platforms, or lending/credit interfaces.	The study focuses on back-end financial systems, internal banking operations, institutional trading, compliance, fraud detection, or non-financial AI applications.
AI/algorithmic component	AI or algorithmic mediation is central to the user-facing decision or interface.	AI is only background infrastructure or is not central to the user-facing service.
Transparency-related component	The study includes a focal disclosure-, transparency-, or explainability-related construct, feature, manipulation, or evaluative dimension, such as explanation, AI disclosure, advisory transparency, interpretability, comprehensibility, information quality, or user control.	The study examines general FinTech trust, TAM, UTAUT, or adoption without a transparency/disclosure/explainability component.
Outcome requirement	The study reports at least one trust-related consumer outcome, such as trust, perceived risk, fairness, credibility, reliance, or adoption intention.	The study is technical XAI, modeling, or fairness work without end-user perceptions, responses, or trust-related outcomes.
Robo-advisor rule	Robo-advisor studies are included even if explainability is not the exact term used, provided that a genuine transparency-related feature is present.	Robo-advisor studies are excluded if they focus only on trust/adoption and do not operationalize a transparency/disclosure/explainability element.
Chatbot rule	FinTech chatbot studies are included when they examine trust together with transparency, explanation, or user control.	Non-financial chatbot studies, or chatbot studies without a trust–transparency link, are excluded.
Lending/credit rule	AI lending or credit studies are included when they examine the consumer-facing side and include fairness, transparency, explainability, or trust perceptions.	Technical credit scoring, risk modeling, or internal lending systems without consumer-facing perceptions are excluded.
Context boundary	Financial e-commerce or crowdfunding studies are included when they clearly function as consumer-facing FinTech contexts and include transparency/trust mechanisms.	General e-commerce recommender studies without a clear financial-service context are excluded.

). Robo-advisor studies were incorporated even if they did not utilize the precise term “explainability,” as long as a legitimate transparency-related component was evident, including explanation, advisory transparency, AI disclosure, comprehensibility, interpretability, information clarity, or user control. Studies on FinTech chatbots could only be included if they connected trust-related results to openness, explanation, or control. AI lending or credit-interface studies were only eligible if they examined the consumer-facing side of the service and included fairness, openness, explainability, or trust perceptions. One reviewer did the data extraction. No study authors were contacted for more information.

Studies were omitted if they concentrated exclusively on back-end financial systems, non-financial AI applications, technical explainable AI devoid of user outcomes, or general FinTech adoption lacking a transparency-related aspect. The review excluded research on fraud detection, risk modeling, internal compliance, institutional trading, and other internal or non-consumer applications, as well as studies on general trust, the Technology Acceptance Model (TAM), or the Unified Theory of Acceptance and Use of Technology (UTAUT) in FinTech that failed to emphasize transparency, disclosure, or explainability. One reviewer did the quality appraisal.

3.3. Screening and Study Selection

To maintain transparency in study identification and selection, the review process is summarized in Figure 1 using a PRISMA-style flow structure [39]. After database-level filtering, 30 Scopus records and 31 Web of Science records remained. These 61 records were then merged, and 20 duplicates were removed, resulting in 41 unique records for screening.

Title and abstract screening were guided by three questions: whether the study examined a consumer-facing financial service, whether it included a transparency/disclosure/explainability-related component, and whether it reported a trust-related consumer outcome. To avoid premature exclusion, borderline records were retained for full-text assessment.

Full-text screening determined the formal synthesis sample. Of the 41 records assessed in full, 26 were excluded because they did not meet the review’s eligibility criteria. The formal synthesis therefore included 9 studies that satisfied all eligibility requirements. In addition, 6 adjacent contextual studies were retained for interpretive use in the discussion because they were highly relevant to trust and adoption in consumer-facing AI-enabled finance, but did not fully meet the stricter transparency/disclosure/explainability criterion. These adjacent studies were not incorporated into the formal synthesis.

3.4. Quality/Risk-of-Bias Appraisal Approach

Quality appraisal used a domain-based qualitative approach rather than a single numerical checklist because the final corpus was methodologically heterogeneous. The objective was to evaluate the evidence’s strength, applicability, and interpretive value rather than to exclude studies based on their design.

Conceptual fit evaluated whether the study actually addressed a consumer-facing AI-financial service and whether the explainability, transparency, or disclosure component was essential rather than incidental. Construct clarity was the degree to which variables pertaining to trust and transparency were clearly defined.

Design adequacy examined whether the selected approach aligned with the research goal. Sample/context transparency focused on how well the FinTech environment and user group were described. Interpretive credibility evaluated the degree to which conclusions were consistent with the available data. Given the final sample’s combination of surveys, qualitative research, design-oriented work, mixed-methods studies, legal-empirical studies, and crowdfunding/platform research, this strategy was appropriate.

3.5. Narrative Synthesis Approach

A narrative synthesis was adopted because the included studies were conceptually related but too heterogeneous for statistical aggregation. The synthesis mapped studies across three core dimensions: transparency cue, trust-related outcome, and FinTech application context. Transparency cues included AI disclosure, explanation, advisory transparency, interpretability, comprehensibility, and user control. Trust-related outcomes included trust, reliance, credibility, fairness, perceived risk, and adoption-related outcomes. FinTech contexts included robo-advisory, AI financial advice, automated credit decisions, and crowdfunding recommendation platforms.

The synthesis also considered methodological and contextual conditioning factors, including user expertise, AI literacy, market setting, and hybrid human–AI configurations. Beyond descriptive mapping, it examined whether transparency cues appeared to support trust calibration—that is, more appropriately aligned reliance based on meaningful understanding—rather than merely increasing trust, comfort, or adoption. The six adjacent contextual studies were not included in the formal synthesis. Instead, they were used selectively in the discussion to situate the core findings within the broader trust and adoption literature on consumer-facing AI in finance.

4. Results

4.1. Overview of Included Studies

The formal synthesis consisted of nine included studies, alongside six adjacent contextual studies retained for interpretive discussion (

Table 2. Characteristics of formally included studies.

Authors	Country	FinTech Context	Method	Sample	Core Transparency Cue	Core Trust-Related Outcome
To et al. [34]	Vietnam	LLM-based personalized investment advisory system	Design science research combining multi-agent system development, historical back-testing, expert evaluation, and investor surveys	50 listed firms; 50 retail investors	EXPL	TR
Mendel et al. [35]	United States	AI-assisted personal finance/retirement investment advising	Between-subjects online advising simulation with AI disclosed vs. undisclosed to clients	113	DISC	RL
Kwon et al. [37]	South Korea	Robo-advisory/AI-based investment advisory service	Online survey with structural equation modeling using a combined TAM and innovation resistance model	158	TRSP	AI
Nazmi et al. [17]	Malaysia	Financial robo-advisors for low-income households	Self-administered bilingual questionnaire analyzed with SEM/PLS-SEM using UTAUT with advisory transparency as a mediator	217	TRSP	AI
Kulkarni et al. [38]	India	Robo-advisory/AI-powered financial robo-advisor for retail investment decision-making	Cross-sectional survey analyzed with PLS-SEM (SmartPLS 4.0)	461	EXPL	AI
Jung et al. [19]	Germany	Robo-advisory/automated web-based investment advisory	Design science study with iterative prototype development and two controlled mixed-method laboratory evaluations	30	TRSP	TR
Lui et al. [33]	United Kingdom	Automated credit decision-making/credit scoring	Doctrinal and comparative legal analysis with UK public and bank-employee surveys	99	EXPL	TR
Nain et al. [40]	India	Robo-advisory/automated wealth management	Semi-structured interviews with industry experts analyzed by content analysis	12	TRSP	TR
Wang et al. [21]	China	E-commerce crowdfunding/crowdfunding recommendation platform	Mixed-method empirical framework combining a structured survey, OLS/SEM/2SLS mediation analysis, XGBoost, and investment-strategy simulation	300 survey respondents + 5000 crowdfunding projects	EXPL	TR

Note: The table includes only the formally included studies; adjacent contextual studies are not shown here. FinTech context, transparency cue, and trust-related outcome are presented using the review coding scheme. FinTech context codes: RA = robo-advisory; AFA = AI financial advice/advisory system; ACI = automated credit/lending interface; CRP = crowdfunding/recommendation platform; CHAT = FinTech chatbot; MB = mobile banking/digital wallet. Transparency cue codes: DISC = AI disclosure; EXPL = explanation/explainable AI; TRSP = advisory or platform transparency; INTP = interpretability/comprehensibility; CTRL = user control/override/customization; INFOQ = information quality/clarity of criteria; RESP = responsibility attribution/disclosure of AI involvement. Trust-related outcome codes: TR = trust; CR = credibility; PR = perceived risk; FA = fairness perception; RL = reliance; AI = adoption intention; CI = continuance intention; PC = psychological comfort; ENG = engagement. Sample entries are reported as extracted and may refer to participants, firms, projects, or mixed sources depending on study design.

). The related studies were preserved to facilitate the interpretation of mechanisms and boundary conditions; solely the formal set was incorporated into the core synthesis. Table 1 shows that the formal literature came from many different places, but it was sparsely distributed [17,19,21,34,35,37,38,40]. It included Vietnam, the United States, South Korea, Malaysia, India, Germany, the United Kingdom, and China, with India being represented twice. As a result, research on transparency and trust in consumer-facing FinTech is developing in both established and emerging digital finance markets; however, it remains disjointed rather than cumulative within any specific context.

The evidence base was also highly uneven in terms of application context. Six of the nine studies included in the review examined robo-advisory and other AI-based investment advice [17,19,35,37,38,40]. The other three studies focused on LLM-based investment advice, automated credit decision-making, and crowdfunding recommendation platforms [21,33,34]. No formally included study concentrated on FinTech chatbots, mobile banking, or digital wallets, suggesting that the existing literature is predominantly focused on advisory decision support rather than consumer-oriented FinTech in a broader context.

Methodologically, the corpus was diverse, encompassing surveys, laboratory or simulation studies, design science, mixed methods, qualitative interviews, and legal-empirical analysis [17,19,21,33,34,35,37,38,40]. Survey designs were predominant, particularly in robo-advisory adoption research [17,37,38], supplemented by smaller design-focused [19,34] and qualitative studies [40], along with a singular legal-empirical study on automated credit decisions. The results showed a similar pattern: trust and intention to adopt were the most common, while reliance and fairness perception were less common and only in specific situations. In general, the formal evidence base is best described as contextually narrow, methodologically mixed, and biased toward building trust and supporting adoption. These things are important for understanding the review as a whole: the current work provides useful early evidence on transparency-related trust effects in consumer-facing FinTech, but remains concentrated in a limited set of advisory settings and only partially addresses the broader question of trust calibration [33,35].

4.2. RQ1: Types of AI Disclosure and Explainability Cues Examined

The formal studies investigated a constrained yet analytically distinct set of transparency-related indicators. This review organized distinct types of terminology into seven families to standardize heterogeneous terminology: AI disclosure (DISC), explanation/explainable AI (EXPL), advisory or platform transparency (TRSP), interpretability/comprehensibility (INTP), user control/customization (CTRL), information quality/clarity of criteria (INFOQ), and responsibility attribution (RESP).

Table 3. Types of transparency cues.

Cue Code	Cue Family	Definition Used in This Review	Example Operationalization	Studies	FinTech Contexts
DISC	AI disclosure	Explicit disclosure that AI is involved in the advisory or decision process	Whether clients know that advisors used AI and whether the AI recommendation is visible	Mendel et al. [35]	RA
EXPL	Explanation/explainable AI	Cues clarifying how an AI-enabled system reaches or supports a recommendation or decision	Natural-language explanations; right to explanation for automated credit decisions; perceived algorithm interpretability and structural assurance	To et al. [34]; Lui et al. [33]; Kulkarni et al. [38]; Wang et al. [21] *	AFA; ACI; RA; CRP
TRSP	Advisory or platform transparency	Openness about service logic, process, costs, information basis, or platform functioning	Transparency of algorithms used in robo-advisor decisions; costs, information, and process transparency; interface features clarifying business models, processes, and assets	To et al. [34]; Lui et al. [33]; Kwon et al. [37]; Nazmi et al. [17]; Nain et al. [40]; Jung et al. [19]	AFA; ACI; RA
INTP	Interpretability/comprehensibility	User-perceived understandability of algorithmic reasoning or recommendation logic	Algorithm interpretability as part of explainable robo-advisor design	Kulkarni et al. [38]	RA
CTRL	User control/override/customization	Degree to which users can influence, inspect, or retain agency over automated decisions	User control over automated investment decisions alongside transparency of algorithmic information	Kwon et al. [37]	RA
INFOQ	Information quality/clarity of criteria	Clarity and usability of decision-relevant information	Clearer cost, information, and process features for low-income users	Nazmi et al. [17]	RA
RESP	Responsibility attribution/AI involvement	Cues about who is responsible when AI is involved in a recommendation	AI-use disclosure shifting responsibility perceptions between the advisor and the AI-supported process	Mendel et al. [35]	RA

Note: The table summarizes only the formally included studies. Cue categories are coded using the review taxonomy: DISC = AI disclosure; EXPL = explanation/explainable AI; TRSP = advisory or platform transparency; INTP = interpretability/comprehensibility; CTRL = user control/override/customization; INFOQ = information quality/clarity of criteria; RESP = responsibility attribution/disclosure of AI involvement. Studies could contribute to more than one cue family when a secondary cue was identifiable. FinTech context codes: RA = robo-advisory; AFA = AI financial advice/advisory system; ACI = automated credit/lending interface; CRP = crowdfunding/recommendation platform. (*) For Wang et al. [21], the originally extracted cue combined recommendation explainability/transparency with platform-governance elements; for table standardization, it is represented under EXPL as the closest primary codebook category.

shows that EXPL and TRSP are the most common types in the literature, while disclosure, control, information quality, and responsibility cues are only found occasionally. In general, transparency has been studied mainly as a broad property of explainability or service openness rather than as a differentiated design architecture with distinct mechanisms.

Only Mendel et al. [35] introduced AI disclosure (DISC), where transparency was defined as whether clients would be aware that their advisor had used AI and whether the AI recommendation would be visible. Because it focuses on the visibility of AI involvement rather than the logic of the output itself, this type of transparency is limited but crucial. Its rarity indicates that the question of whether merely disclosing AI participation modifies trust-related dynamics in consumer finance has received little attention in the literature. Explanation or explainable AI (EXPL) had the broadest contextual spread, such as AI financial advice (AFA), automated credit decisions (ACI), robo-advisory (RA), and crowdfunding recommendation platforms (CRP).

However, implementation differed significantly: To et al. [34] investigated natural-language explanations at the agent and aggregator levels within an LLM-based advisory system; Lui et al. [33] concentrated on the right to explanation for automated credit decisions, particularly the disclosure of criteria and weightings; Kulkarni et al. [38] highlighted perceived algorithm interpretability and structural assurance; and Wang et al. [21] associated recommendation transparency with trust and subsequent platform outcomes. So, explanation is the most advanced cue family, but it is also the most variable in terms of concepts.

Advisory or platform transparency (TRSP) was particularly prevalent in robo-advisory, where it was typically regarded as a service-level attribute rather than a strictly confined interface modification. Kwon et al. [37] characterized it as the extent to which users could verify and comprehend the information and algorithms employed in robo-advisor decisions. Nazmi et al. [17] implemented it via cost, information, and process transparency for economically disadvantaged users. Nain et al. [40] regarded operational transparency as a fundamental aspect of robo-advisory, whereas Jung et al. [19] integrated it into interface elements like a cost calculator, more detailed process descriptions, and enhanced asset and portfolio information. Compared with EXPL, TRSP is less about why a specific decision was made and more about making the overall service environment legible and inspectable.

The cue families that remained were considerably thinner. Only Kulkarni et al. [38] made explicit reference to interpretability/comprehensibility (INTP), emphasizing the distinction between interpretability and explanation: interpretability is concerned with whether users can make sense of it, whereas explanation is concerned with providing justificatory content. Only Kwon et al. [37] discussed user control (CTRL) as the level of agency users retained over automated investment decisions. The most obvious example of information quality/clarity of criteria (INFOQ) was found in Nazmi et al. [17], where advisory transparency for B40 users included clearer cost, information, and process features. Only in Mendel et al. [35] did responsibility attribution (RESP) emerge, where people’s perceptions of who was in charge of the recommendation changed when AI use was disclosed.

In general, the field is most advanced in terms of explanation and greater advisory transparency, particularly in robo-advisory settings [17,19,37,40]. In contrast, direct AI disclosure, responsibility attribution, user control, and information quality are still not well studied and are often seen as secondary rather than central concepts [17,35,37]. The existing literature acknowledges various transparency-related mechanisms; however, it has yet to analyze them in a balanced or systematically comparative manner within consumer-facing FinTech contexts.

4.3. RQ2: How Transparency Cues Influence Trust-Related Outcomes

Across the formal studies, transparency cues generally operate in a trust-supportive direction, but the form of that support varies by outcome and context.

Table 4. Effects of transparency cues on trust-related outcomes.

Study	Transparency Cue	Trust-Related Outcome(s)	Direction of Effect	Summary of Finding	Interpretive Note
To et al. [34]	EXPL	TR; AI	Positive	Explainable and transparent personalized recommendations were associated with higher investor trust and willingness to use, alongside strong system performance.	Supportive evidence, but not based on a direct causal transparency manipulation; more trust- and use-oriented than calibration-focused.
Mendel et al. [35]	DISC	RL; TR	Mixed/conditional	Disclosing advisor AI use did not directly increase AI reliance, but indirectly increased reliance by reducing perceived personal responsibility.	A clear case in which disclosure reshapes responsibility and reliance rather than simply elevating trust.
Kwon et al. [37]	TRSP	AI	Indirect only	Transparency increased perceived usefulness and indirectly supported adoption intention while reducing innovation resistance.	Adoption-oriented and mediated, not a direct transparency-to-trust effect.
Nazmi et al. [17]	TRSP	AI	Positive	Advisory transparency had a significant positive direct effect on robo-advisor adoption intention and mediated key UTAUT-linked predictors.	Transparency functions mainly as an adoption-enabling mechanism, especially in a low-income user setting.
Kulkarni et al. [38]	EXPL	AI	Positive	Perceived algorithm interpretability, together with interactivity, personalization, and autonomy, was associated with reduced behavioral biases relevant to investment decision-making; structural assurance was not significant.	Transparency-related features appear beneficial, but this is not a clean test of disclosure/explanation effects on trust itself.
Jung et al. [19]	TRSP	TR; AI	Positive	Improved cost, process, and information transparency increased trust, satisfaction, and willingness to invest, although many users still preferred some human contact.	Strong evidence for trust-building, but not necessarily for calibrated challenge or appropriate skepticism.
Lui et al. [33]	EXPL	TR; FA	Unclear/indirect	Opacity was framed as undermining trust and fairness in automated credit decisions, with strong support for a right to explanation, especially regarding criteria and weightings.	Explanation is treated as a safeguard for understanding and contestability, but no experimental effect size was estimated.
Nain et al. [40]	TRSP	TR; AI	Positive	Interview evidence identified transparency in operations as a central pillar for building trust and encouraging adoption in robo-advisory.	Qualitative and provider-oriented evidence; useful for interpretation but not strong causal inference.
Wang et al. [21]	EXPL	TR; AI	Positive	Recommendation transparency and perceived platform control increased trust, while trust mediated the path from transparency to crowdfunding success; engagement and update frequency were also important.	Extends the review beyond advisory settings and links transparency to trust-mediated platform outcomes.

Note: The table includes only the formally included studies. Transparency cues are coded as DISC = AI disclosure, EXPL = explanation/explainable AI, and TRSP = advisory or platform transparency. Trust-related outcomes are coded as TR = trust, FA = fairness perception, RL = reliance, and AI = adoption intention. Direction codes were interpreted as follows: Positive = positive association or effect; Indirect only = effect operates through mediators rather than directly; Mixed/conditional = effects vary by pathway or condition; Unclear/indirect = interpretive or normative support without a direct behavioral effect estimate. Where original extracted entries used broader or mixed constructs, the closest codebook-consistent trust outcome and transparency cue were retained for manuscript standardization.

illustrates that positive effects are prevalent; however, they frequently manifest as indirect, conditional, or focused on reassurance and adoption rather than explicit behavioral calibration [21,33,34,35].

Positive responses predominate when it comes to trust-related responses. Higher investor trust and willingness to use AI financial advice were linked to personalized recommendations that were clear and understandable [34]. While operational transparency was found to be a key component of trust and adoption [40], improved cost, process, and information transparency in robo-advisory enhanced trust, satisfaction, and willingness to invest [19]. Transparency in recommendations boosted trust in crowdfunding, which in turn mediated wider platform outcomes [21].

The evidence is thinner and more context-specific for outcomes related to fairness and perceived risk. Fairness was most evident in automated credit decisions, where justification was deemed essential for procedural legitimacy within opaque decision-making frameworks. Lui et al. [33] contended that opacity erodes trust and fairness, advocating robust support for a right to explanation, particularly concerning the criteria and weightings employed in unsuccessful decisions. Conversely, perceived risk was infrequently analyzed directly within the formal corpus, indicating that the literature predominantly addresses the effects of transparency on eliciting favorable responses rather than its efficacy in enhancing users’ assessment of financial risk. In a similar vein, Kulkarni et al. [38] discovered that features associated with interpretability diminished overconfidence and loss-aversion biases, while structural assurance did not exhibit such effects.

The evidence for reliance is limited, yet particularly enlightening. Mendel et al. [35] demonstrated that revealing AI usage did not directly enhance dependence on AI recommendations; however, it did indirectly elevate reliance by diminishing perceived personal responsibility. This is theoretically significant as it demonstrates that transparency can modify reliance by transforming accountability, rather than merely enhancing the perceived trustworthiness of AI.

Adoption-oriented outcomes comprise the largest cluster of findings. Transparency frequently serves as a facilitator of uptake in robo-advisory. Despite having no direct impact on perceived complexity or safety, Kwon et al. [37] discovered that transparency enhanced perceived usefulness, indirectly supported acceptance intention, and decreased innovation resistance. Similarly, Nazmi et al. [17] demonstrated that advisory transparency mediated important UTAUT-linked predictors and directly positively impacted robo-advisor adoption intention, particularly in a low-income and low-literacy group. Similar relationships were found between transparency and intended use or willingness to invest in other studies [19,34,40].

Overall, three points stand out. First, favorable outcomes predominate, particularly for adoption-oriented and trust-building outcomes [17,19,21,34]. Second, although less frequent, mixed or conditional effects are theoretically significant. For example, transparency may influence some evaluations but not others [37] or fail to demonstrate a direct effect while still influencing outcomes through intermediate mechanisms [35]. Third, rather than having a straightforward direct impact, transparency frequently works through perceived usefulness, advisory transparency, responsibility perceptions, or trust itself [17,21,35,37].

Context is crucial. Transparency in robo-advisory is typically associated with adoption support and trust-building through explainability, process clarity, or service openness [17,19,37,38,40]. Fairness, legitimacy, and contestability are more closely associated with explanation in automated credit decisions [33]. Transparency functions alongside platform control and engagement dynamics in crowdfunding recommendation platforms [21]. Overall, transparency cues in consumer-facing FinTech usually work in a positive direction, but most often by supporting trust, reassurance, and uptake rather than by directly demonstrating more accurate, selectively bounded, or challenge-ready trust.

4.4. RQ3: FinTech Application Contexts in Which These Relationships Have Been Studied

The formal evidence base varies greatly among FinTech contexts that interact with consumers.

Table 5. Consumer-facing FinTech application contexts.

Context Code	FinTech Application Context	Studies	Typical Transparency Cues	Typical Trust-Related Outcomes	Evidence Density
RA	Robo-advisory/automated investment advising	[17,19,35,37,38,40]	DISC, EXPL, TRSP	TR, RL, AI	High
AFA	AI financial advice/advisory system	[34]	EXPL	TR, AI	Low
ACI	Automated credit decision-making/credit scoring	[33]	EXPL	TR, FA	Low
CRP	Crowdfunding/recommendation platform	[21]	EXPL	TR, AI	Low

Note: The table includes only the formally included studies. FinTech context codes are as follows: RA = robo-advisory; AFA = AI financial advice/advisory system; ACI = automated credit/lending interface; CRP = crowdfunding/recommendation platform; CHAT = FinTech chatbot; MB = mobile banking/digital wallet. Transparency cue codes are DISC = AI disclosure; EXPL = explanation/explainable AI; TRSP = advisory or platform transparency. Trust-related outcome codes are TR = trust; FA = fairness perception; RL = reliance; AI = adoption intention. Evidence density is classified descriptively based on the number of formally included studies within each application context.

illustrates that it is dominated by robo-advisory and related AI-based investment advising, with very little representation of AI financial advice, automated credit decision-making, and crowdfunding recommendation platforms. Despite their importance to consumer-facing AI finance, no formally included studies addressed FinTech chatbots, mobile banking, or digital wallets.

The empirical focus of the review is robo-advisory/automated investment advising (RA), comprising six of the nine formally incorporated studies [17,19,35,37,38,40]. In this case, the most significant indicators are advisory/platform transparency (TRSP), followed by explanation (EXPL), and AI disclosure (DISC), which is much less common. The most common results are trust (TR) and adoption intention (AI). Reliance (RL) only shows up in the advisor-side disclosure study [35]. Although RA is the only context with high evidence density, it is better described as a clustered early literature than as a mature cumulative field.

The remaining contexts are minimal but crucial for analysis. One study on an LLM-based advisory system, focused on EXPL and results of TR and AI, exemplifies AI financial advice/advisory systems (AFA) [34]. One study also illustrates automated credit decision-making and credit scoring (ACI), where EXPL is linked to perceptions of fairness and trust, making transparency more about legitimacy and contestability than adoption [33]. Recommendation transparency is associated with trust and more general adoption/engagement-related outcomes within a platform ecology shaped by control and governance features, according to one study on crowdfunding/recommendation platforms (CRP) [21].

When considered collectively, the mapped evidence base is still heavily concentrated in robo-advisory, with only sporadic evidence found in adjacent advisory, credit, and platform settings. Accordingly, advisory transparency, explainability, and adoption-oriented outcomes continue to shape the prevailing perception of transparency in consumer-facing FinTech [17,19,37]. Although contexts like credit scoring and crowdfunding offer significant alternatives, particularly in the areas of fairness, contestability, and platform-mediated trust, they are still too few for robust generalization [21,33]. The absence of formal studies in chatbots, mobile banking, and digital wallets further indicates that major segments of consumer-facing AI finance remain largely unexamined from a transparency-and-trust perspective.

4.5. RQ4: Methodological, User-Related, and Contextual Conditioning Factors

As shown in Table 5, transparency does not produce uniform trust-related effects; its consequences depend on methodological design, user characteristics, and context. Much of the apparent variation in findings reflects differences in what was measured, in whom, and under what conditions, rather than simple contradiction.

4.5.1. Methodological Factors

A first source of variation is study design. Studies using surveys, especially in robo-advisory, often found positive links between transparency and outcomes that promote adoption (

Table 6. Factors conditioning the transparency–trust relationship.

Factor Code	Factor Category	Specific Factor	How It Conditions the Relationship	Studies	Interpretive Summary
MECH	Mechanism/mediation	Indirect effects through intermediate perceptions or evaluations	Transparency often works through mediators such as perceived responsibility, perceived usefulness, or advisory transparency rather than acting directly on trust-related outcomes.	[17,21,35,37]	Transparency frequently operates through intermediate psychological or evaluative pathways rather than as a standalone trust trigger.
PERS	Personalization and user fit	Risk profile, investment horizon, tailored recommendation design	When transparency is embedded in personalized recommendations, it may be more likely to support trust and willingness to use.	[34]	Personalization may strengthen the value of explanation, but the evidence remains limited.
STAK	Decision stakes and adverse outcome context	High-stakes or unfavorable financial decisions	Explanation becomes more salient when users face consequential or negative outcomes, such as unsuccessful credit applications.	[33]	In credit contexts, transparency functions less as persuasion and more as a procedural safeguard linked to fairness and justified trust.
EXPV	Prior experience and literacy	Investment experience, financial literacy, digital literacy	User capability shapes whether transparency is meaningful, actionable, or adoption-enabling.	[17,19,38]	Transparency is not equally effective across user groups; its value depends partly on users’ ability to interpret and use it.
VULN	User vulnerability and resource constraints	Risk aversion, low budgets, low savings, low-income status	Transparency appears especially important for financially vulnerable or low-confidence users, but often supports reassurance and adoption more than calibrated challenge.	[17,19]	Transparency may play a compensatory role for vulnerable users, mainly by supporting confidence and uptake.
HUM	Preference for human involvement	Desire for human touch or hybrid advisory	Even where transparency is valued, users or experts may still prefer some human involvement.	[19,40]	Transparency alone may be insufficient where trust depends partly on interpersonal reassurance.
MARK	Market maturity and regulatory environment	Nascent markets, regulatory ambiguity, emerging-market conditions	Low awareness, institutional uncertainty, and evolving regulation shape how transparency is interpreted.	[17,40]	Transparency appears to function differently in emerging or still-maturing FinTech environments than in more institutionalized settings.
METH	Methodological conditioning	Survey, qualitative, design, and legal-empirical study designs	Observed transparency effects vary by design type; survey and design studies more often report positive associations, whereas qualitative and legal-empirical studies emphasize contestability, fairness, and interpretive complexity.	[17,19,21,33,34,37,40]	Apparent positivity should be interpreted in light of design differences, since not all studies test causal effects in the same way.
GOV	Platform and governance features	Platform control, update frequency, learning capability	In platform contexts, transparency interacts with governance and usability features that shape trust, engagement, and downstream outcomes.	[21]	In crowdfunding and recommendation environments, transparency is embedded in a broader platform ecology rather than functioning as an isolated cue.
BIAS	Behavioral and cognitive filtering	Bias reduction and interpretability-linked decision support	Transparency-related features may matter by reducing behavioral biases or improving users’ ability to process recommendations, rather than by directly increasing trust.	[38]	Transparency may influence decision quality indirectly, but the evidence is less cleanly trust-centered than in other studies.

Note: The table summarizes conditioning factors identified across the formally included studies. Factor codes are as follows: MECH = mechanism/mediation; PERS = personalization and user fit; STAK = decision stakes and adverse outcome context; EXPV = prior experience and literacy; VULN = user vulnerability and resource constraints; HUM = preference for human involvement; MARK = market maturity and regulatory environment; METH = methodological conditioning; GOV = platform and governance features; BIAS = behavioral and cognitive filtering. The table is interpretive rather than additive: a study could contribute to more than one conditioning factor, and the listed factors reflect synthesized patterns rather than mutually exclusive causal variables.

). These connections were often through factors including perceived usefulness or advisory transparency [17,37]. Design-oriented and laboratory studies have also identified positive effects, typically within the context of comprehensive usability or interface evaluations rather than isolated transparency manipulations [19,34]. In contrast, qualitative and legal-empirical studies conceptualized transparency not merely as a direct catalyst for trust, but rather as a prerequisite for legitimacy, comprehension, or contestability [33,40].

The predominance of intention-based outcomes over behavioral ones is a second challenge. Instead of focusing on observed reliance, verification, or challenge behavior, a significant amount of the literature relies on adoption intention, willingness to use, or attitudinal trust [17,34,37]. Due to this, the evidence base is less behaviorally diagnostic and more acceptance-oriented. Mendel et al. [35] are the most obvious exception, where disclosure changed reliance indirectly through perceptions of responsibility.

A third issue is whether studies examined manipulated cues or perceived cues. Some studies employed explicit manipulations, such as advisor AI-use disclosure [35], while many depended on users’ perceptions of transparency or interpretability [17,37,38]. In various methodologies, transparency often functioned through intermediary mechanisms such as perceived responsibility, perceived usefulness, advisory transparency, or trust itself, rather than serving as a direct catalyst for trust [17,21,35,37].

4.5.2. User-Related Factors

Additionally, user vulnerability, experience, and capability affect transparency. Transparency was particularly crucial for low-literacy and risk-averse users in the Malaysian B40 context [17]. It was especially appreciated by low-budget, inexperienced, and risk-averse first-time investors in the German design study [19]. Financial literacy significantly moderated bias-related pathways in Kulkarni et al. [38], indicating that interpretability-related features do not function independently of user competence.

Prior experience also appears to matter. Studies with more experienced investors linked transparency to usefulness, interpretability, or adoption support [37,38]. Studies with novice or low-confidence users, on the other hand, focused on reassurance, process clarity, and trust-building [17,19]. Transparency may therefore support inspection for some users, but mainly reduce uncertainty for others.

Preference for human involvement is another factor. Interpersonal reassurance might not be replaced by transparency alone. Many users still preferred some human interaction prior to investing in the German robo-advisory study [19], and a hybrid robo–human model was also preferred in the Indian qualitative study [40]. Transparency may therefore increase trust without removing the perceived need for interpersonal support or human accountability.

4.5.3. Contextual Factors

Context is crucial. According to formal studies [17,34,37,40], may investigations were carried out in developing or emerging digital finance environments, such as Vietnam, Malaysia, India, and South Korea. In these situations, transparency often coexists with institutional ambiguity, dynamic regulation, or low awareness, suggesting that it may partially replace more comprehensive institutional guarantees.

Domain risk and decision stakes further affect the relationship. In robo-advisory, transparency is typically associated with fostering trust and facilitating adoption [17,19,37]. In automated credit decisions, however, explanation is more closely linked to fairness, legitimacy, and contestability in the event of negative outcomes [33].

Finally, the degree of automation matters. Transparency can change the balance of accountability between humans and machines in human-in-the-loop advisory contexts [35]. It interacts with usability and governance features in platform settings, including control, learning capacity, and update frequency [21]. In general, transparency is best viewed as a context-sensitive mechanism whose effects rely on design, users, and setting rather than as a feature that is always advantageous.

4.6. RQ5: Do Transparency Cues Support Trust Calibration or Mainly Acceptance?

A central distinction in this review is between trust calibration and mere acceptance. Calibration refers to cases in which transparency helps users judge AI more appropriately by improving understanding, supporting selective reliance, enabling contestation, or clarifying limits and logic (

Table 7. Trust calibration versus mere acceptance.

Study	Transparency Cue	Observed Outcome Pattern	Calibration Signal	Basis for Classification	Classification
[34]	EXPL	Explanation was associated with higher trust and willingness to use the advisory system.	Weak	Supports trust and intended use, but does not test whether users became more discerning or better calibrated in their reliance.	AT
[35]	DISC	Disclosure did not directly increase reliance, but indirectly increased AI reliance by reducing perceived personal responsibility.	Clear	Directly links disclosure to responsibility and reliance, showing changed accountability dynamics rather than simple uptake.	TC
[37]	TRSP	Transparency increased perceived usefulness and indirectly supported adoption intention while reducing innovation resistance.	Weak	Mediated and adoption-focused; does not test whether transparency improves differentiated trust or reliance.	AT
[17]	TRSP	Advisory transparency directly increased adoption intention and mediated key adoption pathways.	Weak	Transparency mainly supports acceptance among low-income users, not calibrated trust or contestability.	AT
[38]	EXPL	Interpretability-related features were associated with lower behavioral biases and more favorable usage-oriented outcomes.	Limited	May indirectly improve decision quality, but the study is not centered on trust, reliance, or calibrated judgment.	AT
[19]	TRSP	Transparency increased trust, satisfaction, and willingness to invest, although many users still preferred some human contact.	Weak	Shows trust-building and adoption support, but not whether transparency improves calibrated trust or challenge.	AT
[33]	EXPL	Explanation is framed as necessary for trust, fairness, and the ability to understand and contest automated credit decisions.	Strong	Treats explanation as a safeguard for informed judgment, contestability, and fairness in a high-stakes credit context.	TC
[40]	TRSP	Transparency was identified qualitatively as a pillar of trust and adoption in robo-advisory.	Weak	Transparency is discussed as trust-building and adoption-enabling, not as a mechanism for calibrated reliance or challenge.	AT
[21]	EXPL	Recommendation transparency increased trust, which then mediated broader platform outcomes, including crowdfunding success.	Mixed	Transparency is central and trust is explicit, but the outcome pattern extends into engagement and performance rather than clean calibration.	MX

Note: The table includes only the formally included studies. Transparency cue codes are DISC = AI disclosure, EXPL = explanation/explainable AI, and TRSP = advisory or platform transparency. Classification codes are TC = supports trust calibration, AT = mainly supports acceptance, MX = mixed/ambiguous, and NR = not really assessed. TC was assigned when transparency supported more appropriate judgment, informed reliance, contestability, or fairness-oriented evaluation. AT was assigned when transparency mainly increased trust, comfort, or adoption without clear evidence of better-calibrated judgment. MX was used when transparency was central and trust-related, but outcomes extended beyond calibration into broader engagement or performance pathways.

). Acceptance refers to cases in which transparency mainly increases trust, comfort, willingness to use, or adoption intention without showing that users have become more discerning or better able to respond to decision quality and stakes. This matters because transparency can appear beneficial while still leaving users poorly equipped to rely on AI well.

Table 7 shows that only a small number of formal studies clearly support trust calibration. Mendel et al. [35] and Lui et al. [33] make the strongest cases. Mendel et al. [35] demonstrated that disclosure did not directly enhance reliance; rather, it did so indirectly by diminishing perceived personal responsibility, thereby rendering accountability and reliance dynamics—rather than mere uptake—the principal concern. Lui et al. [33] regarded explanation in automated credit decisions as a protective measure for trust, equity, and contestability, associating transparency with informed judgment in high-stakes scenarios rather than mere reassurance. These studies are most akin to calibration as they associate transparency with responsibility, fairness, and challenge, rather than solely with positive attitudes.

The majority of the remaining studies are more accurately categorized as supporting acceptance. All connected transparency to increased trust, perceived usefulness, willingness to invest, or adoption intention [17,19,34,37,40]. However, they did not examine whether users became more discerning or better calibrated in their reliance. Kulkarni et al. [38] is analogous, although features related to interpretability were linked to diminished behavioral biases, indicating restricted calibration significance. Wang et al. [21] should be regarded as mixed: transparency enhanced trust and influenced broader platform outcomes, yet did not distinctly establish more suitably constrained reliance.

According to Mendel et al. [35] and Lui et al. [33], cues related to explanation and interpretability are more likely to shift toward calibration than cues related to general transparency, particularly when they are associated with comprehension, accountability, fairness, or contestability. According to Kwon et al. [37], Nazmi et al. [17], and Jung et al. [19], acceptance is more frequently supported by broader transparency, usability, and comfort-oriented cues in robo-advisory than calibration. The primary endpoints of many studies are trust, willingness to use, or adoption intention; these are significant, but they do not determine whether users are identifying system limitations, reacting appropriately to uncertainty, or adjusting reliance under error or unfavorable consequences. Therefore, the review’s primary analytical benefit is that current efforts to increase transparency in consumer-facing FinTech are still far more focused on making AI acceptable than on demonstrating that it makes trust appropriately informed, bounded, and actionable.

4.7. Role of Adjacent Contextual Literature

The adjacent studies were omitted from the formal synthesis as they did not directly evaluate a focal transparency-, disclosure-, or explainability-related cue in relation to a trust-related consumer outcome (

Table 8. Adjacent contextual literature and its interpretive role.

Study	Why Relevant	Why Not Formally Included	How Used in Discussion
Bashir et al. [26]	Shows trust formation as a central mechanism in robo-advisory adoption, especially via trust propensity and age.	No direct transparency, disclosure, or explainability cue.	Supports interpretation of trust as a mediator in robo-advisory adoption, especially in emerging markets.
Abbas et al. [27]	Identifies transparent communication, visible assurances, privacy, audits, and user education as trust-relevant conditions.	Qualitative ecosystem evidence rather than a direct transparency-to-trust test.	Used as background on the broader trust ecology surrounding transparency, privacy, and security.
Arora et al. [28]	Shows that perceived trust and trialability support psychological comfort and continuance in AI robo-advisory.	Examines trust and continuance without a direct transparency or explainability intervention.	Used to interpret psychological comfort as a trust-adjacent continuance mechanism.
Alonso-Robisco and Carbó [29]	Shows that post hoc explainability varies in reliability, with implications for how explanations should be trusted.	Methodological explainability study without consumer trust or adoption outcomes.	Used as a bridge paper on transparency-by-design, explanation quality, and explainability reliability in credit AI.
Aini [41]	Identifies AI literacy and psychological traits as antecedents of trust in robo-advisors.	Trust-formation study without a direct transparency, disclosure, or explainability treatment.	Used as an adjacent mechanism paper on AI literacy → trust → behavior in consumer-facing finance.

Note: The table reports the adjacent contextual studies, which were not included in the formal synthesis but were retained to support interpretation of mechanisms, boundary conditions, and conceptual distinctions. They are not treated as equivalent to the formally included evidence base because they do not directly test focal transparency, disclosure, or explainability cues as required for formal inclusion, or because they address adjacent methodological, fairness, literacy, or trust-formation issues. Their role is interpretive rather than evidentiary for the main review conclusions.

). They are still useful, though, because they help explain mechanisms, boundary conditions, and conceptual differences that the formal evidence alone cannot fully explicate. Their role is therefore interpretive rather than evidentiary.

In robo-advisory, these studies indicate that the establishment of trust is influenced not only by transparency-related design but also by trust propensity, age, AI literacy, trialability, and psychological comfort [26,28,41]. This elucidates why the formal robo-advisory literature frequently identifies positive associations between transparency and adoption-oriented outcomes: trust is integrated within a comprehensive mechanism framework encompassing user predispositions, perceived usability, and comfort with AI-enabled finance. Abbas et al. [27] similarly emphasizes transparent communication, visible assurances, privacy protection, audits, and user education, demonstrating that transparency is seldom observed as an isolated indicator but frequently exists within a broader trust ecosystem.

The adjacent credit and lending studies sharpen the fairness- and explainability-related interpretation of the formal evidence. Alonso-Robisco [29] demonstrates that post hoc explainability can exhibit varying degrees of reliability, whereas Abbas [27] illustrates that statistical fairness and perceived fairness may not align. These studies collectively establish a more stringent criterion: explainability must be not only technically accessible but also methodologically sound and experientially significant. This elucidates the proximity of the formal credit study to trust calibration, in contrast to numerous robo-advisory studies, wherein transparency is predominantly associated with acceptance rather than justification or contestability [27,29].

The adjacent literature contextualizes the formal findings within a more extensive theoretical framework. It underscores the significance of user predispositions, AI literacy, psychological comfort, privacy and security assurances, and the differentiation between technical and experiential fairness [26,27,28,29,41]. These studies elucidate the variability of transparency effects across different contexts and clarify that positive trust-related findings do not inherently signify calibrated trust, while maintaining the distinction between the fundamental evidence base and the expansive trust and adoption literature concerning consumer-facing AI in finance.

5. Discussion

5.1. Main Synthesis: Transparency Is Studied More as a Trust-Building Device than as a Calibration Mechanism

The main finding of this review is that transparency in consumer-facing FinTech is studied primarily as a mechanism for trust-building and adoption support rather than as a basis for trust calibration [17,19,34,37,40]. Across the formal corpus, transparency-related cues are generally associated with positive trust-related outcomes, but these outcomes are concentrated mainly in trust, willingness to use, and adoption intention rather than in more diagnostic indicators such as reliance, contestability, or fairness-oriented evaluation [17,33,34,35,37]. Overall, the available evidence suggests that transparency has been examined more often as a way of making AI-enabled financial services appear acceptable and usable than as a way of supporting appropriately bounded trust.

This overall pattern reflects the structure of the evidence base itself. The literature remains concentrated in robo-advisory and adjacent AI-enabled investment advising, with only limited evidence from automated credit decision-making and crowdfunding recommendation platforms [17,19,21,33,35,37,38,40]. Across these studies, explanation/explainable AI and broader advisory or platform transparency are the most frequently examined cues, whereas AI disclosure, responsibility attribution, user control, and information-quality cues are comparatively underexamined [17,35,37]. Trust and adoption-oriented outcomes are similarly overrepresented, while reliance and fairness appear only occasionally and in more specific settings [33,35].

A further synthesis point is that transparency effects are often indirect rather than straightforward. Positive relationships commonly operate through intermediate mechanisms such as perceived usefulness, advisory transparency, trust, or responsibility perceptions [17,21,35,37]. Taken together, the literature is best characterized as an early but promising body of work that provides useful evidence on transparency-related trust effects, while still offering only limited support for claims about calibrated trust in consumer-facing AI finance [19,21,34,40].

5.2. Theoretical Interpretation: Transparency by Design in Consumer-Facing FinTech

A key theoretical implication of the review is that transparency in consumer-facing FinTech should not be treated as a single undifferentiated construct (Table 7). Although the literature often groups disclosure, explanation, service transparency, interpretability, and control under a common transparency umbrella, these cues are not functionally equivalent [34,35,37,38]. They differ in what they reveal, which mechanism they activate, and what kind of trust-related response they are likely to shape [21,33]. The main theoretical value of the review, therefore, lies in reframing transparency as a bundle of distinct design functions rather than as a uniform property that simply produces “more trust.”

Disclosure, for example, does not do the same work as explanation. Disclosure makes AI involvement visible, but does not necessarily clarify how a system reached its output or whether that output merits reliance [35]. Its main effect may instead be to shift perceptions of responsibility and accountability. Explanation, by contrast, is intended to illuminate decision logic through rationales, criteria, or interpretive cues [33,38]. Yet explanation should not be equated with understanding, because users may interpret the mere presence of an explanation as a sign of legitimacy even when they do not genuinely comprehend the system’s reasoning [33,38].

Broader service or process transparency appears to function differently again. In robo-advisory, transparency is often framed as openness about platform operations, cost structures, service processes, or decision logic [17,19,37,40]. Such transparency can reduce uncertainty and support trust and adoption, especially among novice or risk-averse users, but it is not the same as enabling users to judge when trust is warranted, when skepticism is appropriate, or when challenge is justified [17,19]. In that sense, service transparency often supports reassurance and legitimacy more directly than calibration.

The review therefore suggests that calibration is more closely linked to cues involving control, responsibility, accountability, and fairness-relevant explanation than to openness alone (Figure 2). Cues that preserve agency, clarify responsibility, or support comprehension of adverse decisions appear more closely aligned with appropriately bounded trust than cues that mainly enhance perceived clarity or usability [33,35,37]. This is especially evident in high-stakes contexts such as automated credit decisions, where explanation matters not only for confidence, but also for contestability, fairness evaluation, and justified response [33]. The broader implication is that transparency-by-design in consumer-facing FinTech should be theorized not as a single continuum of openness, but as a differentiated architecture of cues, mechanisms, and trust consequences [21,35].

5.3. Trust Calibration Versus Mere Acceptance as the Key Contribution of the Review

The distinction between trust calibration and mere acceptance is the review’s central analytical contribution (

Table 9. Evidence gaps in the literature on AI transparency and trust in consumer-facing FinTech.

Gap Domain	Observed Gap	Evidence Basis from the Review	Why the Gap Matters	Priority Level
FinTech context coverage	The formal evidence base is heavily concentrated in robo-advisory, with very limited coverage of other consumer-facing FinTech settings.	Most formally included studies fall in RA, with only one study each in AFA, ACI, and CRP; no formal studies were identified in CHAT or MB.	This concentration limits the external validity of the review and makes it difficult to infer whether transparency works similarly across advisory, lending, chatbot, wallet, and platform-based financial services.	High
Transparency cue coverage	Explanation and broad advisory/platform transparency dominate, while disclosure, responsibility attribution, user control, and information-quality cues are underexamined.	The formal corpus is dominated by EXPL and TRSP. DISC, RESP, CTRL, and INFOQ appear only sparsely and often as secondary rather than focal cues.	The field currently cannot determine which transparency mechanisms are most effective, for whom, and under what conditions.	High
Trust outcome coverage	The literature emphasizes trust and adoption-related outcomes more than fairness, reliance, risk, contestation, or overreliance.	Across the formal studies, TR and especially AI recur most often, whereas RL appears centrally only in S2 and FA only in S7; PR, CR, CI, PC, and ENG are largely absent or peripheral in the formal set.	This outcome pattern makes it difficult to distinguish trust quality from trust quantity and weakens claims about appropriate reliance.	Very high
Trust calibration assessment	Very few studies directly test whether transparency improves appropriately bounded, reflective, or contestable trust.	In the formal synthesis, only a small subset was classified as clearly supporting TC; most studies were classified as AT, with one mixed case.	This is the central conceptual gap in the literature: transparency is often assumed to be beneficial without testing whether it improves judgment rather than merely acceptance.	Very high
Comparative cue testing	Few studies compare different transparency formats head-to-head within the same task or system.	Most studies examine one transparency construct at a time, often bundled with other design features such as personalization, usability, governance, or control.	Without direct comparison, the field cannot identify whether disclosure, explanation, transparency, or control are interchangeable, complementary, or differently effective.	High
Causal identification and design rigor	Much of the literature relies on surveys, qualitative studies, design evaluations, or normative/legal analysis rather than tightly controlled behavioral tests.	The corpus includes several SURV, QUAL, MM, and LEGAL-EMP studies, but relatively few clean experimental designs directly isolating transparency effects on trust-related behavior.	Positive transparency findings may reflect perception-level associations rather than causal effects on trust, reliance, or contestation.	High
Behavioral realism	Real or consequential financial behavior is rarely tested; many studies rely on intention, ratings, simulations, or hypothetical interactions.	Adoption intention, trust ratings, and qualitative judgments dominate; only limited evidence addresses downstream behavior, and even then mostly outside repeated real-use settings.	Trust calibration is most meaningful when decisions carry consequences, uncertainty, or error costs.	High
Longitudinal and repeated-interaction evidence	The literature is largely static and cross-sectional, with little evidence on how transparency affects trust over time.	Most formal studies are one-shot surveys, laboratory tasks, or single-evaluation designs; repeated exposure, trust updating, and post-error trust repair are largely absent.	Trust calibration is dynamic. A cue that boosts initial trust may not sustain appropriate reliance after repeated use, changing accuracy, or system failure.	Very high
Adverse and high-stakes decision contexts	High-stakes, contested, or unfavorable outcomes are underrepresented outside automated credit decisions.	S7 stands out because explanation is tied to unsuccessful credit applications, fairness, and contestability; most other studies focus on advisory uptake or trust-building in non-adverse contexts.	Calibration matters most when consumers are vulnerable to denial, loss, unfair treatment, or poor recommendations.	High
User heterogeneity and vulnerable segments	Consumer differences are acknowledged but not systematically tested across literacy, experience, income, or vulnerability profiles.	A few studies include low-income users, inexperienced users, risk-averse users, or emerging-market populations, but cross-study comparability is limited and formal moderator testing is sparse.	Transparency may help some users and confuse others; without segment-sensitive evidence, design recommendations remain blunt.	High
Human–AI relational design	The role of hybrid advisory, human fallback, and human-touch preferences remains underdeveloped.	Several studies note a continuing preference for human involvement or hybrid support, but do not systematically test how transparency interacts with human presence.	Trust in consumer finance is often relational. Transparency may work differently in fully automated versus human-in-the-loop systems.	Medium–High
Explanation quality and fidelity	The literature pays limited attention to whether explanations are accurate, faithful, comprehensible, or merely present.	Formal studies often treat explanation as present/valued, while adjacent credit studies show that explanation reliability and fairness perception can diverge from model-level properties.	Assuming that any explanation is beneficial risks overclaiming the value of transparency and obscures when explanations may mislead.	Very high
Platform and governance integration	Transparency is rarely studied as part of a broader governance ecology including control, updates, learning support, and platform signaling.	S9 is the main example linking recommendation transparency with platform control, engagement, update frequency, and success.	Consumer-facing FinTech operates in socio-technical environments, so isolated cue testing may miss the broader architecture that shapes trust.	Medium–High
Measurement standardization	The field lacks a consistent bundle of trust-calibration measures across studies.	Different studies use trust, fairness, reliance, adoption, usefulness, or adjacent constructs in non-comparable ways; several outcomes fall outside a standardized calibration framework.	This limits cumulative synthesis and makes it difficult to compare which transparency designs support informed trust rather than simple positivity.	High

Note: The table synthesizes evidence gaps from the completed formal synthesis tables rather than reproducing raw study-level extraction. Gap identification was based on patterns across application contexts, transparency cue types, trust-related outcomes, conditioning factors, and calibration classifications. Priority levels are interpretive judgments derived from the degree to which each gap constrains cumulative inference about transparency and trust calibration in consumer-facing FinTech.

). In consumer-facing FinTech, more trust is not necessarily better trust. Financial decisions involve uncertainty, unequal expertise, and the possibility of harm; accordingly, effective transparency design should not merely increase confidence in AI-enabled systems, but should also help align reliance with system logic, limits, reliability, and stakes [33,35]. A transparent system that increases confidence without improving discernment may still encourage overreliance or uncritical acceptance [17,37].

This is why adoption-oriented outcomes are insufficient to establish calibrated trust. Many of the reviewed studies show that transparency enhances willingness to use, perceived usefulness, or general trust [17,19,25,37,40]. These outcomes are meaningful, but they do not show whether users became better able to judge when trust was warranted, when challenge was appropriate, or when human support was needed [19,40]. By contrast, outcomes such as reliance, fairness perception, contestability, understanding, and responsibility attribution are more diagnostic because they bear more directly on how trust is exercised in consequential settings [33,35].

Within the formal corpus, only a small subset of studies clearly approaches calibration. Mendel et al. [35] show that disclosure influenced reliance indirectly through perceived responsibility, making accountability rather than positivity the key mechanism. Lui et al. [33] position explanation in automated credit decisions as necessary for comprehension, fairness, and contestability, thereby linking transparency to informed response rather than mere reassurance. These studies are distinctive because they connect transparency to more appropriate judgment, not simply to more favorable attitudes toward AI-enabled services.

Most of the remaining studies are better interpreted as supporting acceptance. In robo-advisory and related advisory settings, transparency increased trust, perceived utility, investment propensity, or adoption intention, but did not demonstrate that users became more discerning, more skeptical when warranted, or better able to evaluate AI under uncertainty [17,19,21,34,37,40]. Kulkarni et al. [38] is somewhat closer to calibration insofar as interpretability-related features reduced behavioral biases, yet the study still centered more on usage-oriented outcomes than on reliance, fairness, or contestation. Wang et al. [21] is best understood as mixed, because transparency improved trust and broader platform outcomes without clearly establishing more appropriately bounded reliance.

Overall, the literature provides much stronger support for transparency as an acceptance mechanism than as a calibration mechanism. The main risk is therefore conceptual as well as practical: the field may over-credit transparency by treating higher uptake as evidence of better judgment. The present evidence shows more clearly that transparency can make AI-enabled finance acceptable than that it can make trust appropriately informed, bounded, and actionable.

5.4. Context Matters: Why Transparency Works Differently Across FinTech Settings

The review also shows that transparency does not serve a single stable function across consumer-facing FinTech settings (Table 9). Instead, its apparent value depends heavily on the role the system plays, the kinds of decisions users face, and whether the context emphasizes convenience, recommendation support, or adverse and contestable outcomes. For that reason, transparency-by-design cannot be interpreted independently of application context.

In robo-advisory, transparency is typically framed as service clarity, process openness, or explainability that helps users feel more comfortable with automated financial advice [19,37]. The dominant outcomes in this literature are therefore trust- and adoption-related variables [17,40]. Because robo-advisory usually involves recommendation support rather than immediate adverse consequences, transparency is often interpreted in terms of usability, reassurance, and willingness to adopt [18,19]. This helps explain why advisory settings generate relatively positive findings while still offering limited evidence on calibration.

Credit and lending contexts shift the function of transparency in a more consequential direction. Here, explanation is tied more directly to fairness, accountability, and contestability, especially when decisions are high-stakes or unfavorable [33]. Users are not only deciding whether to engage with an AI-enabled service; they may also be trying to understand a denial, assess whether it was fair, and determine whether it can be challenged. In such settings, transparency acts less as a comfort feature and more as a procedural safeguard [27,33]. Although credit-related work remains sparse in the formal corpus, it is conceptually significant because it more directly exposes the difference between mere acceptance and calibrated trust.

Crowdfunding recommendation platforms suggest a third contextual pattern. In this setting, transparency operates within a broader platform ecology in which trust depends not only on recommendation transparency, but also on platform control, engagement, update frequency, and governance-related dynamics [21]. Transparency is therefore not merely an interface cue; it is embedded within a wider socio-technical environment that shapes both action and legitimacy. This implies that transparency-by-design may require different theoretical treatment depending on whether the system functions primarily as an advisor, a decision-maker, or a platform mediator [21].

At the same time, the absence of formal studies in chatbots, mobile banking, and digital wallets limits the external reach of current conclusions. These are important consumer-oriented FinTech sectors in which AI-driven guidance, automation, and personalization may increasingly shape trust, yet the formal evidence base remains sparse. Current conclusions about transparency-by-design in consumer-facing FinTech therefore rest primarily on advisory systems, with only limited evidence from alternative domains. This gap not only points to a future research agenda, but also cautions against overgeneralizing the present findings. Figure 3 summarizes the review’s conceptual synthesis of transparency-by-design pathways in consumer-facing FinTech.

5.5. Conditioning Factors and Boundary Conditions

Apparent inconsistency in the transparency–trust literature is better explained by boundary conditions than by direct contradiction (Table 9). Transparency does not operate uniformly; its effects depend on the type of cue employed, the mechanism it activates, the users who encounter it, and the setting in which it is embedded [33,35]. A useful synthesis is therefore: transparency cue → mechanism (e.g., understanding, fairness, control, reassurance) → trust-related outcome → either calibrated reliance or simple acceptance.

A first set of boundary conditions is methodological. Much of the literature relies on surveys and on perceived rather than experimentally manipulated cues, which likely contributes to the predominance of positive findings [17,37,38]. Survey and design studies often capture general perceptions of openness, utility, or usability, and therefore tend to support trust-building and adoption-oriented interpretations [19]. By contrast, qualitative and legal-empirical studies more often foreground fairness, contestability, and interpretive complexity [33,40]. The heavy reliance on intention-based outcomes rather than behavioral measures further limits what can be inferred. Where studies examine reliance or contestability more directly, transparency effects become more conditional and less uniformly positive [33,35].

A second set of boundary conditions is user-related. Transparency does not function identically for users with limited literacy, lower income, less experience, or greater risk aversion compared with users who are more experienced or financially secure [17,19,38]. In more vulnerable populations, transparency often plays a compensatory role by reducing hesitation and facilitating uptake rather than by cultivating challenge-ready trust [17,19]. At the same time, explanation and interpretability are unlikely to support calibrated judgment unless users possess sufficient capability to decode and use them meaningfully [38,41]. User preferences for human support matter as well: even when transparency is valued, users may still prefer hybrid or human-assisted arrangements in consequential financial contexts [19,40].

A third set of boundary conditions is contextual. Regulatory environment, market maturity, and institutional setting all shape how transparency is interpreted [17,33,34,40]. In nascent or evolving FinTech markets, transparency may partially substitute for stronger institutional assurances [17,26]. In higher-risk domains such as credit, it takes on greater procedural significance because users require fairness, justification, and grounds for contestation [27,33]. In platform settings, transparency is further intertwined with governance, control, and feedback structures rather than functioning as an isolated signal [21].

Taken together, these patterns reinforce a broader conclusion: transparency should be viewed as a context-sensitive design tool rather than as a universally beneficial feature [21,33,35]. Its effects depend on how it is operationalized, for whom it is designed, and in what decision environment it is encountered. This also helps explain why the main evidence gaps identified in the review concern cue differentiation, behavioral realism, user heterogeneity, explanation quality, and trust-calibration measurement.

5.6. Implications for Design and Practice in Consumer-Facing FinTech

First, transparency should not be treated as a generic add-on by designers, businesses, and regulators. Disclosure, explanation, process transparency, user control, and fairness-relevant justification are different forms of transparency and should not be used interchangeably [33,35,37]. More fundamentally, different transparency cues serve different design goals: some primarily support reassurance and adoption, some support comprehension and informed use, and others are more directly tied to accountability, fairness, and contestability. The practical task is therefore to match the transparency cue to the design objective rather than assume that any form of openness will have the same effect.

A second implication is that consumer-facing FinTech should be designed not only for reassurance, but also for comprehension and contestability. In advisory contexts, broad service transparency may reduce uncertainty and support adoption, but this alone is not sufficient to ensure informed reliance [17,19,40]. In higher-stakes domains such as automated credit decisions, explanation must also support justification and challenge [33]. In hybrid human–AI systems, disclosure matters because it shapes responsibility attribution between the human intermediary and the AI-supported process [35].

A third implication is that transparency design should be context-sensitive. In advisory systems, it is most useful when it clarifies process, costs, and recommendation logic, especially for novice or uncertain users [17,19,34]. In credit decisions, it should focus more directly on fairness, justification, and contestability [33]. In platform-based recommendation environments, it may need to be embedded within a wider governance architecture that includes control, update visibility, and learning support rather than being presented as an isolated cue [21].

For practitioners, the design implications differ across application types. In robo-advisors, transparency should primarily clarify service logic, portfolio processes, fees, and recommendation rationales in ways that reduce uncertainty without implying unjustified certainty. In credit interfaces, transparency should be designed less as reassurance and more as a procedural safeguard that supports explanation, fairness evaluation, and the possibility of challenge or appeal. In platform-based recommendation systems such as crowdfunding, transparency should be combined with governance features such as platform control, update visibility, and user-engagement support, because trust depends on the wider socio-technical environment rather than on recommendation clarity alone [21,33,35].

Finally, these findings matter especially for vulnerable or low-literacy users. Such groups may particularly value transparency, but this should not be mistaken for support for superficial reassurance alone [17,19]. The practical goal is not merely to make automated financial services appear more trustworthy, but to ensure that trust is supported by meaningful understanding and by appropriate review, fallback, or challenge options where needed [33,35].

5.7. Methodological Implications for Future Research

The review identifies several methodological priorities (

Table 10. Evidence-based future research agenda: design-ready gaps, recommended study designs, and priority measures.

Research Priority	What Future Studies Should Build/Test	Recommended Design	Priority Measures	Expected Contribution
Compare transparency cue types directly	Test AI disclosure, explanation, advisory transparency, user control, responsibility cues, and information-quality cues within the same consumer-facing FinTech task rather than in separate studies.	Multi-arm randomized experiment within the same interface and decision task, ideally in robo-advisory, credit, chatbot, or wallet settings.	Trust, reliance, fairness perception, perceived risk, adoption intention, comprehension, confidence, challenge/appeal behavior.	Identifies which transparency mechanisms genuinely improve informed judgment rather than merely increasing positivity toward AI.
Move from acceptance to calibration	Build studies that test whether transparency improves appropriate reliance, not just trust or adoption intention.	Behavioral experiment with variable-quality AI outputs, including correct, incorrect, uncertain, and borderline recommendations.	Appropriate reliance, overreliance, underreliance, verification behavior, override behavior, contestation, calibration error.	Directly addresses the core gap in the literature by distinguishing calibrated trust from simple acceptance.
Expand beyond robo-advisory	Study transparency in underrepresented FinTech contexts, especially automated credit, mobile banking, digital wallets, chatbots, fraud alerts, and embedded finance interfaces.	Context-specific survey experiments, field simulations, or platform-based studies across multiple FinTech domains.	Trust, fairness, reliance, user engagement, perceived legitimacy, decision quality, complaint/appeal behavior.	Improves external validity and shows whether transparency effects generalize beyond robo-advisory.
Study high-stakes and adverse decisions	Build designs around loan denials, fraud flags, suspicious-transaction notices, disputed recommendations, and rejected applications.	Scenario experiment, simulated platform study, or quasi-field design with adverse outcome conditions and response options.	Fairness perception, trust, perceived legitimacy, emotional response, willingness to appeal, comprehension of reasons, reliance adjustment.	Tests transparency where it matters most: under harm, uncertainty, contestation, and consumer vulnerability.
Separate explanation presence from explanation quality	Compare faithful, specific, comprehensible explanations against vague, generic, post hoc, or potentially misleading ones.	Controlled experiment varying explanation fidelity, specificity, completeness, and readability while holding the interface constant.	Explanation comprehension, trustworthiness judgments, fairness, reliance, error detection, explanation satisfaction.	Prevents the field from assuming that any explanation is beneficial and clarifies when explanation quality supports or undermines trust.
Integrate transparency with governance architecture	Study transparency together with platform control, update frequency, audit signals, learning support, and escalation/human-support options.	Platform-style experiment, mixed-method simulation, or interface prototype comparison with governance features manipulated systematically.	Trust, engagement, perceived control, legitimacy, reliance, willingness to continue, actual decision outcomes.	Better reflects real FinTech environments where trust is shaped by socio-technical design rather than a single cue.
Test user-sensitive transparency design	Examine how transparency works for low-literacy, novice, low-income, risk-averse, expert, and digitally experienced users.	Stratified experiment, moderated survey experiment, or mixed-method design with predefined user segments.	Comprehension, cognitive load, trust, reliance, fairness, confidence, usability, decision quality.	Supports segment-sensitive design guidance and helps avoid one-size-fits-all transparency recommendations.
Study repeated interaction and trust dynamics	Move beyond one-shot designs to test how trust changes after repeated use, changing performance, and system failure or recovery.	Longitudinal panel study, repeated-session lab design, diary study, or platform log study with staged recommendation quality changes.	Trust trajectory, reliance trajectory, trust repair, switching behavior, continued use, tolerance of error, confidence updating.	Captures the dynamic nature of trust calibration and reveals whether early transparency benefits persist or decay over time.
Build more behaviorally realistic tasks	Use tasks that involve portfolio choice, acceptance/rejection of advice, loan appeal decisions, switching behavior, or allocation choices rather than only ratings or intentions.	Incentivized behavioral experiment, simulated financial decision environment, or field-like digital platform task.	Actual choice, reliance, switching, override, deliberation time, allocation quality, economic outcomes.	Strengthens behavioral validity and shows whether transparency changes what users actually do under meaningful trade-offs.
Examine hybrid human–AI trust arrangements	Test how transparency interacts with human advisors, escalation options, and human oversight visibility.	Factorial experiment comparing AI-only, human-only, and human-in-the-loop advisory or credit-decision settings.	Trust, responsibility attribution, reliance, perceived accountability, satisfaction, human fallback preference.	Clarifies whether transparency works differently in automated versus hybrid systems and how accountability is socially distributed.
Standardize trust-calibration measurement	Develop a common measurement bundle that can be reused across consumer-facing FinTech transparency studies.	Multi-study framework paper, coordinated replication, or common-protocol empirical program.	Trust, reliance, fairness, contestability, comprehension, confidence calibration, verification, overreliance/underreliance.	Improves comparability across studies and enables stronger cumulative claims about transparency-by-design.
Link transparency to decision quality and welfare	Test whether transparency improves not only perceptions but also financial judgment quality and consumer welfare, especially among vulnerable users.	Experimental or quasi-experimental design with objective performance benchmarks and subgroup analysis.	Decision accuracy, portfolio quality, harmful-choice avoidance, bias reduction, perceived risk, welfare-relevant outcomes.	Moves the literature from attitude change toward practical consumer protection and meaningful FinTech design implications.

). Initially, subsequent research should directly compare types of transparency cues instead of analyzing disclosure, explanation, advisory transparency, and control in isolated studies. Currently, the majority of studies examine only a single cue family, constraining the ability to determine whether these mechanisms are interchangeable, complementary, or distinct [17,34,35,37].

Second, the field requires more robust causal designs. A significant portion of the literature depends on surveys, perceived-cue metrics, qualitative data, or design assessments, which, while useful, are inadequate for isolating the causal transparency effects on trust-related behaviors [17,34,37,40]. There is a need for more regulated behavioral and multi-arm experimental research, particularly to distinguish explanation from interpretability and disclosure from accountability [33,35].

Third, subsequent research should progress from intentions to reliance and behavior. The prevailing literature primarily focuses on trust ratings, willingness to use, and adoption intention, which are inadequate for assessing calibrated trust [17,19,37]. Research should progressively focus on reliance, override, verification, appeal, switching, and other manifestations of actual or simulated financial behavior [33,35].

Fourth, additional research on trust updating and repeated interaction is necessary in the literature. Although trust calibration is dynamic, most formal studies are essentially one-shot; a cue that increases initial trust may not support appropriate reliance after repeated exposure, inconsistent performance, or error [17,37].

Fifth, researchers should examine explanation quality, not just explanation presence. The formal literature often treats explanation as simply present or valued, while adjacent work shows that explanation reliability and user-perceived fairness may diverge from model-level properties [27,29,33,34]. Future studies should therefore manipulate fidelity, specificity, clarity, and contestability directly.

Finally, future work should use more realistic and higher-stakes tasks. Trust calibration is most significant when users encounter uncertain, consequential, or detrimental decisions; however, much of the existing evidence derives from low-stakes intentions or simplified assessments [17,37,40]. Stronger studies would involve portfolio allocation, loan appeal, disputed recommendations, switching behavior, or incentive-compatible tasks [21].

Overall, future research should be designed not merely to show that transparency is liked, but to determine which transparency designs activate which mechanisms, under what conditions, and with what consequences for actual financial judgment and reliance [21,33,35].

5.8. Limitations of the Review

This review has several limitations. The most important is the size and distribution of the formal evidence base. The formal synthesis includes only nine studies, and these are concentrated mainly in robo-advisory and related investment-advising contexts, with only limited formal evidence from automated credit decisions and crowdfunding recommendation platforms. This constrains the generalizability of the review and means that its conclusions are better supported for advisory settings than for consumer-facing FinTech more broadly.

A second limitation concerns methodological heterogeneity. The included studies use diverse methods, including surveys, qualitative interviews, design science, mixed methods, legal-empirical analysis, and platform-based research. Although this diversity was analytically useful, it reduced comparability and made statistical aggregation inappropriate. Several studies were qualitative, legal-empirical, or design-oriented rather than directly causal, which means that some transparency effects must be interpreted as contextual or evaluative rather than strictly causal [19,33,34,40].

A third limitation concerns the review process itself. Screening, data extraction, and quality appraisal were conducted by a single reviewer, which may have increased the risk of subjective judgment or inconsistency despite the use of explicit eligibility criteria and a structured review process. The review was also not prospectively registered.

A fourth limitation concerns retrieval scope. The search was conducted in Scopus and Web of Science only. Although these databases provided broad interdisciplinary coverage, omission of other potentially relevant databases and repositories may have limited retrieval of some studies, particularly from adjacent technical, economics, legal, and working-paper studies. The exclusion of non-English and inaccessible records further limits the coverage and generalizability of the review.

Finally, coding required harmonization across constructs labeled differently across studies. Terms such as explanation, transparency, interpretability, structural assurance, and recommendation transparency were not used consistently, and some results had to be standardized into broader trust-related categories. In addition, the adjacent contextual studies did not satisfy the stricter transparency/disclosure/explainability criterion and were therefore excluded from the formal synthesis. They were retained only to support interpretation of mechanisms and boundary conditions in the discussion. This preserved analytical clarity, but it also means that some broader insights into trust formation remain interpretive rather than part of the core evidence base.

6. Conclusions

This review examined AI disclosure, explainability, and related transparency cues in relation to trust in consumer-facing FinTech. The available evidence base is concentrated primarily in robo-advisory and AI-enabled investment advisory contexts, while remaining much thinner in automated credit decisions and crowdfunding recommendation platforms. The most frequently studied cues are explanation/explainable AI and broader advisory or platform transparency, whereas direct disclosure, responsibility attribution, user control, and information-quality cues remain underexamined.

Across the formal corpus, transparency cues are generally associated with more positive trust-related outcomes, especially greater trust, willingness to use, adoption intention, and broader reassurance or legitimacy effects. However, within the current evidence base, support appears stronger for trust-building and acceptance than for genuine trust calibration. Only a limited number of studies directly address more diagnostic outcomes such as reliance, fairness, accountability, or contestability.

The main conclusion of the review is therefore that, within the currently available literature, transparency in consumer-facing FinTech has been supported more convincingly as a mechanism for making AI systems acceptable than as a basis for ensuring appropriately calibrated trust. Future research and design should move beyond treating transparency as a generic trust facilitator and instead examine how it may support meaningful understanding, fairness-sensitive evaluation, informed reliance, and contextually relevant contestability. In consumer-facing finance, the goal is not simply greater uptake of AI-enabled services, but trust calibrated to what such systems can, cannot, and should not be expected to do. To realize the promise of transparency by design in consumer-facing FinTech, future empirical research must move beyond trust and adoption measures alone and directly assess overreliance, underreliance, fairness perceptions, and contestability behavior.