Investigating Artificial Intelligence Digital Sovereignty in Mobile Shopping Apps: A Case Study of Nigeria

This paper, authored by George Grispos and Sajda Qureshi of the University of Nebraska-Omaha and published as a preprint for the Thirty-second Americas Conference on Information Systems (2026), investigates how Artificial Intelligence (AI) in Nigerian mobile e-commerce applications affects digital sovereignty. The study is motivated by the rapid expansion of mobile applications in Africa, where mobile connectivity continues to expand and is projected to reach approximately 915 million mobile subscribers by 2030, representing 53% of the population. The researchers examine digital sovereignty through platform transparency as a key indicator of user awareness and control. The central research question asks: how does AI impact digital sovereignty in the context of socio-economic development?

Nigeria is presented as a particularly important case because it has experienced rapid growth in mobile device connectivity and digital commerce, while also continuing to face persistent technology-related fraud and scams that undermine trust in digital systems. With a population exceeding 240 million, about 64% active mobile users, and smartphone penetration at 85%, Nigeria represents one of Africa's largest and most digitally-connected markets. The country's e-commerce market ranks 38th globally, is projected to reach nearly 15 billion by 2029, and over 70% of e-commerce transactions are made on a mobile device.

The paper defines digital sovereignty as the ability to exercise authority and control over digital technologies and infrastructure. While originally linked to state power, recent research now views digital sovereignty as shared among governments, corporations, and individuals, who all influence how digital systems are built and governed. The study focuses on individual-level digital sovereignty, which refers to a person's ability to control their personal data and digital identity, and to the extent to which technology support or constrain that control. This connects to broader theories of individual autonomy and identity control, including the concept of self-sovereign identity. The authors narrow their focus to transparency as a measurable indicator of user awareness and control, defining algorithmic transparency as the disclosure of information about algorithms to enable monitoring, checking, criticism, or intervention by interested parties.

The methodology is described as interpretive, drawing on Klein and Myers' (1999) principle of abstraction and generalization. It combines three data sources:

1. Mobile forensic analysis: Three Android e-commerce applications—Jumia, Konga, and Jiji—were installed and executed on a rooted Android Virtual Device (AVD). The AVD was rooted using a Magisk-based method to allow access to protected application directories. Applications were subjected to synthetic interactions, which includes browsing listings, modifying shopping carts, and navigating interface menus, with no user accounts created. Data extraction was performed using the Android Debug Bridge (ADB), and application directories were analyzed for SQLite databases, XML files, cache and temporary files, logs, and metadata.

2. Transparency document analysis: Publicly available information for each app—including Google Play Store listing, linked privacy policies, and official platform websites—was reviewed to identify explicit or implicit disclosures of AI, automated decision-making, personalization, or algorithmic processing.

3. Contextual document analysis: The 2025 United Nations Global Survey on AI and Human Development and the UN Human Development Index were analyzed to situating the forensic findings within Nigeria's broader development context and societal perceptions of AI.

The analysis focused on three AI functional categories: (a) recommender and personalization systems, (b) conversational AI chatbots, and (c) engagement analytics platforms, classified as either locally executed or server-dependent. The authors note that "This study does not directly observe AI at runtime, but infers its presence from SDK names, filenames, logs, and other application data. The findings therefore indicate AI integration, rather than confirmed model execution."

Konga: The forensic analysis revealed a SQLite database called NCSmartech, related to Netcore Smartech, a customer engagement and personalization platform used by Konga for AI-driven product discovery, automated campaigns, and behavioral segmentation. A table called event contained timestamped activities describing product searches and interactions, navigation data, and session identifiers. A separate ChatDatabase contained JSON records that appear to contain product placement metadata... and recommendation blocks. Additional shared preferences XML files associated with Firebase including Remote Config files were recovered, with Firebase documentation indicating that the Remote Config personalization feature applies machine learning to tailor parameter delivery for individual users. The evidence suggests Konga uses AI-driven personalization and engagement features within its application.

Jumia: The jumia.db database contained tables RecentlySearch and RecentlyView that appear to store user search terms and viewed product identifiers, which may represent implicit feedback. A LiveChatConfigEntity table contains chat integration settings identifying Sprinklr as the provider, consistent with Jumia's July 2024 announcement of a customized AI-powered platform incorporating conversational bots from Sprinklr. Additionally, artifacts indicated integration with MoEngage, an AI-powered marketing platform for customer engagement. The MOEInteractions database contained tables storing a persistent device and user identifier, timestamped engagement attribute values, recent search terms, viewed products, cart activity, and hierarchical product category metadata.

Jiji: The history.db database's views table stores JSON records of viewed product listings, including listing IDs, product attributes, seller metadata, timestamps, and chat-availability flags. Integration with the Intercom messaging platform, which supports automated workflows and AI-enabled conversational services was identified through XML files and JSON records containing messaging identifiers, configuration metadata, and timestamped engagement events. Analysis also suggested interaction with Google's ML Kit framework, which provides machine learning capabilities for mobile applications, confirmed by a file containing a string named ml sdk instance id.

The authors conclude that Across all three applications, the identified components are predominantly server-dependent systems, relying on external platforms for data processing and personalization, rather than on-device AI execution. Stored user interaction data do not directly show that AI is being used, but this type of data is commonly applied in personalization and recommendation systems.

The document analysis identified limited disclosure of AI usage across the three applications. Specifically:

• None of the Google Play Store descriptions for the applications explicitly referenced any AI-driven functionality or automated decision-making, or algorithmic systems.

• Jiji provided the clearest indication of AI use, with its privacy policy explicitly stating that automated tools powered by AI operate within its platform.

• Jumia's privacy policy "outlines the use of customer data for personalization, behavioral analysis, and service improvement. While these practices imply the presence of recommendation systems or profiling mechanisms, the document does not explicitly reference AI or machine-learning technologies."

• Konga exhibited the lowest observable level of transparency. Neither its Google Play Store listing, nor its privacy policy documentation contain explicit references to AI, automated decision-making or algorithmic processing.

Overall, when present, [disclosure] tends to appear only within detailed legal privacy policies, rather than prominently in user-facing materials.

The UN survey data revealed several patterns in Nigeria:

• Moderate AI awareness: While nearly 40% of the respondents indicated knowing only 'a little' about AI, nearly 22% reported no knowledge at all, with only 13% indicating a high level of familiarity with AI.

• Uneven interaction: Nearly 41% of the respondents indicated that they had never knowingly interacted with AI, compared to just over 25% who answered interacting with AI more than once per week.

• Consumer-platform dependence: AI usage is primarily in consumer-facing domains, including social media (39%), education platforms (26%), and entertainment services (21%), with substantially lower use in government services (7%) and healthcare (12%).

• High future expectations: Nearly 57% indicating that AI is very likely to be used in education and 58% suggesting entertainment or social media applications.

• Low public trust: Less than half of the respondents (48%) indicating either strong or moderate confidence, while nearly 52% expressed limited or no confidence in AI.

The authors note that the Nigerian public's trust in AI is still cautious, and that familiarity with AI does not necessarily lead to confidence in the reliability, transparency, or regulatory oversight of these systems.

The research finds that AI affects individual digital sovereignty through limited user awareness, strong dependence on consumer digital platforms, uneven direct interactions with AI systems, and relatively low public trust in AI technologies. The authors support analytical generalization... rather than statistical generalization, generalizing the relationship between the use of AI, limited transparency, and reduced individual digital sovereignty identified in the analyzed applications. This generalization is most relevant to mobile-first digital economies that rely heavily on digital platforms, have developing regulatory frameworks, and show moderate levels of AI awareness, such as Nigeria and similar emerging markets.

Several implications for socio-economic development are identified:

1. Technological trust is critical: AI systems collect personal information through mobile applications and feed this data into large language models that interact with users, raising concerns about data control and consent.

2. Technical considerations: AI relies on user input to support conversational agents that recommend products and services, increasing the scale and efficiency of data sharing.

3. Discriminative models and predictive profiling: AI systems using predictive analytics may aggregate data from multiple sources to construct detailed user profiles containing thousands of data points. The authors connect this to Zuboff's (2015) concept of surveillance capitalism, in which data-driven systems not only predict consumer behavior, but also shape consumer actions... raising concerns about the potential for AI-driven social control.

The paper also notes that "AI features such as recommender systems are not inherently harmful; concerns arise when their operation lacks transparency. Many of these risks exist without AI, but AI amplifies them through automation, scale, and predictive capabilities."

The research advances the understanding of individual digital sovereignty by providing empirical evidence that links AI transparency with user trust and individuals' control over their personal data and digital devices. The authors conclude that the document analysis findings align with the forensic evidence recovered from the examination of the three Nigerian e-commerce applications, with recovered artifacts indicating the use of external personalization engines, automated customer interaction system and behavioral targeting infrastructure by the three providers. Collectively, the results suggest that Nigerian users frequently encounter AI within mobile applications such as Konga, Jumia, and Jiji, often without explicit awareness of interacting with these technologies.

The study acknowledges limitations, noting "This study focuses on a single country, which limits how broadly the findings can be generalized. However, Nigeria serves as an important example of a mobile-first digital economy with rapid platform growth and developing governance structures."

Future research plans include exploring technical and architectural approaches that will provide users with clearer transparency, stronger consent mechanisms, and improve content controls in mobile e-commerce applications, complementing forensic analysis with behavioral or black-box evaluation techniques, such as controlled interaction scenarios, to observe how system outputs (e.g., recommendations) adapt over time under different conditions, extending analysis to a broader range of e-commerce applications across Africa, and applying the methodology to larger samples of widely used mobile applications to further examine how AI adoption influences user awareness, trust, and control in evolving digital markets.

Based on the paper’s findings, here are specific improvements for AI systems in mobile e-commerce and similar platforms:

1. Implement runtime AI activity disclosure

• Improvement: Add in-app, user-facing indicators that trigger whenever AI-driven features (recommendations, chatbots, personalized search, engagement tracking) are active.

• What the system can do: Show a badge or notification like “AI personalized this recommendation” or “This chat is powered by an AI assistant” at the moment of interaction, not just in a privacy policy.

2. Shift from server-dependent AI to on-device processing where feasible

• Improvement: Refactor personalization, recommendation, and chatbot features to execute locally on the user’s device using frameworks like ML Kit, Core ML, or on-device LLMs, instead of sending behavioral data to external platforms.

• What the system can do: Generate recommendations and responses without transmitting user search history, viewed products, or cart data to third-party servers, giving users stronger control and reducing data exposure.

3. Introduce plain-language AI transparency summaries

• Improvement: Replace legal-only disclosures with simple, concise “AI Transparency” sections in app stores and inside settings, explicitly naming the AI components (e.g., “We use Netcore Smartech for product personalization,” “We use Sprinklr for AI chat”).

• What the system can do: Show users exactly which AI vendor is used, what data feeds it, and how to disable or adjust personalization, without requiring them to read long privacy policies.

4. Create granular consent and opt-out controls for AI features

• Improvement: Build separate toggles for each AI category: recommender systems, conversational chatbots, and engagement analytics/behavioral profiling. Require explicit opt-in, rather than automatic inclusion.

• What the system can do: Let users independently disable product recommendations, AI chat, or analytics tracking. The system then respects those choices by not invoking server-side AI services or storing the related interaction data.

5. Add a data-usage dashboard for AI-driven processes

• Improvement: Provide a within-app dashboard that shows what interaction data is stored, which AI processes use it, and which external services receive it (e.g., Firebase Remote Config, MoEngage, Intercom).

• What the system can do: Display “Recently viewed products are used to personalize your feed” and allow users to delete that data locally and remotely, giving them direct control over their digital footprint.

6. Design AI features for contextual trust-building in low-trust markets

• Improvement: Include trust cues embedded in AI interactions—such as confidence levels, explanations for recommendations, and warnings when a conversation is automated—targeting markets like Nigeria where distrust in AI is high.

• What the system can do: Automatically label AI-generated responses and recommendations with “This is an automated suggestion” and provide a “Why am I seeing this?” explanation, increasing user awareness and aligned with the paper’s finding that transparency builds trust.

7. Enable offline and local calibration of AI personalization

• Improvement: Allow the AI system to learn user preferences from on-device interactions only, using federated learning or local models, and send only aggregated, anonymized updates.

• What the system can do: Personalize search and product recommendations without building per-user profiles on remote servers, reducing the surveillance-capitalism risks identified in the paper.

8. Standardize AI disclosure metadata in app store listings

• Improvement: Add machine-readable “AI disclosure” fields to app metadata (like Google Play’s Data Safety form) that explicitly list any AI models, personalization engines, or automated decision-making systems used.

• What the system can do: Automatically generate a standardized, comparable AI transparency label for each app, helping users make informed choices before downloading and pressuring platforms to increase disclosure.

These improvements directly address the paper’s core issues: limited transparency, server-dependent data processing, user unawareness, and low public trust. The improved AI systems would be more transparent, give users explicit control, minimize third-party data sharing, and build trust in mobile-first digital economies.

The use of e-commerce mobile applications is expanding in Nigeria, creating both opportunities and risks, including fraud and reduced user control over digital technologies, raising concerns about digital sovereignty. This research examines how Artificial Intelligence (AI) in Nigerian mobile applications affects digital sovereignty, examined through platform transparency as a key indicator of user awareness and control. Using an interpretive approach, the research combines the forensic analysis of selected Android applications with contextual document analysis to identify AI features and evaluate disclosure practices. The findings show that AI is widely implemented in the applications, yet transparency about its use remains limited. A socio-economic analysis of Nigeria further shows an increasing dependence on consumer digital platforms, moderate AI awareness, and uneven patterns of interaction. By providing empirical evidence on AI transparency and platform practices, this study advances understanding of individual digital sovereignty and highlights challenges for protecting user control in AI-driven digital environments.