CLOSING THE PENSION GAP: CAN AI DRIVE INFORMAL SECTOR INCLUSION

 INTRODUCTION

Ghana’s digital transformation agenda has, over the past decade, moved from ambition to measurable progress. Foundational systems—ranging from the Ghana Card and mobile money interoperability to digitised public services and online tax administration—have reshaped how citizens interact with the state. Beyond improving administrative efficiency, these developments have established critical building blocks for broader financial inclusion, particularly among underserved segments of the economy. According to the World Bank’s Global Findex data, over 80% of adults in Ghana now have access to financial accounts and regularly use digital payment platforms, reflecting the scale of this transformation.

Yet a fundamental challenge remains—the persistent pension coverage gap within Ghana’s informal sector. According to the International Labour Organization (ILO), approximately 78–80% of Ghana’s workforce is employed in the informal sector, while pension coverage within this segment remains low, estimated at just over 10% in recent years. A significant proportion of workers therefore continue to operate outside formal pension arrangements. Participation is limited, contribution patterns are irregular, and engagement with structured financial systems is often minimal. While digital infrastructure has reduced barriers to access, the question of how to translate connectivity into sustained pension participation remains unresolved. The issue is no longer simply one of access; it is about designing systems that respond meaningfully to the economic realities of informal workers.

This challenge is increasingly shaping the next phase of pension system modernisation. Attention is shifting beyond digitisation toward how data, platforms, and institutional processes can be aligned to support inclusion at scale. The objective is not merely to create digital access points. It is to enable systems that are responsive, coordinated, and capable of supporting individuals whose income patterns fall outside traditional models.

Within this evolving landscape, artificial intelligence (AI) is emerging as a potentially transformative layer. If digitalisation enables access and interoperability supports coordination, AI introduces the capacity for systems to analyse, learn, and respond more adaptively. Properly deployed, it can support personalised engagement, improve contribution consistency, strengthen oversight, and enhance system integrity. For example, AI-enabled systems can leverage mobile money and USSD-based platforms—already widely used across Ghana—to deliver timely prompts and contribution recommendations tailored to irregular income flows.

However, AI also raises important governance questions. Systems designed to optimise efficiency may, if not carefully structured, reinforce existing exclusions or introduce new forms of bias. Its deployment must therefore be situated within a broader institutional and policy framework that balances innovation with accountability, inclusion, and public trust.

This article examines how artificial intelligence can be applied within Ghana’s pension administration landscape to advance informal sector inclusion. It explores practical applications in improving enrolment, strengthening contribution patterns, and enhancing oversight, while also addressing the structural constraints related to data quality, system integration, and governance. At its core, the discussion asks a more fundamental question: can AI help move pension systems from access to sustained participation? The argument advanced is that AI is not a solution in itself, but an instrument—whose effectiveness depends on how it is designed, governed, and integrated within the broader system.

 

 

DEFINING THE INFORMAL SECTOR IN THE CONTEXT OF PENSION INCLUSION

For the purposes of pension policy in Ghana, the informal sector may be understood as comprising individuals and small-scale economic actors engaged in legitimate income-generating activities. These activities operate outside formal employment structures and are not integrated into standard payroll-based systems.

As reflected in Ghana's National Pensions Act, 2008 (Act 766), such economic activity is typically characterised by a low level of organisation, limited separation between labour and capital, and small-scale operations. In practical terms, the sector consists largely of self-employed individuals, microbusiness operators, and workers in unregistered or loosely structured enterprises. According to the International Labour Organization (ILO), this segment accounts for approximately 78–80% of total employment in Ghana, underscoring its centrality to the national economy.

From a pension perspective, the defining features of the informal sector extend beyond the absence of formal registration to the nature of economic participation itself. Workers in this segment of the economy—predominantly self-employed—typically:
• Earn irregular and unpredictable incomes
• Operate outside formal payroll and employer-sponsored benefit systems
• Bear full responsibility for their own long-term financial security, including retirement savings
• Have limited access to structured financial and pension products

It is important to distinguish this group from formal self-employed professionals—such as lawyers, accountants, and consultants—who, although self-employed, generally operate within more structured, higher-income environments and have greater access to financial planning tools. The primary target for micro-pension interventions is instead the vulnerable informal self-employed population: the market trader in Makola, the seamstress in Nima, the tro-tro driver, the hairdresser, the mobile phone repairer, or the small-scale food vendor. For this group, income insecurity and limited financial access significantly constrain participation in traditional pension arrangements.

This segment spans both rural and urban economic activity, yet it is unified by the absence of stable, institutionalised mechanisms for income security in old age. The challenge, therefore, lies not simply in legal informality, but in the underlying economic structure within which these individuals operate.

It is this structural reality—rather than informality as a legal category—that presents the central challenge for pension inclusion. Designing effective pension systems for this segment requires approaches that accommodate income variability, reduce administrative friction, and support voluntary, flexible participation.

FROM DIGITAL INFRASTRUCTURE TO INTELLIGENT INCLUSION

Digital transformation in public systems typically evolves through three stages:

Stage One: Basic Digitisation — where analogue processes are converted into digital formats, improving efficiency within individual institutions.

Stage Two: Optimised Service Delivery Through Digitalisation with Interoperability — where systems exchange and apply data in coordinated ways, reducing duplication and enabling more seamless public services. Ghana has made notable progress across both stages. This progress is reflected in the rapid expansion of digital financial services; according to the World Bank, over 80% of adults in Ghana now have access to financial accounts and actively use digital payment platforms.

Stage Three: Intelligent Systems (Now Emerging) — a third stage is now emerging, characterised by the transition toward intelligent systems. Such systems are defined not merely by automation, but by their ability to analyse data, identify patterns, and respond dynamically to user behaviour. In pension administration, this distinction is critical. While digitisation enables registration and digital contributions, it does not address the deeper challenge of irregular participation, particularly among informal sector workers.

The concept of intelligent inclusion reflects a shift from systems that provide access to those that actively support participation. Rather than relying on individuals to navigate administrative processes, intelligently designed systems can identify patterns—such as income variability or contribution gaps—and respond with timely, context-sensitive interventions. In practice, this may include prompting contributions during periods of higher income or flagging emerging gaps in saving behaviour before they become prolonged.

This transition, however, is not purely technological. It depends on the quality of underlying data, the degree of system integration, and the strength of governance frameworks. Without these foundations, intelligent systems risk amplifying fragmentation rather than resolving it.

Within this evolving framework, artificial intelligence is not simply an additional tool; it becomes a critical enabling layer that underpins the shift from access to sustained participation.

FUNCTIONAL DIMENSIONS OF AI IN PENSION ADMINISTRATION

Artificial intelligence in pension administration is best understood not as a single technology, but as a set of capabilities. These capabilities enable systems to process data, generate insights, and support decision-making. Its application can be viewed across four functional dimensions.

The first is operational AI, which enhances efficiency in routine processes such as onboarding, customer support, and transaction handling.

For instance, Kwame, an informal worker, dials *92046# on his basic phone. An AI-powered USSD chatbot guides him through pension registration in Twi, asking simple questions one at a time. Within three minutes, he is registered without visiting an office or speaking to a person.

While valuable, its impact is largely incremental.

The second is analytical AI, which enables deeper insights into user behaviour. By analysing contribution histories and income patterns, systems can identify individuals at risk of disengagement or detect emerging gaps in participation.

In practice, analysis of contribution data may reveal that market traders in Kumasi Central Market exhibit a 40% drop-off rate after their third contribution. This pattern can then be flagged to administrators, who design targeted re-engagement strategies for this segment, including peer testimonials from successful contributors within the same market. This is particularly relevant in the informal sector, where contribution behaviour is often irregular and difficult to predict.

The third is regulatory AI, which supports oversight through anomaly detection, compliance monitoring, and early risk identification.

Consider a scenario in which an AI system monitoring a pension scheme detects that a group of newly registered members in a specific market area consistently make only a single minimum contribution and then become inactive. The pattern is repeated across dozens of accounts linked to the same registration agent. The system flags this behaviour to administrators, who discover that the agent is prioritising rapid enrolment to meet targets or earn commissions, without ensuring genuine participation. By identifying such patterns early, regulatory AI enables targeted intervention, improves compliance, and helps maintain the integrity of the system.

The fourth is behavioural AI, which uses personalised prompts and recommendations to influence user engagement. By aligning contribution suggestions with individual income patterns, it supports more realistic and sustainable participation.

In another case, the system observes that Kofi, a tro-tro driver, receives mobile money transfers averaging GHS 150–250 every Friday evening. At 7 p.m., when his balance exceeds GHS 200, he receives a prompt: “Kofi, you’re doing well this week. Save GHS 20 for your pension? Reply 1 to contribute.” By aligning the prompt with periods of higher income, the system makes engagement more likely.

The deployment of analytical and behavioural AI requires careful attention to data privacy, informed consent, and transparency. These systems necessarily analyse workers' financial patterns and transaction behaviours—creating detailed profiles of their economic lives. Without robust safeguards, such capabilities risk enabling surveillance, unauthorized profiling, or commercial exploitation of vulnerable workers' data. Workers must understand what data is collected, how it is used, and retain the right to opt out of AI-driven analysis while still participating in pension schemes. Privacy protections and meaningful consent are not merely legal requirements—they are fundamental to maintaining the trust upon which voluntary pension participation depends. Regulatory frameworks must therefore establish clear boundaries on data use, mandate transparency in AI operations, and enforce strict accountability for any misuse.

Taken together, these dimensions illustrate that AI's value lies in integrating operational efficiency, behavioural insight, and regulatory intelligence into a coherent system. Its effectiveness, however, depends on alignment with policy objectives, the availability of reliable data, and the presence of robust governance frameworks.

INCLUSION RISKS AND THE INFORMAL SECTOR REALITY

Despite its potential, AI introduces risks that are particularly pronounced in informal sector contexts. Informal economies are characterised by variability rather than uniformity. Income flows are irregular, work arrangements are fluid, and data patterns are often incomplete. According to the International Labour Organization (ILO), this segment accounts for approximately 78–80% of total employment in Ghana, meaning that such variability is not exceptional but systemic. AI systems trained on formal sector data may therefore interpret normal economic behaviour as risk or non-compliance.

This creates a fundamental tension. Patterns that appear irregular within a dataset may in fact reflect the economic realities of informal work. Intermittent contributions, for example, often result from income constraints or seasonal earnings rather than disengagement. If AI systems fail to account for this, they risk reinforcing exclusion through misclassification—penalising behaviour that is, in context, entirely rational.

A related challenge is data invisibility. Not all informal economic activity is digitally captured, despite significant advances in financial inclusion. Individuals operating outside formal platforms or digital payment systems may remain unseen, limiting the reach and effectiveness of AI-driven interventions.

This raises a further design consideration: accessibility. Even where digital systems exist, barriers related to language, literacy, and interface design can limit meaningful participation. AI offers the potential to address these constraints through multilingual interfaces, voice-enabled interactions, and reduced reliance on text-based systems. However, without deliberate implementation, these capabilities may remain underutilised. Inclusive design must therefore be treated as a core requirement, ensuring that systems are accessible to users with varying levels of literacy and across diverse linguistic contexts.

Addressing these risks requires deliberate design choices. Datasets must be representative, models must be continuously tested for bias, and alternative access channels must be maintained. Inclusion cannot depend solely on digital visibility or standardised user behaviour.

Ultimately, AI must be designed to interpret complexity rather than simplify it. Without this, it risks replicating—and potentially deepening—the very exclusions it seeks to address.

DATA, SYSTEM INTEGRATION, AND THE LIMITS OF AI

AI’s effectiveness is fundamentally dependent on data quality and system integration.

Pension administration involves multiple data sources, including identity systems, payment platforms, and financial records. Where these operate in isolation, data remains fragmented. AI applied within such environments risks generating incomplete or misleading insights. In practice, individuals may transact across mobile money platforms, bank accounts, and informal savings mechanisms, with no single system capturing the full scope of their financial activity.

Partial data leads to partial intelligence.

This limitation is particularly pronounced in informal sector contexts, where economic activity is often distributed across multiple channels. Without integration, no single dataset provides a comprehensive view of an individual’s financial behaviour. As a result, AI systems may produce insights that are technically accurate within a dataset but incomplete in reality.

AI cannot substitute for integration. It performs best in environments where data flows are coherent and standards are harmonised. In the absence of these conditions, its impact remains constrained.

A further limitation lies in the nature of predictive capability. AI generates probabilistic insights rather than certainty. In dynamic economic environments—especially those characterised by income variability and informality—over-reliance on such outputs can create false confidence and lead to inappropriate decision-making.

Effective deployment therefore requires parallel investment in data governance, system integration, and institutional capacity. AI should be understood as a dependent layer within a broader ecosystem, not a standalone solution.

GOVERNANCE, ETHICS, AND PUBLIC TRUST

The deployment of AI in pension administration is, at its core, a governance issue rather than a purely technological one.

Transparency is central. Systems must be able to explain how decisions are made, particularly where they affect access, eligibility, or outcomes. In the context of automated processes, such as contribution monitoring or eligibility assessments, opaque systems risk eroding trust and limiting user confidence.

Accountability must remain clear. AI should support decision-making, not replace it. Human oversight is essential, particularly in sensitive areas such as compliance, enforcement, and dispute resolution, where contextual judgement remains critical.

Fairness requires continuous monitoring. Bias embedded in data can translate directly into biased outcomes, especially in informal sector contexts where data is often incomplete or uneven. Governance frameworks must therefore include mechanisms for detecting, assessing, and correcting such biases over time.

Data protection is equally critical. As digital financial activity expands—driven in part by the rapid growth of mobile money and digital transactions—AI systems increase the scope for data analysis and user profiling. This requires robust safeguards around consent, data access, and security to protect individuals from misuse or unintended exposure.

Institutional capacity is also essential. Regulators and administrators must possess the technical and operational understanding required to oversee the systems they deploy. Without this capacity, reliance on external technologies may introduce new vulnerabilities and weaken oversight.

AI in pension systems is not merely a technological upgrade; it is ultimately a test of institutional integrity, requiring systems that are transparent, accountable, fair, and worthy of public trust.

FROM AUTOMATION TO ANTICIPATORY SOCIAL PROTECTION

Artificial intelligence enables a shift from reactive to anticipatory models of social protection, marking an important evolution in how pension systems engage with users.

Traditional systems depend on individuals initiating engagement. By contrast, AI allows systems to respond to patterns and life events, enabling timely and context-sensitive interventions. In practice, this may include identifying contribution gaps early or prompting engagement during periods of higher income.

Such responsiveness reduces friction and aligns services more closely with real-world behaviour. For informal sector workers, it can help bridge persistent gaps in awareness, consistency, and participation.

However, anticipatory systems must be carefully governed. Interventions must respect user consent and preserve individual autonomy, particularly where behavioural prompts are driven by predictive insights. Not all system-generated signals should result in action.

Importantly, anticipation does not eliminate uncertainty. Informal economies remain dynamic and only partially observable, limiting the completeness of any predictive model. Human engagement and institutional responsiveness therefore remain essential components of effective pension systems.

The objective is not full automation, but the development of systems that are more responsive, adaptive, and supportive of user needs.

CONCLUSION

Ghana’s pension system is approaching a critical inflection point. The foundations of digital transformation are in place. The challenge now is to leverage these foundations to expand inclusion, particularly within the informal sector.

Artificial intelligence offers practical pathways to support this transition. It can improve enrolment, strengthen contribution behaviour, and enhance oversight. When effectively aligned with existing digital infrastructure, it enables more adaptive and responsive pension systems.

Its effectiveness, however, depends on more than technology alone. Data quality, system integration, and governance frameworks remain decisive. Without these, AI risks reinforcing existing exclusions rather than reducing them. The critical question is therefore not whether AI will be used, but how it will be governed and integrated within the broader system.

For informal sector workers, this is not merely a technical consideration. It is a question of long-term financial security. AI must be deployed in ways that reflect economic realities, protect individual rights, and sustain public trust.

The answer, therefore, is conditional but clear: AI can drive informal sector inclusion in Ghana—but only if it is designed, governed, and implemented in ways that prioritise inclusion, accountability, and trust. Ultimately, AI should be understood as an instrument rather than an end in itself. Its value will be measured not by its sophistication, but by its contribution to a pension system that is more inclusive, more responsive, and more trusted.