The impact of AI in loan underwriting and collections in Zambia

The lending system in Zambia is under pressure. For many citizens, accessing a loan is still a frustrating and slow process marked by paperwork, long wait times, and high rejection rates. Despite steady progress in financial inclusion over the years, the system is still heavily directed toward urban, formally employed individuals with collateral and clean credit histories.

Traditional underwriting methods simply haven’t kept pace with the country’s evolving financial needs. Loan officers rely on manual reviews and rigid credit criteria, leaving out the majority of Zambians who operate informally or don’t have standard financial records. Approval timelines can extend into weeks, and even when loans are granted, interest rates currently around 29.12% as of January 2025 make repayment burdensome. Collections aren’t any smoother either. Without proper borrower data or communication tools, many lenders struggle to recover payments, especially during periods of economic instability.

These inefficiencies are both operational and structural. A 2020 University of Zambia study linked loan defaults to poor loan appraisal processes, minimal borrower education, and the macroeconomic environment. And while the World Bank notes that financial inclusion rose from 60% in 2015 to 70% in 2020, inclusion in rural areas still lags significantly. According to FinScope, 30% of the population remains financially excluded, with rural financial access sitting at just 44.2%.

Given these circumstances, Artificial Intelligence (AI) is reforming how credit is assessed and collected in Zambia. This includes making smarter, faster, and more inclusive lending decisions. AI tools can analyze patterns to build a clearer picture of a borrower’s financial behaviour.

As fintechs expand and mobile usage grows across Zambia, AI is positioned to play a much larger role in closing long-standing gaps in access and efficiency. This article explores how AI is being applied to underwriting and collections in Zambia, the limitations it seeks to solve, and the opportunities it presents for building a better financial system.

Read further: How AI Agents will empower lenders

How AI credit scoring works

At its core, AI credit scoring uses more data to make smarter decisions, especially when traditional credit information doesn’t exist. Unlike manual underwriting processes that rely heavily on payslips, collateral, or formal banking history, AI credit scoring systems use machine learning algorithms to analyze a wide range of behavioural and digital signals in real time.

In Zambia, where a large number of the population is either unbanked or operates in the informal economy, this approach is a turning point. Take mobile money transactions, for instance. An AI model can assess how frequently someone uses services like Airtel Money or MTN MoMo, how much they send and receive, and whether their cash flows show financial stability. Even consistent utility payments like topping up a ZESCO meter or paying water bills can signal responsible financial behaviour.

Telecom data also plays a role. Patterns in airtime purchases, average call durations, and even SMS frequency may offer clues about a user’s economic activity. Add geolocation into the mix, such as consistent movements between home and a market stall or frequent visits to a supplier, and AI can infer informal business operations. Social network analysis deepens this further. If a borrower is financially connected to others who have strong repayment histories, that information can improve their credit score.

Lenders are also starting to factor in device data. Something as simple as the type of phone used, the apps installed, or browsing behaviour might contribute to a digital risk profile. All these data points feed into machine learning models that don’t just make binary “yes or no” decisions but calculate nuanced risk scores in seconds.

One example of this in practice is Lupiya, a fast-growing neobank in Zambia. Lupiya uses AI to assess customers who would otherwise be excluded from credit due to a lack of paperwork or traditional financial history. By analyzing mobile usage and behavioural patterns, Lupiya has been able to serve over 120,000 customers, many of whom have never interacted with a formal lender. What makes this approach powerful is that it redefines what “creditworthy” means by making sense of the digital habits people already have in their day-to-day lives.

AI in debt collection

Debt collection has never been a particularly pleasant experience for lenders or borrowers. In Zambia, much like in other parts of Africa, traditional collection methods have often leaned heavily on persistence and pressure. Think repeated phone calls, SMS reminders that border on harassment, or even public shaming in some extreme cases. Not only are these tactics stressful for borrowers, they’re often ineffective, especially when people genuinely can’t pay rather than won’t pay. What’s beginning to change the game is how Artificial Intelligence is being applied.

Rather than bombarding everyone with the same generic messages, AI-powered systems can now tailor communication based on a borrower’s behaviour. These systems analyze patterns like when someone usually receives income, how they tend to spend it, how responsive they’ve been to past reminders, and even what channels they’re most likely to engage with. So instead of calling someone at random, the system might decide that a WhatsApp message at 5 p.m. on a Friday (after payday) is more likely to get a response. It sounds simple, but the results can be drastically better both in terms of repayments and preserving customer trust.

Beyond just improving how lenders reach out, AI also helps anticipate trouble before it happens. This is where predictive analytics comes in. By scanning through payment behaviour and transaction history, AI tools can spot early signs that someone is heading toward default; maybe their payments are slowing, or they’ve stopped topping up mobile airtime as often, which could be a proxy for financial strain. Once flagged, the lender can step in early to offer support, like a flexible payment plan or short-term relief. This proactive approach not only improves recovery rates, it shows empathy. And in a financial system where trust is often fragile, that matters.

In Zambia, communication is also about language. With over 70 spoken languages across the country, and many people more comfortable communicating in Bemba, Nyanja, Lozi, or Tonga than in English, any AI-powered collection strategy needs to reflect that linguistic reality. The Bank of Zambia’s consumer protection chatbot, which operates in eight local languages, is a great example of how technology can bridge the gap. It ensures people aren’t left out just because they don’t speak the dominant language.

The regulatory environment in Zambia

As AI-powered lending gains momentum in Zambia, regulators are working to strike a balance: encouraging financial innovation while safeguarding consumer rights and ensuring accountability. This balance is important in a country where digital lending is expanding, often in spaces where traditional operations haven’t fully reached.

Current framework and recent developments

Zambia’s AI lending platforms operate under a mix of financial laws and more recent digital-focused policies. The Banking and Financial Services Act of 2017 remains the foundational legislation for all licensed financial service providers. It governs key areas like licensing, consumer protection, disclosures, and operational practices. While it doesn’t specifically reference artificial intelligence, it still shapes the compliance expectations for any entity, traditional or technology-driven offering loans in Zambia.

Complementing that is the National Payment Systems Act of 2007, which provides the legal framework for digital payment systems. This law is especially relevant because most AI lending platforms in Zambia rely on mobile money infrastructure like MTN MoMo and Airtel Money for disbursements and collections. As a result, the law indirectly governs many of the core mechanisms that enable AI-driven lending to function.

But AI brings in a new layer of complexity, especially around how personal data is collected and used. That’s where the Data Protection Act of 2021 becomes central. For AI lenders that draw on alternative data such as mobile phone usage, geolocation, and behavioural patterns, this act lays down clear rules for data privacy, consent, storage, and processing. It requires lenders to demonstrate transparency in how data is used and to obtain explicit consent from users before collecting or sharing personal information.

This is a significant development in a country where many borrowers are new to formal financial systems and may not fully understand how their data is being used to determine loan eligibility or interest rates. For AI models to be trusted, they must be explainable, and under this law, borrowers now have the legal right to question how decisions about them are made.

The role of the regulatory sandbox

Recognizing that rigid regulation can slow progress and discourage innovation, the Bank of Zambia (BoZ) introduced a regulatory sandbox, a controlled testing environment where fintech startups and AI-driven platforms can pilot their solutions under the supervision of regulators. Instead of requiring companies to meet every single regulatory condition upfront, the sandbox allows them to operate within relaxed rules for a limited time and scale, while their products are monitored and evaluated.

This framework is particularly valuable for AI-powered lenders. Developing, training, and deploying credit scoring models based on alternative data like telecom usage, geolocation, or mobile money transactions carries both opportunity and risk. The sandbox makes it possible to test scoring algorithms, repayment triggers, and borrower engagement tools on a limited user base, gather feedback, correct issues, and improve model accuracy, all before launching at full scale.

For example, a lending startup using AI to score borrowers based on mobile phone behaviour can enter the sandbox, run a pilot with a few thousand users, and test whether its algorithm fairly assesses borrowers across different regions, income levels, or gender groups. Regulators observe the pilot, track outcomes like approval rates and defaults, and assess compliance with principles such as data transparency, fairness, and consumer protection. This gives the startup room to innovate while still being held accountable.

The sandbox also helps BoZ close the regulatory knowledge gap around developing technologies. By participating directly in the testing phase, the central bank gets firsthand insight into how these AI systems work, what data they require, and what their risks are in the Zambian context. This learning process helps shape more informed regulations going forwar,d policies that reflect how innovation works on the ground.

The sandbox moves Zambia away from a reactive approach where new innovations are only addressed after they create problems and toward a proactive based model. Regulators are engaging directly with fintechs to understand what’s coming, guide safe experimentation, and co-develop policy responses that make room for progress without sacrificing stability.

Zambia joins a growing list of countries, including Nigeria, Kenya, and South Africa, that have adopted regulatory sandboxes to guide fintech development. But what sets Zambia’s model apart is how it is being integrated with broader national goals like increasing financial inclusion, enforcing data protection, and supporting responsible AI use through the National AI Strategy.

The national AI strategy

One of the boldest moves Zambia has made in the technology space is introducing its National AI Strategy (2024–2026). It’s a cross-sector plan that touches everything from farming and healthcare to education, but it’s especially relevant to financial services, where AI is already shaking up how people access credit.

This strategy sets out clear values like fairness, transparency, and accountability, which matter even more when AI’s are helping decide who gets a loan, what interest rate they pay, or how collections are handled. The government is saying: “If we’re going to use AI in these areas, we need to get it right.”

For lenders, this means AI models can’t be systems with no transparency. If a system turns someone down for a loan, there needs to be a way to explain why. If data is being collected to power that decision, people need to know how it’s being used and that it’s being handled securely and ethically.

The strategy also emphasizes local context, which is huge. Zambia doesn’t want to just import ready-made AI tools that were built with totally different markets in mind. Instead, there’s a push to invest in local talent, digital infrastructure, and partnerships so the AI tools built or adopted here actually work for Zambians and reflect the realities of how people live, borrow, and earn.

It’s forward-thinking. Rather than playing catch-up, Zambia is laying the groundwork now to guide how AI grows in the country, especially in sensitive sectors like lending, where the risks are real but so are the opportunities.

Consumer protection initiatives

Even as Zambia opens the door to AI-driven lending, regulators are making it clear that innovation should not come at the cost of basic rights and protections. As financial technologies become more complex, so too does the potential for misuse, whether through opaque credit scoring, unclear loan terms, or aggressive debt recovery methods. That’s why consumer protection is incoming as a parallel focus in Zambia’s financial modernisation efforts.

A key step in this direction is the deployment of Proto’s AI-powered consumer protection platform, introduced in partnership with the Competition and Consumer Protection Commission (CCPC) and the Bank of Zambia. More than just a support tool, this multilingual system allows borrowers to submit complaints, receive updates, and communicate with institutions in any of eight local languages, including Bemba, Nyanja, Tonga, and Lozi. It levels the playing field for consumers who might otherwise struggle with language barriers or digital literacy when resolving financial issues.

But what makes this platform especially powerful is its ability to provide regulators with an all-around view of the market. Because the system is used across multiple financial institutions, the Bank of Zambia can track patterns such as recurring complaints about certain lenders, signs of discriminatory lending practices, or spikes in unresolved disputes. This helps authorities take faster, more targeted action, closing regulatory gaps before they grow into systemic problems.

At a time when AI is being used to assess creditworthiness and automate collections, it’s equally important that it’s used to protect consumers from harm and hold financial providers accountable. This system does exactly that, embedding fairness and feedback directly into the financial ecosystem. In effect, Zambia is using AI to monitor how that access is delivered. It’s a practical, forward-thinking model: regulation and innovation growing together, with the borrower always at the centre.

Read further: Service providers for lenders in Zambia: Credit bureaus, payments providers, KYC, IT providers

Challenges and limitations

AI in lending comes with a lot of promise: faster decisions, wider access, better risk management. But in practice, turning that promise into progress takes a lot of work, especially in a market like Zambia. There are several limitations that still stand in the way. And while AI can help close financial inclusion gaps, it can also widen them if not handled carefully.

Data quality and the risk of bias

AI should be able to assess people more fairly by looking beyond traditional credit scores and using alternative data, things like mobile money activity, airtime purchases, or even utility payments. But that only works if the data is consistent and reliable.

In Zambia, the reality is messier. Mobile money usage can vary widely, and many users operate across different providers without a central record. Some rural users share mobile devices, which makes it hard to link behaviour to a single person. Data collection standards aren’t always uniform, especially among newer fintech players, and poor-quality data can easily influence how AI systems assess risk.

Even more concerning is the issue of algorithmic bias. If an AI system is trained on historical lending data that reflects past discrimination, like fewer loans to women or rural borrowers, it can end up replicating those patterns, just at a faster scale. In fact, research shows that women’s financial inclusion in Zambia still trails men’s by over three percentage points, despite gains over the years. Without deliberate checks, AI can widen those gaps instead of closing them.

Infrastructure gaps and the digital divide

Zambia’s digital infrastructure is improving, but it’s not where it needs to be for widespread, high-performing AI adoption. Internet penetration is still relatively low, and it is estimated at around 25%; rural areas are the most affected. These are the same places where financial inclusion is already lagging, with only 44.2% of rural adults formally included in financial services, compared to 80.9% in urban areas.

This connectivity gap limits the ability of AI lending systems to process real-time data or deliver mobile-based services reliably. If a platform depends on borrowers using smartphone apps, but many users have feature phones or unstable networks, the experience breaks down.

There’s also the issue of digital literacy. Not everyone is familiar with mobile apps, loan terms, or what it means to share data with an algorithm. Many people, especially older adults or those in rural communities, may not fully understand how their financial information is being used or what recourse they have if something goes wrong. This lack of digital awareness creates a power imbalance that AI tools can unintentionally exploit.

Regulatory ambiguity and compliance costs

Zambia’s regulatory environment is evolving. The Bank of Zambia has taken a more dynamic approach than many of its peers. But there are still areas for improvement, especially when it comes to how AI should be governed in financial services.

The Data Protection Act of 2021 is a big step forward. It establishes how personal data must be collected, stored, and shared. But compliance isn’t cheap, and smaller fintech startups may struggle to meet the law’s requirements without raising costs or slowing down innovation. This can create an uneven playing field where larger, well-funded platforms have the resources to comply, but smaller platforms, which are often the most innovative, fall behind.

Cross-border lending adds another layer of complexity. Companies like eShandi and Lupiya, have to navigate different data privacy rules, and financial regulations in each country they operate in. Without regional alignment, scaling AI responsibly becomes a regulatory burden.

Consumer risks in automated lending

One of AI’s biggest strengths is speed, which can also be its greatest weakness. Automated credit decisions can approve or reject a loan in seconds. But for borrowers, especially first-timers, this can feel opaque or even unfair. If you’re turned down, and the system doesn’t explain why, how do you know if the decision was accurate? And how do you fix it?

That lack of transparency can reduce trust. Borrowers may feel like they’re dealing with a faceless machine that offers no explanation or appeal process. In some cases, decisions are final and unchallengeable, especially in platforms that prioritize volume over customer engagement.

There are also concerns around interest rates and loan terms, particularly in short-term digital loans. Research into Zambia’s digital lending space has flagged instances where interest rates are high, and loan terms are poorly communicated. Some borrowers don’t realize how much they owe until repayment is due, and by then, it’s too late; they’re caught in a debt spiral.

Economic impact and financial inclusion outcomes

Artificial Intelligence is transforming who gets access to credit and what that access means for Zambia’s economy at large. The numbers are beginning to tell a story of meaningful progress, though it’s not without gaps or challenges.

Measurable progress in financial access

Over the past decade, Zambia has made some of the most significant gains in financial inclusion on the continent. Back in 2009, only 37.3% of adults had access to formal or informal financial services. By 2020, that number had jumped to 69.4%, according to FinScope data. While many factors contributed to that growth, like mobile money and fintech innovation, AI has played an important role behind the scenes.

What makes AI different is its ability to work with the kind of data that’s actually available in Zambia. Instead of relying on formal credit histories or payslips (which many people don’t have), AI models can make lending decisions based on alternative signals, things like mobile money usage, airtime purchases, or even how regularly someone tops up their electricity. That has opened the door for thousands of borrowers who were previously invisible to traditional banks.

The growth of mobile money, which rose from 14% to 58.4% between 2015 and 2020, didn’t just improve payment access; it also created a data trail. Every transaction, transfer, or bill payment helps build a borrower profile that AI can learn from. As more people use mobile money, more data becomes available, and more accurate lending decisions follow. That, in turn, encourages even more usage. It’s a cycle that feeds itself, and AI sits at the centre of it.

Gender and rural inclusion

Between 2015 and 2020, women’s formal financial inclusion in Zambia rose from 33% to 59%. That’s a significant increase in just five years. AI has helped drive this by removing some of the friction women often face in traditional banking, like needing collateral or formal employment documentation.

As of 2020, 71.2% of men were financially included, compared to 67.9% of women. The divide may seem narrow on paper, but it shows deeper structural issues like limited phone ownership, uneven access to digital tools, or cultural barriers that still make it harder for women to engage with financial platforms confidently.

The rural-urban divide also paints a similar picture. In urban areas, financial inclusion is high, 84.4% as of 2020. But rural inclusion lags far behind at 55.9%. And this isn’t because AI tools don’t work in rural settings, it’s because the underlying infrastructure often isn’t there to support them. Unreliable internet, lower smartphone ownership, and weaker agent networks all make it harder for AI-powered platforms to function where they’re arguably needed most. That said, the groundwork being laid through national strategies and mobile network expansion suggests the rural gap could narrow over time.

Broader economic effects

The impact of AI-enabled lending doesn’t stop with personal loans. In fact, one of the biggest economic ripple effects has been in the small business sector. Many Zambian SMEs struggle to grow because they can’t access working capital. With AI, lenders can now assess the risk of informal traders or gig workers even if they’ve never had a bank account or credit score.

This has made it possible for entrepreneurs, market vendors, small-scale farmers, and delivery workers to access short-term financing that helps them restock, expand, or manage cash flow. And when these businesses grow, they hire more people, support local supply chains, and contribute to the overall economy. That’s the multiplier effect AI lending is quietly enabling.

On the lender side, AI also improves efficiency. By automating parts of the underwriting process and using predictive tools for collections, platforms can cut costs and reduce the risk of defaults. These savings allow them to offer more competitive interest rates, making formal lending a viable alternative to informal loan sharks who often charge exorbitant fees.

In other words, AI isn’t just helping people access credit. It’s helping reshape the economics of lending itself, making it more scalable, more sustainable, and more inclusive.

Read further: Effective loan collections for lenders in Zambia

Future outlook

One of the clearest shifts we’re seeing is how AI is beginning to plug into larger digital ecosystems, rather than operating in isolation. Think of open banking: as Zambia’s financial system becomes more interconnected, AI systems will be able to draw from a wider pool of customer data from mobile wallets and savings accounts to insurance histories and even retail purchases. The more context AI has, the better it can understand risk and personalize financial offerings. For borrowers, this means more relevant loan products, faster approvals, and fewer unnecessary rejections.

On the back end, blockchain technology is being explored as a way to solve certain problems in the lending space, particularly around identity verification and transaction transparency. In markets where fraud and data tampering are real risks, blockchain can provide a tamper-proof record of financial activity. This helps AI models trust the data they’re using, which is essential for credit scoring that’s both accurate and fair.

Meanwhile, AI capabilities themselves are evolving fast. Natural Language Processing (NLP), for example, is moving beyond simple chatbot scripts. We’re now seeing AI tools that can understand local dialects, respond in conversational Bemba or Nyanja, and pick up on tone and context in conversations. That’s a big win in a country as linguistically diverse as Zambia, where language barriers have historically excluded many people from digital services.

Another developing area is computer vision. AI systems that can process and interpret images. In lending, this could be used to verify assets or evaluate small businesses remotely. Imagine a market trader snapping a photo of their inventory or a farmer sharing a video of their farm setup, and an AI system analyzing it to determine business activity and loan viability. This reduces the need for costly, in-person inspections and makes the underwriting process faster and more accessible.

There’s also growing interest in federated learning, which addresses one of AI’s biggest obstacles: data privacy. Instead of pulling sensitive data into one central database, federated learning lets AI models improve across decentralized devices. This means better privacy protections for users and less regulatory friction for platforms, a win-win, especially in a regulatory environment that’s still finding its footing.

Getting there won’t happen by accident. It’ll take coordinated action from all sides of the ecosystem. Banks and fintechs need to invest in proof-of-concept projects, testing new AI tools in real-world settings, particularly in rural or underserved areas where traditional models don’t work well. Regulators, led by the Bank of Zambia, need to establish clear, practical guidelines around AI, ones that encourage innovation but also protect users from harm. 

And the government has a role to play too, especially in expanding internet access, supporting AI education, and ensuring that the benefits of new technologies are widely shared. The Second National Financial Inclusion Strategy (2024–2028) lays out a strong roadmap for digital transformation. If Zambia can align this agenda with a deliberate focus on ethical, transparent, and inclusive AI, it could reshape the financial lives of millions. 

Read further: A cultural view of loan defaults in Zambia

From experimentation to everyday use

AI isn’t a miracle drug for Zambia’s lending and collections challenges, but it’s proving to be a practical tool where it matters most: reducing friction, improving accuracy, and reaching people who’ve long been left out of the formal financial system. What’s encouraging is that this shift isn’t just theoretical. We’re already seeing signs of impact from smarter underwriting decisions and personalized repayment strategies to clearer regulatory efforts and stronger consumer protections.

But the next phase will be less about technology itself and more about how it’s applied. How do we make sure AI models are fair? How do we build systems that are transparent, inclusive, and able to evolve with Zambia’s complex social and economic landscape? These are the questions that matter now.

The groundwork is being laid by startups testing bold ideas, by regulators opening space for experimentation, and by borrowers who are increasingly open to digital solutions. The opportunity is here. What happens next depends on how consistently all players, technology providers, financial institutions, regulators, and consumers can align innovation with trust, efficiency with fairness, and scale with inclusion.