Intelligent Pathways From Access to Agency

Access has scaled. Agency hasn’t. Investing in intelligent systems can address this gap while generating both financial and impact returns.

Our Causes

Since 2011, the share of adults with a financial account has jumped by 28 percentage points. Yet one core measure—whether a person can raise emergency funds within 30 days—has barely moved. In low- and middle-income countries, it remains stuck at 56%. Meanwhile, 1.3 billion people remain unbanked, largely in rural districts, informal economies, and digital deserts.

This is the paradox at the heart of Findex 2025: the rails are in place, but the resilience hasn’t followed. The gap between access and real financial security is no longer just a policy issue—it’s a product, pricing, and delivery challenge. It’s about the usage!

For providers, driving usage is about client engagement and cost efficiency. For investors, driving usage is a growth thesis. The next wave of both financial and developmental returns will come from organizations that embed intelligence—real-time, contextual, adaptive—into their operating models. When designed responsibly, AI systems can personalize at scale, unlock thin-file credit, reduce servicing costs, and convert behavioral data into predictive value.

For providers, driving usage is about client engagement and cost efficiency. For investors, driving usage is a growth thesis. The next wave of both financial and developmental returns will come from organizations that embed intelligence—real-time, contextual, adaptive—into their operating models. When designed responsibly, AI systems can personalize at scale, unlock thin-file credit, reduce servicing costs, and convert behavioral data into predictive value.

Where AI meets real-world behavior

Findex 2025 makes one thing uncomfortably clear: access doesn’t guarantee use, and use doesn’t guarantee resilience. People save, pay, and borrow in ways that are fragmented, seasonal, irregular—and hard to interpret through traditional risk lenses.

But these patterns aren’t noise. They’re signal. They reflect how people manage liquidity in the real world. And that’s where intelligent systems can start earning their keep. Not by replacing human judgment, but by surfacing what human systems overlook.

When AI is trained to recognise real-life rhythms—daily market sales, weekend gig income, harvest cycles—it can move beyond static product menus. Instead of one-size-fits-none lending, we get timely disbursements. Instead of broad segmentations, we get dynamic thresholds and personalised nudges.

Here are five such behaviors flagged by Findex 2025, and what they imply for intelligent system design, if we choose to take on the mission:

When access becomes agency

Imagine Amara, a vendor in rural Kenya using an intelligent retail app on a platform. The platform knows she typically sells produce on Thursdays. With that data, a system can front-load an inventory loan on Tuesday and collect repayment on Saturday. No paperwork. No collateral. No branch visit. That’s what agency looks like: a system adapting to a person, not the other way around.

Platforms like Palenca enable workers to share verified earnings histories and prompt lenders to structure repayment windows around peak earning hours—not calendar months. Or consider Bud’s Drive engine, which scours transaction streams to surface “lazy cash” and nudge savers toward higher‑yield accounts—boosting deposits and loyalty in one move.

This kind of hyper-personalized alignment—between economic rhythm and service design—isn’t a fringe case. It’s the missing link between infrastructure and economic resilience.

The efficiency dividend

Serving low‑income, geographically dispersed customers can look prohibitively expensive. Ticket sizes were small, customer acquisition was manual, and support costs often outweighed revenue. AI is changing that.

By embedding intelligence into the core of workflows, providers are finding that the unit economics flip faster than expected.

• Loan underwriting: Banco Covalto cut credit‑approval times by 90 percent after embedding generative AI in its workflow.
• Fraud management: Real‑time models now freeze suspicious accounts before human review, slashing losses and investigation hours.
• Customer support: Bank of America’s Erica and NatWest’s Cora have clocked a combined 3.2 billion interactions annually, freeing staff for higher‑order conversations.
• Fundacion Genesis in Guatemala and Annapurna Finance in India are building intelligent customer service chatbots to scale their customer support teams and layer on financial and business education tools.

When cost-to-serve drops by an order of magnitude, the economics of inclusion start to look very different. What once seemed too expensive to scale suddenly becomes viable—even attractive. That’s not charity. That’s operating leverage.

Alternative data is the new collateral

Traditional underwriting relies on documents the unbanked do not possess. AI thrives on the fragments they do leave behind: prepaid top‑ups, utility payments, satellite imagery, social graphs.

• Kueski in Mexico mines e‑commerce footprints and social media patterns to score short‑term loans in minutes.
• Stori issues credit cards to “thin‑file” applicants by predicting repayment behavior from phone‑sensor data.
• In Brazil, Nubank (now the region’s most valuable bank) rewrote risk models to treat behavioral data as first‑class collateral.

The shift is strategic. When data is plentiful but messy, probabilistic identity outperforms documentary identity—and expands the addressable market by hundreds of millions of people.

For providers, this unlocks lending where formal credit files don’t exist. For investors, it creates proprietary data moats and defensible underwriting edges. And for customers, it opens doors that traditional scoring systems have kept closed for decades.

Guard‑rails that keep ambition honest

The same AI systems that enable scale and precision can also deepen exclusion, if not handled with care. Findex 2025 may be a milestone, but it also functions as a caution: where gaps have narrowed, they’ve narrowed unevenly. Where data has multiplied, it hasn’t always translated into better decisions.

Cyber fraud, algorithmic bias, and energy-intensive model training are no longer edge-case concerns. They’re live risks that can erode trust—and the business case—before returns materialize.

Responsible AI isn’t a slogan. It’s operational discipline. And at this stage, it should be non-negotiable. That means:

• Auditing features for bias—especially those that serve as proxies for gender, geography, or income. The global gender gap in account ownership is now just four points. Let’s not reopen it in code.
• Building explainability in from the start. Retrieval-augmented generation and interpretable models give providers—and regulators—transparency into why a model did what it did. Opaque credit scores don’t just lose trust. They invite scrutiny and, eventually, regulation.
• Designing for climate alignment. As AI compute scales, so does its energy footprint. Financial inclusion that relies on carbon-intensive infrastructure undercuts the resilience story.

Ambition is welcome. But without clear guardrails, inclusion at scale can become exclusion at speed. That’s not a risk serious institutions can afford to take.

What this means for investors

For investors backing financial service providers—or building them—the case for AI is no longer speculative. It’s structural.

The Findex data reveals a massive behavioural market that’s currently underserved not because it’s too risky or too poor, but because legacy systems can’t parse the signal. AI changes that. Done right, it builds margin, defensibility, and mission into the same architecture. Here’s how:

• Volume economics: Once onboarding, scoring, and support are automated, marginal cost per customer drops to cents. That opens room for $2 wallets and $20 loans—without bleeding operating margin.
• Data flywheels: Every new interaction—loan, payment, crop report—feeds future prediction accuracy. The more the system learns, the more valuable it becomes—operationally and commercially.
• Regulatory alignment: Real-time payment rails, e-KYC mandates, and data-rights regimes are becoming standard in emerging markets. The firms that embed these into their systems now will gain speed, trust, and licensing advantage.
• Blended capital: Development funders can underwrite early experimentation and model training. Equity investors can then scale those models into profitable systems, with less friction and lower cost of acquiring a customer (CAC). The capital stack is already working in the field.

You don’t have to choose between purpose and performance. It’s real-world infrastructure for inclusion—with operating leverage and portfolio growth built in.

The road from milestone to mandate

Findex 2025 confirms what many of us have suspected for years: the digital infrastructure is largely in place. The pipes have been laid. Phones are in hand. Accounts are open. But meaningful usage—the kind that builds financial resilience, not just transaction counts—still lags behind.

The distance between an 86% phone ownership rate and a 56% resilience rate isn’t a gap. It’s an investable delta. One that will be closed not by onboarding flows or awareness campaigns, but by systems that understand, anticipate, and respond to real economic behavior.

That’s where AI comes in—not as a silver bullet, but as an intelligent layer that makes inclusion adaptive, contextual, and cost-effective. When embedded responsibly, it doesn’t just scale access. It converts access into agency.

For providers, this is a design challenge. For investors, it’s a growth thesis. For both, it’s a chance to turn a digital milestone into a meaningful mandate.

The rails are already built. Now it’s time to run something better across them.