when the lights go out on the loan

Problem statement

Pay-as-you-go (PAYGo) solar lets off-grid households buy a solar home system on installments, unlocking the system in increments as they pay via mobile money — the dominant model for extending electricity to people the grid will not reach for years. But the financing engine is failing under its own customers: roughly 40% of PAYGo customers in Kenya and Uganda struggle to keep up with payments, and the difficulty compounds over the loan's life — only about 20% of customers in their first three months fall behind, rising to 40% by months four to six. The companies extending these micro-loans frequently cannot distinguish a customer hit by a temporary income shock (who would recover with forbearance) from a genuine defaulter (who will not pay), so they apply blunt responses that either write off recoverable customers or chase unrecoverable ones, eroding both portfolios and the households' access to light.

Why this matters

PAYGo is the financing backbone of off-grid energy access for hundreds of millions of people, and its viability determines whether private capital keeps flowing to electrify the last mile. When repayment collapses, three things break at once: the household loses light (the system locks), the company's portfolio-at-risk balloons and investors retreat, and the broader promise that the poor are "bankable" through asset finance is undermined. The stakes compound because the customer base is precisely the population least able to absorb a missed payment: only about a third of struggling customers in these markets report stable monthly income, 40% already carry other debts, and over half report that climate shocks — drought, floods, crop disease, extreme heat — have hit their earnings. The same households the model exists to serve are the ones the model is most likely to strand.

What’s been tried and why it hasn’t worked

The industry's instinct during its growth phase was to maximize sales — agents were paid on units sold, not on whether customers ultimately paid off their systems — which front-loaded the portfolio with customers whose ability to pay was never properly assessed. Credit-scoring and income-assessment at onboarding has been weak: companies sold to whoever would sign, and the resulting portfolios manifest more credit risk as they mature, because repayment is strong at the start and decays over time, so growth masked the rot until cohorts aged. Remediation efforts then run aground on an information gap: more than half of Ugandan and a third of Kenyan customers don't accurately know their own remaining balance or payment timeline, and companies lack the touchpoints to learn early which silent non-payers are in temporary distress versus permanent default. Generic, late-stage collection — locking the unit, sending reminders — treats both groups identically, pushing recoverable customers into default and wasting effort on the unrecoverable. The core unsolved problem is early, low-cost discrimination between distress and default, paired with a tailored response for each.

What would unlock progress

Progress hinges on turning the rich, real-time signal these companies already collect — payment cadence, top-up size, system-usage telemetry, mobile-money patterns — into an early-warning model that flags a struggling customer in the critical months 2–4 and routes them to the right intervention (a restructured payment plan for distress, decisive action for default) before arrears harden. The adjacent precedent is mainstream digital lending and microfinance, where behavioral and alternative-data scoring plus graduated, automated collection workflows already triage borrowers; the gap is adapting these to the PAYGo data shape and the reality that the asset itself (the solar system) is both collateral and the household's only light.

Entry points for student teams

A data-science team could build, on anonymized or synthetic PAYGo repayment data, an early-warning classifier that predicts which customers will fall into arrears in the next 30–60 days from their first months of payment telemetry, and prototype a decision-tree that maps each risk tier to a specific intervention. A service-design team could redesign the first-90-days customer journey — the touchpoints, balance-transparency tools, and agent incentives — to surface distress early and shift agent pay toward sustained repayment rather than the sale. A development-finance team could model the growth-versus-sustainability tradeoff for a representative portfolio, quantifying how much sales growth a company should forgo to keep portfolio-at-risk viable for investors. Relevant skills: machine learning, behavioral economics, service design, and impact finance.