Remittances and the data they leave

Around a quarter of Nepal’s GDP arrives every year as remittances — money sent home by Nepalis working in the Gulf, Malaysia, Korea, India, and increasingly Europe. This is the single largest source of foreign exchange in the country. It is also, considered as data, an unusually rich and unusually under-used asset.

The shape of the data

A single remittance carries, at minimum: a sender, a receiver, an amount, a corridor (which country to which district), a frequency relative to that sender’s history, a method (bank, IME, Western Union, Hundi), and a date. Aggregated across IME Pay, Prabhu Money Transfer, Western Union, MoneyGram, and the bank-to-bank rails, the flow tells you — in something close to real time — how labour-exporting Nepal is going.

Three patterns are visible in this data that nobody else in the country can see:

Which districts are sending which kinds of workers to which destinations — a granular picture of Nepali labour migration that the government formal datasets do not have.
How household income changes month-over-month, by destination — when a worker loses a job in Qatar, the inflow to Kavre stops a month later, and the model knows before any survey would.
Where remittance is being spent versus saved — by tracing the receiver’s subsequent transactions on the same platform.

What this data could power

Five concrete uses, in rough order of feasibility.

Better credit for remittance-receiving households. A household that has received Rs. 50,000 a month from a son in Doha for three years is a different credit risk than one that has not. A cooperative could lend against this future flow at much lower rates than a bank would, if the platform shared (with consent) a verifiable inflow record. This is the most immediately commercial use of the data.

Real-time labour-migration intelligence. When inflow from a specific corridor — say, Saudi Arabia to Banke — drops sharply over two months, it almost always means something has happened to the labour market there: a contract cancellation, an immigration crackdown, a sectoral slowdown. A government that watches this data weekly would know before any embassy did.

Early warning for household stress. When a household that had been receiving steady remittance for years sees the flow stop, and then begins drawing down savings on the same platform, the model can flag this as financial distress — and a cooperative or NGO could intervene before the household is pushed into informal high-interest borrowing. This needs careful consent design but is otherwise a clean public-good use.

Geographic and seasonal forecasting. Remittance flows are strongly seasonal — Tihar, school admission season, monsoon agricultural inputs. A model that forecasts inflow by district and season would let cooperatives and merchants plan inventory and credit lines with much better information than they have today.

Macroeconomic nowcasting. Nepal’s official GDP figures are released with a lag of months. Real-time remittance data is one of the cleanest leading indicators the country has. A central bank that integrated remittance-platform data into its dashboards would, in effect, have a higher-resolution view of the economy than the official statistical system provides.

Why none of this is happening at scale yet

Three reasons, all fixable.

The data sits in commercial silos. Each remittance platform sees only its own flow. The country-level picture requires aggregation that no single firm has incentive to build, and no public body has the authority to compel. A privacy-preserving aggregation layer — perhaps run under the central bank — would unlock most of the uses above.

The legal framework for consented use is incomplete. Nepali law has no clear standard for what a remittance platform may do with anonymised flow data, what a user can consent to, or what a researcher may access under what conditions. Until this is clarified, the platforms default to “use it for fraud and AML only, do not touch otherwise.”

The talent gap is real. Economists who can work with this kind of high-frequency administrative data are scarce in the country. Most of the analytical capacity is in the central bank and a handful of universities; little of it is currently focused on AI-augmented economic measurement.

The fix is not heroic. It is a small inter-platform working group, a privacy-engineering investment, and a regulator willing to set the rules of the road. The country could be running on remittance-powered economic nowcasting in eighteen months if it chose to.

Check your understanding

Quick check

—

What is the most immediately practical use of aggregated remittance-flow data for AI-driven public good in Nepal?

Training a chatbot that explains remittance fees.
Using verified inflow histories (with consent) to offer better credit to remittance-receiving households, and using corridor-level flow data to give central-bank and cooperative actors real-time labour and macroeconomic intelligence.
Replacing all human staff in money transfer companies.
Sending automated marketing messages to migrant workers.

What comes next

Chapter 3 closes here. The country’s economy runs on payments, credit, and remittances; AI is deeply inside the first, slowly arriving in the second, and barely scratching the third. Chapter 4 leaves money entirely. We turn to the country’s other defining substrate: the land itself — mountains, monsoons, and the farms that feed most of the population.